load model from drive and convert

.gitignore

@@ -0,0 +1 @@
+checkpoint-*/

README.md

@@ -0,0 +1,51 @@
+---
+license: apache-2.0
+tags:
+- generated_from_trainer
+model-index:
+- name: mGPT-gpt-pierre_DS-msgs-df_Ep-2_Bs-4
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# mGPT-gpt-pierre_DS-msgs-df_Ep-2_Bs-4
+
+This model is a fine-tuned version of [sberbank-ai/mGPT](https://huggingface.co/sberbank-ai/mGPT) on the None dataset.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 5e-05
+- train_batch_size: 4
+- eval_batch_size: 4
+- seed: 42
+- distributed_type: multi-GPU
+- gradient_accumulation_steps: 8
+- total_train_batch_size: 32
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: cosine_with_restarts
+- lr_scheduler_warmup_ratio: 0.05
+- num_epochs: 2
+
+### Framework versions
+
+- Transformers 4.18.0
+- Pytorch 1.11.0+cu113
+- Datasets 2.1.0
+- Tokenizers 0.12.1

config.json

@@ -0,0 +1,34 @@
+{
+  "_name_or_path": "sberbank-ai/mGPT",
+  "activation_function": "gelu_new",
+  "architectures": [
+    "GPT2LMHeadModel"
+  ],
+  "attn_pdrop": 0.1,
+  "bos_token_id": 50256,
+  "embd_pdrop": 0.1,
+  "eos_token_id": 50256,
+  "gradient_checkpointing": false,
+  "initializer_range": 0.02,
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "gpt2",
+  "n_ctx": 2048,
+  "n_embd": 2048,
+  "n_head": 16,
+  "n_inner": null,
+  "n_layer": 24,
+  "n_positions": 2048,
+  "reorder_and_upcast_attn": false,
+  "resid_pdrop": 0.1,
+  "scale_attn_by_inverse_layer_idx": false,
+  "scale_attn_weights": true,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.18.0",
+  "use_cache": false,
+  "vocab_size": 100000
+}

latest

@@ -0,0 +1 @@
+global_step3202

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d6db98cdd92c575f5bc1c7aa998e2491670c6a70c1663a8f33b4827fc72a1ee
+size 6073088630

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "eos_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "unk_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "pad_token": "<|endoftext|>"}

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "bos_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "eos_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "add_prefix_space": false, "model_max_length": 512, "special_tokens_map_file": null, "name_or_path": "sberbank-ai/mGPT", "errors": "replace", "tokenizer_class": "GPT2Tokenizer"}

trainer_state.json

@@ -0,0 +1,3856 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.29535224801497,
+  "global_step": 3201,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.4595617233228861e-05,
+      "loss": 5.7174,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.0881607426192572e-05,
+      "loss": 4.073,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.4558675968977286e-05,
+      "loss": 3.8184,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.7167597619156276e-05,
+      "loss": 3.485,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.9191234466457723e-05,
+      "loss": 3.4225,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.0844666161941e-05,
+      "loss": 3.4162,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.2242622741451356e-05,
+      "loss": 3.1498,
+      "step": 35
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.345358781211998e-05,
+      "loss": 3.1313,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.4521734704725705e-05,
+      "loss": 3.1651,
+      "step": 45
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.547722465942143e-05,
+      "loss": 3.0759,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.6341570461211475e-05,
+      "loss": 3.2154,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.71306563549047e-05,
+      "loss": 3.0505,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.785654501111715e-05,
+      "loss": 3.0904,
+      "step": 65
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.8528612934415064e-05,
+      "loss": 3.1834,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.9154293202206144e-05,
+      "loss": 3.1904,
+      "step": 75
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 3.973957800508369e-05,
+      "loss": 3.1185,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 4.028936856743207e-05,
+      "loss": 2.8454,
+      "step": 85
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.080772489768941e-05,
+      "loss": 3.1069,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.1298047923155946e-05,
+      "loss": 3.1566,
+      "step": 95
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.1763214852385144e-05,
+      "loss": 2.9029,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.220568147719977e-05,
+      "loss": 3.0556,
+      "step": 105
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.262756065417519e-05,
+      "loss": 3.1721,
+      "step": 110
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.3030683326266976e-05,
+      "loss": 2.9709,
+      "step": 115
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.3416646547868404e-05,
+      "loss": 2.9186,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.378685169968659e-05,
+      "loss": 3.1484,
+      "step": 125
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.414253520408086e-05,
+      "loss": 2.9578,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.4484793440474126e-05,
+      "loss": 3.0733,
+      "step": 135
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.481460312737877e-05,
+      "loss": 3.0312,
+      "step": 140
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.513283812620485e-05,
+      "loss": 2.9714,
+      "step": 145
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.544028339516985e-05,
+      "loss": 2.9715,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.573764665433774e-05,
+      "loss": 3.1439,
+      "step": 155
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.60255681980474e-05,
+      "loss": 2.9821,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.630462919695989e-05,
+      "loss": 2.9551,
+      "step": 165
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.6575358760395774e-05,
+      "loss": 3.0681,
+      "step": 170
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.683823997468022e-05,
+      "loss": 3.0571,
+      "step": 175
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.709371509065312e-05,
+      "loss": 3.0496,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.7342190000274217e-05,
+      "loss": 2.9148,
+      "step": 185
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.758403811611965e-05,
+      "loss": 2.9517,
+      "step": 190
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.781960374686557e-05,
+      "loss": 2.8606,
+      "step": 195
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.8049205045348844e-05,
+      "loss": 2.9665,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.827313659255119e-05,
+      "loss": 2.9399,
+      "step": 205
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.849167167016348e-05,
+      "loss": 2.9148,
+      "step": 210
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.8705064265710865e-05,
+      "loss": 2.8763,
+      "step": 215
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.8913550847138896e-05,
+      "loss": 2.9652,
+      "step": 220
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.911735193795457e-05,
+      "loss": 2.854,
+      "step": 225
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.9316673519230676e-05,
+      "loss": 3.0612,
+      "step": 230
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.951170828082003e-05,
+      "loss": 2.9249,
+      "step": 235
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.970263674083211e-05,
+      "loss": 2.8752,
+      "step": 240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.988962824967385e-05,
+      "loss": 3.0823,
+      "step": 245
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 5e-05,
+      "loss": 2.9281,
+      "step": 250
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 5e-05,
+      "loss": 3.0877,
+      "step": 255
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 5e-05,
+      "loss": 2.8499,
+      "step": 260
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 5e-05,
+      "loss": 2.9582,
+      "step": 265
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 5e-05,
+      "loss": 2.9118,
+      "step": 270
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 5e-05,
+      "loss": 3.0349,
+      "step": 275
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 5e-05,
+      "loss": 2.7885,
+      "step": 280
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 5e-05,
+      "loss": 3.0644,
+      "step": 285
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 5e-05,
+      "loss": 2.9174,
+      "step": 290
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 5e-05,
+      "loss": 2.8956,
+      "step": 295
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 5e-05,
+      "loss": 2.8977,
+      "step": 300
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 5e-05,
+      "loss": 2.7825,
+      "step": 305
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 5e-05,
+      "loss": 2.9082,
+      "step": 310
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 5e-05,
+      "loss": 2.8638,
+      "step": 315
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 5e-05,
+      "loss": 3.0091,
+      "step": 320
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 5e-05,
+      "loss": 2.8423,
+      "step": 325
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 5e-05,
+      "loss": 3.0593,
+      "step": 330
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 5e-05,
+      "loss": 3.0978,
+      "step": 335
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 5e-05,
+      "loss": 2.9759,
+      "step": 340
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 5e-05,
+      "loss": 2.7671,
+      "step": 345
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 5e-05,
+      "loss": 2.7337,
+      "step": 350
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 5e-05,
+      "loss": 2.7853,
+      "step": 355
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 5e-05,
+      "loss": 2.7746,
+      "step": 360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 5e-05,
+      "loss": 2.8793,
+      "step": 365
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 5e-05,
+      "loss": 2.8478,
+      "step": 370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 5e-05,
+      "loss": 2.9319,
+      "step": 375
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 5e-05,
+      "loss": 2.9938,
+      "step": 380
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 5e-05,
+      "loss": 2.9035,
+      "step": 385
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 5e-05,
+      "loss": 2.8904,
+      "step": 390
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 5e-05,
+      "loss": 2.9606,
+      "step": 395
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 5e-05,
+      "loss": 3.0521,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 5e-05,
+      "loss": 2.9584,
+      "step": 405
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 5e-05,
+      "loss": 2.8218,
+      "step": 410
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 5e-05,
+      "loss": 2.9647,
+      "step": 415
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 5e-05,
+      "loss": 2.792,
+      "step": 420
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 5e-05,
+      "loss": 2.656,
+      "step": 425
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 5e-05,
+      "loss": 2.905,
+      "step": 430
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 5e-05,
+      "loss": 2.7492,
+      "step": 435
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 5e-05,
+      "loss": 2.924,
+      "step": 440
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 5e-05,
+      "loss": 2.672,
+      "step": 445
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 5e-05,
+      "loss": 2.9088,
+      "step": 450
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 5e-05,
+      "loss": 2.8127,
+      "step": 455
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 5e-05,
+      "loss": 2.7125,
+      "step": 460
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 5e-05,
+      "loss": 2.9101,
+      "step": 465
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 5e-05,
+      "loss": 2.878,
+      "step": 470
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 5e-05,
+      "loss": 2.9539,
+      "step": 475
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 5e-05,
+      "loss": 2.7836,
+      "step": 480
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 5e-05,
+      "loss": 2.9817,
+      "step": 485
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 5e-05,
+      "loss": 2.9691,
+      "step": 490
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 5e-05,
+      "loss": 2.7756,
+      "step": 495
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 5e-05,
+      "loss": 2.8108,
+      "step": 500
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 5e-05,
+      "loss": 2.7651,
+      "step": 505
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 5e-05,
+      "loss": 2.8161,
+      "step": 510
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 5e-05,
+      "loss": 2.7832,
+      "step": 515
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 5e-05,
+      "loss": 2.7455,
+      "step": 520
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 5e-05,
+      "loss": 2.8838,
+      "step": 525
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 5e-05,
+      "loss": 2.7434,
+      "step": 530
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 5e-05,
+      "loss": 2.8708,
+      "step": 535
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 5e-05,
+      "loss": 2.8642,
+      "step": 540
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 5e-05,
+      "loss": 2.7339,
+      "step": 545
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 5e-05,
+      "loss": 2.7499,
+      "step": 550
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 5e-05,
+      "loss": 2.7786,
+      "step": 555
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 5e-05,
+      "loss": 2.8487,
+      "step": 560
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 5e-05,
+      "loss": 2.8406,
+      "step": 565
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 5e-05,
+      "loss": 2.8463,
+      "step": 570
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 5e-05,
+      "loss": 2.7209,
+      "step": 575
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 5e-05,
+      "loss": 2.9581,
+      "step": 580
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 5e-05,
+      "loss": 2.8408,
+      "step": 585
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 5e-05,
+      "loss": 2.9146,
+      "step": 590
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 5e-05,
+      "loss": 2.8323,
+      "step": 595
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 5e-05,
+      "loss": 2.8473,
+      "step": 600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 5e-05,
+      "loss": 2.8023,
+      "step": 605
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 5e-05,
+      "loss": 2.859,
+      "step": 610
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 5e-05,
+      "loss": 2.8659,
+      "step": 615
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 5e-05,
+      "loss": 2.6733,
+      "step": 620
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 5e-05,
+      "loss": 2.7982,
+      "step": 625
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 5e-05,
+      "loss": 2.8614,
+      "step": 630
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 5e-05,
+      "loss": 2.8252,
+      "step": 635
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 5e-05,
+      "loss": 2.7545,
+      "step": 640
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 5e-05,
+      "loss": 2.9128,
+      "step": 645
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 5e-05,
+      "loss": 2.6117,
+      "step": 650
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 5e-05,
+      "loss": 2.8768,
+      "step": 655
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 5e-05,
+      "loss": 2.8761,
+      "step": 660
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 5e-05,
+      "loss": 2.7655,
+      "step": 665
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 5e-05,
+      "loss": 2.7963,
+      "step": 670
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 5e-05,
+      "loss": 2.8481,
+      "step": 675
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 5e-05,
+      "loss": 2.7366,
+      "step": 680
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 5e-05,
+      "loss": 2.9392,
+      "step": 685
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 5e-05,
+      "loss": 2.8233,
+      "step": 690
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 5e-05,
+      "loss": 2.7672,
+      "step": 695
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 5e-05,
+      "loss": 2.789,
+      "step": 700
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 5e-05,
+      "loss": 2.7065,
+      "step": 705
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 5e-05,
+      "loss": 2.8101,
+      "step": 710
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 5e-05,
+      "loss": 2.6257,
+      "step": 715
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 5e-05,
+      "loss": 2.7759,
+      "step": 720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 5e-05,
+      "loss": 2.7739,
+      "step": 725
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 5e-05,
+      "loss": 2.6239,
+      "step": 730
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 5e-05,
+      "loss": 2.7561,
+      "step": 735
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 5e-05,
+      "loss": 2.8663,
+      "step": 740
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 5e-05,
+      "loss": 2.7199,
+      "step": 745
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 5e-05,
+      "loss": 2.7612,
+      "step": 750
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 5e-05,
+      "loss": 2.8215,
+      "step": 755
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 5e-05,
+      "loss": 2.7875,
+      "step": 760
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 5e-05,
+      "loss": 2.7699,
+      "step": 765
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 5e-05,
+      "loss": 2.7907,
+      "step": 770
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 5e-05,
+      "loss": 3.023,
+      "step": 775
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 5e-05,
+      "loss": 2.7708,
+      "step": 780
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 5e-05,
+      "loss": 2.8424,
+      "step": 785
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 5e-05,
+      "loss": 2.8524,
+      "step": 790
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 5e-05,
+      "loss": 2.7407,
+      "step": 795
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 5e-05,
+      "loss": 2.8392,
+      "step": 800
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 5e-05,
+      "loss": 2.7107,
+      "step": 805
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 5e-05,
+      "loss": 2.7377,
+      "step": 810
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 5e-05,
+      "loss": 2.8285,
+      "step": 815
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 5e-05,
+      "loss": 2.7663,
+      "step": 820
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 5e-05,
+      "loss": 2.819,
+      "step": 825
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 5e-05,
+      "loss": 2.7749,
+      "step": 830
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 5e-05,
+      "loss": 2.8127,
+      "step": 835
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 5e-05,
+      "loss": 2.7182,
+      "step": 840
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 5e-05,
+      "loss": 2.7712,
+      "step": 845
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 5e-05,
+      "loss": 2.6442,
+      "step": 850
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 5e-05,
+      "loss": 2.9364,
+      "step": 855
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 5e-05,
+      "loss": 2.9922,
+      "step": 860
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 5e-05,
+      "loss": 2.7308,
+      "step": 865
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 5e-05,
+      "loss": 2.6962,
+      "step": 870
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 5e-05,
+      "loss": 2.6137,
+      "step": 875
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 5e-05,
+      "loss": 2.7204,
+      "step": 880
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 5e-05,
+      "loss": 2.7692,
+      "step": 885
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 5e-05,
+      "loss": 2.794,
+      "step": 890
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 5e-05,
+      "loss": 2.7687,
+      "step": 895
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 5e-05,
+      "loss": 2.7605,
+      "step": 900
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 5e-05,
+      "loss": 2.7807,
+      "step": 905
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 5e-05,
+      "loss": 2.7144,
+      "step": 910
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 5e-05,
+      "loss": 2.6465,
+      "step": 915
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 5e-05,
+      "loss": 2.8023,
+      "step": 920
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 5e-05,
+      "loss": 2.6612,
+      "step": 925
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 5e-05,
+      "loss": 2.8403,
+      "step": 930
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 5e-05,
+      "loss": 2.7133,
+      "step": 935
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 5e-05,
+      "loss": 2.9143,
+      "step": 940
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 5e-05,
+      "loss": 2.7716,
+      "step": 945
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 5e-05,
+      "loss": 2.7732,
+      "step": 950
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 5e-05,
+      "loss": 2.788,
+      "step": 955
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 5e-05,
+      "loss": 2.7164,
+      "step": 960
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 5e-05,
+      "loss": 2.6721,
+      "step": 965
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 5e-05,
+      "loss": 2.9455,
+      "step": 970
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 5e-05,
+      "loss": 2.6689,
+      "step": 975
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5e-05,
+      "loss": 2.629,
+      "step": 980
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5e-05,
+      "loss": 2.8129,
+      "step": 985
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5e-05,
+      "loss": 2.6629,
+      "step": 990
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5e-05,
+      "loss": 2.8313,
+      "step": 995
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 5e-05,
+      "loss": 2.9289,
+      "step": 1000
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5e-05,
+      "loss": 2.7845,
+      "step": 1005
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5e-05,
+      "loss": 2.7841,
+      "step": 1010
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5e-05,
+      "loss": 2.78,
+      "step": 1015
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5e-05,
+      "loss": 2.8525,
+      "step": 1020
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 5e-05,
+      "loss": 2.7313,
+      "step": 1025
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5e-05,
+      "loss": 2.8678,
+      "step": 1030
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5e-05,
+      "loss": 2.8533,
+      "step": 1035
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5e-05,
+      "loss": 2.7356,
+      "step": 1040
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5e-05,
+      "loss": 2.7197,
+      "step": 1045
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 5e-05,
+      "loss": 2.7888,
+      "step": 1050
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5e-05,
+      "loss": 2.8152,
+      "step": 1055
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5e-05,
+      "loss": 2.6529,
+      "step": 1060
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5e-05,
+      "loss": 2.715,
+      "step": 1065
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5e-05,
+      "loss": 2.7427,
+      "step": 1070
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 5e-05,
+      "loss": 2.7064,
+      "step": 1075
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5e-05,
+      "loss": 2.8462,
+      "step": 1080
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5e-05,
+      "loss": 2.6574,
+      "step": 1085
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5e-05,
+      "loss": 2.8997,
+      "step": 1090
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 5e-05,
+      "loss": 2.7303,
+      "step": 1095
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5e-05,
+      "loss": 2.7889,
+      "step": 1100
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5e-05,
+      "loss": 2.7992,
+      "step": 1105
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5e-05,
+      "loss": 2.62,
+      "step": 1110
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5e-05,
+      "loss": 2.6867,
+      "step": 1115
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 5e-05,
+      "loss": 2.5923,
+      "step": 1120
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5e-05,
+      "loss": 2.6593,
+      "step": 1125
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5e-05,
+      "loss": 2.8382,
+      "step": 1130
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5e-05,
+      "loss": 2.5861,
+      "step": 1135
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5e-05,
+      "loss": 2.8416,
+      "step": 1140
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 5e-05,
+      "loss": 2.9029,
+      "step": 1145
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5e-05,
+      "loss": 2.6753,
+      "step": 1150
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5e-05,
+      "loss": 2.6871,
+      "step": 1155
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5e-05,
+      "loss": 2.6733,
+      "step": 1160
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5e-05,
+      "loss": 2.6566,
+      "step": 1165
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 5e-05,
+      "loss": 2.8209,
+      "step": 1170
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5e-05,
+      "loss": 2.8289,
+      "step": 1175
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5e-05,
+      "loss": 2.8023,
+      "step": 1180
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5e-05,
+      "loss": 2.6979,
+      "step": 1185
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5e-05,
+      "loss": 2.6954,
+      "step": 1190
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 5e-05,
+      "loss": 2.7828,
+      "step": 1195
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5e-05,
+      "loss": 2.7909,
+      "step": 1200
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5e-05,
+      "loss": 2.6024,
+      "step": 1205
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5e-05,
+      "loss": 2.7298,
+      "step": 1210
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5e-05,
+      "loss": 2.5615,
+      "step": 1215
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 5e-05,
+      "loss": 2.6645,
+      "step": 1220
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5e-05,
+      "loss": 2.6223,
+      "step": 1225
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5e-05,
+      "loss": 2.8121,
+      "step": 1230
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5e-05,
+      "loss": 2.7913,
+      "step": 1235
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5e-05,
+      "loss": 2.6475,
+      "step": 1240
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 5e-05,
+      "loss": 2.7489,
+      "step": 1245
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 5e-05,
+      "loss": 2.5228,
+      "step": 1250
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 5e-05,
+      "loss": 2.7824,
+      "step": 1255
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 5e-05,
+      "loss": 2.7072,
+      "step": 1260
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 5e-05,
+      "loss": 2.8302,
+      "step": 1265
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 5e-05,
+      "loss": 2.562,
+      "step": 1270
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 5e-05,
+      "loss": 2.6751,
+      "step": 1275
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 5e-05,
+      "loss": 2.6708,
+      "step": 1280
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 5e-05,
+      "loss": 2.7931,
+      "step": 1285
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 5e-05,
+      "loss": 2.8918,
+      "step": 1290
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 5e-05,
+      "loss": 2.7633,
+      "step": 1295
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 5e-05,
+      "loss": 2.6468,
+      "step": 1300
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 5e-05,
+      "loss": 2.829,
+      "step": 1305
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 5e-05,
+      "loss": 2.5203,
+      "step": 1310
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 5e-05,
+      "loss": 2.8926,
+      "step": 1315
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 5e-05,
+      "loss": 2.5314,
+      "step": 1320
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 5e-05,
+      "loss": 2.6843,
+      "step": 1325
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 5e-05,
+      "loss": 2.6962,
+      "step": 1330
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 5e-05,
+      "loss": 2.9101,
+      "step": 1335
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 5e-05,
+      "loss": 2.7593,
+      "step": 1340
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 5e-05,
+      "loss": 2.7953,
+      "step": 1345
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 5e-05,
+      "loss": 2.6882,
+      "step": 1350
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 5e-05,
+      "loss": 2.6233,
+      "step": 1355
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 5e-05,
+      "loss": 2.7277,
+      "step": 1360
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 5e-05,
+      "loss": 2.5897,
+      "step": 1365
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 5e-05,
+      "loss": 2.618,
+      "step": 1370
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 5e-05,
+      "loss": 2.7698,
+      "step": 1375
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 5e-05,
+      "loss": 2.6026,
+      "step": 1380
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 5e-05,
+      "loss": 2.8144,
+      "step": 1385
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 5e-05,
+      "loss": 2.7199,
+      "step": 1390
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 5e-05,
+      "loss": 2.612,
+      "step": 1395
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 5e-05,
+      "loss": 2.7256,
+      "step": 1400
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 5e-05,
+      "loss": 2.7822,
+      "step": 1405
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 5e-05,
+      "loss": 2.7898,
+      "step": 1410
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 5e-05,
+      "loss": 2.6511,
+      "step": 1415
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 5e-05,
+      "loss": 2.68,
+      "step": 1420
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 5e-05,
+      "loss": 2.6535,
+      "step": 1425
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 5e-05,
+      "loss": 2.7386,
+      "step": 1430
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 5e-05,
+      "loss": 2.7598,
+      "step": 1435
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 5e-05,
+      "loss": 2.7257,
+      "step": 1440
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 5e-05,
+      "loss": 2.7083,
+      "step": 1445
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 5e-05,
+      "loss": 2.6456,
+      "step": 1450
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 5e-05,
+      "loss": 2.8651,
+      "step": 1455
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 5e-05,
+      "loss": 2.6377,
+      "step": 1460
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 5e-05,
+      "loss": 2.7162,
+      "step": 1465
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 5e-05,
+      "loss": 2.7777,
+      "step": 1470
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 5e-05,
+      "loss": 2.6719,
+      "step": 1475
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 5e-05,
+      "loss": 2.5657,
+      "step": 1480
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 5e-05,
+      "loss": 2.6202,
+      "step": 1485
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 5e-05,
+      "loss": 2.7201,
+      "step": 1490
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 5e-05,
+      "loss": 2.8134,
+      "step": 1495
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 5e-05,
+      "loss": 2.9,
+      "step": 1500
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 5e-05,
+      "loss": 2.8065,
+      "step": 1505
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 5e-05,
+      "loss": 2.7656,
+      "step": 1510
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 5e-05,
+      "loss": 2.7519,
+      "step": 1515
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 5e-05,
+      "loss": 2.6922,
+      "step": 1520
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 5e-05,
+      "loss": 2.5151,
+      "step": 1525
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 5e-05,
+      "loss": 2.6186,
+      "step": 1530
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 5e-05,
+      "loss": 2.6513,
+      "step": 1535
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 5e-05,
+      "loss": 2.6852,
+      "step": 1540
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 5e-05,
+      "loss": 2.7875,
+      "step": 1545
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 5e-05,
+      "loss": 2.8141,
+      "step": 1550
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 5e-05,
+      "loss": 2.6425,
+      "step": 1555
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 5e-05,
+      "loss": 2.72,
+      "step": 1560
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 5e-05,
+      "loss": 2.7421,
+      "step": 1565
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 5e-05,
+      "loss": 2.7139,
+      "step": 1570
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 5e-05,
+      "loss": 2.6792,
+      "step": 1575
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 5e-05,
+      "loss": 2.5915,
+      "step": 1580
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 5e-05,
+      "loss": 2.6162,
+      "step": 1585
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 5e-05,
+      "loss": 2.617,
+      "step": 1590
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 5e-05,
+      "loss": 2.7125,
+      "step": 1595
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 5e-05,
+      "loss": 2.7146,
+      "step": 1600
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 5e-05,
+      "loss": 2.5648,
+      "step": 1605
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 5e-05,
+      "loss": 2.8611,
+      "step": 1610
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 5e-05,
+      "loss": 2.5752,
+      "step": 1615
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 5e-05,
+      "loss": 2.6658,
+      "step": 1620
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 5e-05,
+      "loss": 2.7633,
+      "step": 1625
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 5e-05,
+      "loss": 2.6336,
+      "step": 1630
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 5e-05,
+      "loss": 2.771,
+      "step": 1635
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 5e-05,
+      "loss": 2.8223,
+      "step": 1640
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 5e-05,
+      "loss": 2.7626,
+      "step": 1645
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 5e-05,
+      "loss": 2.7405,
+      "step": 1650
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 5e-05,
+      "loss": 2.7567,
+      "step": 1655
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 5e-05,
+      "loss": 2.8065,
+      "step": 1660
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 5e-05,
+      "loss": 2.6281,
+      "step": 1665
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 5e-05,
+      "loss": 2.5907,
+      "step": 1670
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 5e-05,
+      "loss": 2.8343,
+      "step": 1675
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 5e-05,
+      "loss": 2.5374,
+      "step": 1680
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 5e-05,
+      "loss": 2.6573,
+      "step": 1685
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 5e-05,
+      "loss": 2.7715,
+      "step": 1690
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 5e-05,
+      "loss": 2.7022,
+      "step": 1695
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 5e-05,
+      "loss": 2.6918,
+      "step": 1700
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 5e-05,
+      "loss": 2.7902,
+      "step": 1705
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 5e-05,
+      "loss": 2.7458,
+      "step": 1710
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 5e-05,
+      "loss": 2.6979,
+      "step": 1715
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 5e-05,
+      "loss": 2.7609,
+      "step": 1720
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 5e-05,
+      "loss": 2.6818,
+      "step": 1725
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 5e-05,
+      "loss": 2.6761,
+      "step": 1730
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 5e-05,
+      "loss": 2.7184,
+      "step": 1735
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 5e-05,
+      "loss": 2.585,
+      "step": 1740
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 5e-05,
+      "loss": 2.7434,
+      "step": 1745
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 5e-05,
+      "loss": 2.6304,
+      "step": 1750
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 5e-05,
+      "loss": 2.841,
+      "step": 1755
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 5e-05,
+      "loss": 2.7202,
+      "step": 1760
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 5e-05,
+      "loss": 2.5044,
+      "step": 1765
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 5e-05,
+      "loss": 2.655,
+      "step": 1770
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 5e-05,
+      "loss": 2.5988,
+      "step": 1775
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 5e-05,
+      "loss": 2.588,
+      "step": 1780
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 5e-05,
+      "loss": 2.7623,
+      "step": 1785
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 5e-05,
+      "loss": 2.7245,
+      "step": 1790
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 5e-05,
+      "loss": 2.6365,
+      "step": 1795
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 5e-05,
+      "loss": 2.6877,
+      "step": 1800
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 5e-05,
+      "loss": 2.5983,
+      "step": 1805
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 5e-05,
+      "loss": 2.7249,
+      "step": 1810
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 5e-05,
+      "loss": 2.7584,
+      "step": 1815
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 5e-05,
+      "loss": 2.7737,
+      "step": 1820
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 5e-05,
+      "loss": 2.7943,
+      "step": 1825
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 5e-05,
+      "loss": 2.6356,
+      "step": 1830
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 5e-05,
+      "loss": 2.535,
+      "step": 1835
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 5e-05,
+      "loss": 2.7207,
+      "step": 1840
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 5e-05,
+      "loss": 2.7745,
+      "step": 1845
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 5e-05,
+      "loss": 2.6661,
+      "step": 1850
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 5e-05,
+      "loss": 2.6929,
+      "step": 1855
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 5e-05,
+      "loss": 2.6875,
+      "step": 1860
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 5e-05,
+      "loss": 2.7406,
+      "step": 1865
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 5e-05,
+      "loss": 2.8286,
+      "step": 1870
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 5e-05,
+      "loss": 2.7516,
+      "step": 1875
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 5e-05,
+      "loss": 2.7069,
+      "step": 1880
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 5e-05,
+      "loss": 2.6228,
+      "step": 1885
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 5e-05,
+      "loss": 2.7762,
+      "step": 1890
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 5e-05,
+      "loss": 2.694,
+      "step": 1895
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 5e-05,
+      "loss": 2.6888,
+      "step": 1900
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 5e-05,
+      "loss": 2.7838,
+      "step": 1905
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 5e-05,
+      "loss": 2.6155,
+      "step": 1910
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 5e-05,
+      "loss": 2.5779,
+      "step": 1915
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 5e-05,
+      "loss": 2.6237,
+      "step": 1920
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 5e-05,
+      "loss": 2.6635,
+      "step": 1925
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 5e-05,
+      "loss": 2.6905,
+      "step": 1930
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 5e-05,
+      "loss": 2.6694,
+      "step": 1935
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 5e-05,
+      "loss": 2.4783,
+      "step": 1940
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 5e-05,
+      "loss": 2.7565,
+      "step": 1945
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 5e-05,
+      "loss": 2.7145,
+      "step": 1950
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 5e-05,
+      "loss": 2.6378,
+      "step": 1955
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 5e-05,
+      "loss": 2.7469,
+      "step": 1960
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 5e-05,
+      "loss": 2.6073,
+      "step": 1965
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 5e-05,
+      "loss": 2.7848,
+      "step": 1970
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 5e-05,
+      "loss": 2.6595,
+      "step": 1975
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 5e-05,
+      "loss": 2.8059,
+      "step": 1980
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 5e-05,
+      "loss": 2.7439,
+      "step": 1985
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 5e-05,
+      "loss": 2.583,
+      "step": 1990
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 5e-05,
+      "loss": 2.7066,
+      "step": 1995
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 5e-05,
+      "loss": 2.6711,
+      "step": 2000
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 5e-05,
+      "loss": 2.6781,
+      "step": 2005
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 5e-05,
+      "loss": 2.6504,
+      "step": 2010
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 5e-05,
+      "loss": 2.601,
+      "step": 2015
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 5e-05,
+      "loss": 2.7151,
+      "step": 2020
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 5e-05,
+      "loss": 2.7105,
+      "step": 2025
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 5e-05,
+      "loss": 2.4789,
+      "step": 2030
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 5e-05,
+      "loss": 2.587,
+      "step": 2035
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 5e-05,
+      "loss": 2.7408,
+      "step": 2040
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 5e-05,
+      "loss": 2.777,
+      "step": 2045
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 5e-05,
+      "loss": 2.6372,
+      "step": 2050
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 5e-05,
+      "loss": 2.6225,
+      "step": 2055
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 5e-05,
+      "loss": 2.4768,
+      "step": 2060
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 5e-05,
+      "loss": 2.7418,
+      "step": 2065
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 5e-05,
+      "loss": 2.758,
+      "step": 2070
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 5e-05,
+      "loss": 2.6905,
+      "step": 2075
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 5e-05,
+      "loss": 2.8158,
+      "step": 2080
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 5e-05,
+      "loss": 2.809,
+      "step": 2085
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 5e-05,
+      "loss": 2.7938,
+      "step": 2090
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 5e-05,
+      "loss": 2.6229,
+      "step": 2095
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 5e-05,
+      "loss": 2.6269,
+      "step": 2100
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 5e-05,
+      "loss": 2.6908,
+      "step": 2105
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 5e-05,
+      "loss": 2.7486,
+      "step": 2110
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 5e-05,
+      "loss": 2.7088,
+      "step": 2115
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 5e-05,
+      "loss": 2.5786,
+      "step": 2120
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 5e-05,
+      "loss": 2.725,
+      "step": 2125
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 5e-05,
+      "loss": 2.6511,
+      "step": 2130
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 5e-05,
+      "loss": 2.4968,
+      "step": 2135
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 5e-05,
+      "loss": 2.7425,
+      "step": 2140
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 5e-05,
+      "loss": 2.5695,
+      "step": 2145
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 5e-05,
+      "loss": 2.5162,
+      "step": 2150
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 5e-05,
+      "loss": 2.504,
+      "step": 2155
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 5e-05,
+      "loss": 2.7378,
+      "step": 2160
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5e-05,
+      "loss": 2.6437,
+      "step": 2165
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5e-05,
+      "loss": 2.629,
+      "step": 2170
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5e-05,
+      "loss": 2.5375,
+      "step": 2175
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5e-05,
+      "loss": 2.7463,
+      "step": 2180
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 5e-05,
+      "loss": 2.5113,
+      "step": 2185
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5e-05,
+      "loss": 2.7163,
+      "step": 2190
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5e-05,
+      "loss": 2.8432,
+      "step": 2195
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5e-05,
+      "loss": 2.7086,
+      "step": 2200
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5e-05,
+      "loss": 2.784,
+      "step": 2205
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 5e-05,
+      "loss": 2.7021,
+      "step": 2210
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5e-05,
+      "loss": 2.6244,
+      "step": 2215
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5e-05,
+      "loss": 2.633,
+      "step": 2220
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5e-05,
+      "loss": 2.7068,
+      "step": 2225
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5e-05,
+      "loss": 2.8057,
+      "step": 2230
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 5e-05,
+      "loss": 2.677,
+      "step": 2235
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 5e-05,
+      "loss": 2.7191,
+      "step": 2240
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 5e-05,
+      "loss": 2.7879,
+      "step": 2245
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 5e-05,
+      "loss": 2.7059,
+      "step": 2250
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 5e-05,
+      "loss": 2.7086,
+      "step": 2255
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 5e-05,
+      "loss": 2.6386,
+      "step": 2260
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 5e-05,
+      "loss": 2.6557,
+      "step": 2265
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 5e-05,
+      "loss": 2.7518,
+      "step": 2270
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 5e-05,
+      "loss": 2.6355,
+      "step": 2275
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 5e-05,
+      "loss": 2.6018,
+      "step": 2280
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 5e-05,
+      "loss": 2.5666,
+      "step": 2285
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5e-05,
+      "loss": 2.7285,
+      "step": 2290
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5e-05,
+      "loss": 2.6315,
+      "step": 2295
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5e-05,
+      "loss": 2.7684,
+      "step": 2300
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5e-05,
+      "loss": 2.8144,
+      "step": 2305
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 5e-05,
+      "loss": 2.6321,
+      "step": 2310
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5e-05,
+      "loss": 2.534,
+      "step": 2315
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5e-05,
+      "loss": 2.6426,
+      "step": 2320
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5e-05,
+      "loss": 2.7651,
+      "step": 2325
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5e-05,
+      "loss": 2.9408,
+      "step": 2330
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 5e-05,
+      "loss": 2.5735,
+      "step": 2335
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 5e-05,
+      "loss": 2.7254,
+      "step": 2340
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 5e-05,
+      "loss": 2.5567,
+      "step": 2345
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 5e-05,
+      "loss": 2.6888,
+      "step": 2350
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 5e-05,
+      "loss": 2.615,
+      "step": 2355
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 5e-05,
+      "loss": 2.7708,
+      "step": 2360
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 5e-05,
+      "loss": 2.6665,
+      "step": 2365
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 5e-05,
+      "loss": 2.5748,
+      "step": 2370
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 5e-05,
+      "loss": 2.6752,
+      "step": 2375
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 5e-05,
+      "loss": 2.5538,
+      "step": 2380
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 5e-05,
+      "loss": 2.7701,
+      "step": 2385
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 5e-05,
+      "loss": 2.7305,
+      "step": 2390
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 5e-05,
+      "loss": 2.7094,
+      "step": 2395
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 5e-05,
+      "loss": 2.9193,
+      "step": 2400
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 5e-05,
+      "loss": 2.6647,
+      "step": 2405
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 5e-05,
+      "loss": 2.6341,
+      "step": 2410
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 5e-05,
+      "loss": 2.4908,
+      "step": 2415
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 5e-05,
+      "loss": 2.5294,
+      "step": 2420
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 5e-05,
+      "loss": 2.461,
+      "step": 2425
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 5e-05,
+      "loss": 2.6605,
+      "step": 2430
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5e-05,
+      "loss": 2.5215,
+      "step": 2435
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5e-05,
+      "loss": 2.539,
+      "step": 2440
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5e-05,
+      "loss": 2.644,
+      "step": 2445
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5e-05,
+      "loss": 2.5958,
+      "step": 2450
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 5e-05,
+      "loss": 2.528,
+      "step": 2455
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5e-05,
+      "loss": 2.6468,
+      "step": 2460
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5e-05,
+      "loss": 2.6027,
+      "step": 2465
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5e-05,
+      "loss": 2.4891,
+      "step": 2470
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5e-05,
+      "loss": 2.7302,
+      "step": 2475
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 5e-05,
+      "loss": 2.2587,
+      "step": 2480
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 5e-05,
+      "loss": 2.3727,
+      "step": 2485
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 5e-05,
+      "loss": 2.2232,
+      "step": 2490
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 5e-05,
+      "loss": 2.2597,
+      "step": 2495
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 5e-05,
+      "loss": 2.0625,
+      "step": 2500
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 5e-05,
+      "loss": 2.2327,
+      "step": 2505
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 5e-05,
+      "loss": 2.0604,
+      "step": 2510
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 5e-05,
+      "loss": 2.0927,
+      "step": 2515
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 5e-05,
+      "loss": 2.2176,
+      "step": 2520
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 5e-05,
+      "loss": 2.2972,
+      "step": 2525
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 5e-05,
+      "loss": 2.1105,
+      "step": 2530
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 5e-05,
+      "loss": 2.1953,
+      "step": 2535
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 5e-05,
+      "loss": 2.4188,
+      "step": 2540
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 5e-05,
+      "loss": 2.2246,
+      "step": 2545
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 5e-05,
+      "loss": 2.3411,
+      "step": 2550
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 5e-05,
+      "loss": 2.2848,
+      "step": 2555
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 5e-05,
+      "loss": 2.0621,
+      "step": 2560
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 5e-05,
+      "loss": 2.3601,
+      "step": 2565
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 5e-05,
+      "loss": 2.0717,
+      "step": 2570
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 5e-05,
+      "loss": 2.0666,
+      "step": 2575
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 5e-05,
+      "loss": 2.157,
+      "step": 2580
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 5e-05,
+      "loss": 2.2869,
+      "step": 2585
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 5e-05,
+      "loss": 2.2177,
+      "step": 2590
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 5e-05,
+      "loss": 2.2843,
+      "step": 2595
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 5e-05,
+      "loss": 2.1964,
+      "step": 2600
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 5e-05,
+      "loss": 2.2299,
+      "step": 2605
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 5e-05,
+      "loss": 2.1429,
+      "step": 2610
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 5e-05,
+      "loss": 2.1783,
+      "step": 2615
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 5e-05,
+      "loss": 2.0898,
+      "step": 2620
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 5e-05,
+      "loss": 2.2693,
+      "step": 2625
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 5e-05,
+      "loss": 2.2216,
+      "step": 2630
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 5e-05,
+      "loss": 2.1387,
+      "step": 2635
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 5e-05,
+      "loss": 2.2065,
+      "step": 2640
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 5e-05,
+      "loss": 2.0899,
+      "step": 2645
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 5e-05,
+      "loss": 2.1705,
+      "step": 2650
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 5e-05,
+      "loss": 2.1863,
+      "step": 2655
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 5e-05,
+      "loss": 2.2563,
+      "step": 2660
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 5e-05,
+      "loss": 2.2669,
+      "step": 2665
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 5e-05,
+      "loss": 2.2752,
+      "step": 2670
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 5e-05,
+      "loss": 2.3229,
+      "step": 2675
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 5e-05,
+      "loss": 2.1955,
+      "step": 2680
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 5e-05,
+      "loss": 2.1868,
+      "step": 2685
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 5e-05,
+      "loss": 2.3266,
+      "step": 2690
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 5e-05,
+      "loss": 2.1649,
+      "step": 2695
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 5e-05,
+      "loss": 2.1767,
+      "step": 2700
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 5e-05,
+      "loss": 2.2341,
+      "step": 2705
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 5e-05,
+      "loss": 2.2082,
+      "step": 2710
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 5e-05,
+      "loss": 2.1698,
+      "step": 2715
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 5e-05,
+      "loss": 2.1964,
+      "step": 2720
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 5e-05,
+      "loss": 2.2899,
+      "step": 2725
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 5e-05,
+      "loss": 2.2172,
+      "step": 2730
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 5e-05,
+      "loss": 2.1917,
+      "step": 2735
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 5e-05,
+      "loss": 2.2461,
+      "step": 2740
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 5e-05,
+      "loss": 2.143,
+      "step": 2745
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 5e-05,
+      "loss": 2.1732,
+      "step": 2750
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 5e-05,
+      "loss": 2.2312,
+      "step": 2755
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 5e-05,
+      "loss": 2.1724,
+      "step": 2760
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 5e-05,
+      "loss": 2.1024,
+      "step": 2765
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 5e-05,
+      "loss": 2.234,
+      "step": 2770
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 5e-05,
+      "loss": 2.2232,
+      "step": 2775
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 5e-05,
+      "loss": 2.1624,
+      "step": 2780
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 5e-05,
+      "loss": 2.1253,
+      "step": 2785
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 5e-05,
+      "loss": 2.2269,
+      "step": 2790
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 5e-05,
+      "loss": 2.3,
+      "step": 2795
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 5e-05,
+      "loss": 2.2228,
+      "step": 2800
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 5e-05,
+      "loss": 2.2687,
+      "step": 2805
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 5e-05,
+      "loss": 2.0817,
+      "step": 2810
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 5e-05,
+      "loss": 2.3083,
+      "step": 2815
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 5e-05,
+      "loss": 2.1453,
+      "step": 2820
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 5e-05,
+      "loss": 2.229,
+      "step": 2825
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 5e-05,
+      "loss": 2.0037,
+      "step": 2830
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 5e-05,
+      "loss": 2.2472,
+      "step": 2835
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 5e-05,
+      "loss": 2.2924,
+      "step": 2840
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 5e-05,
+      "loss": 2.3999,
+      "step": 2845
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 5e-05,
+      "loss": 2.2672,
+      "step": 2850
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 5e-05,
+      "loss": 2.2466,
+      "step": 2855
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 5e-05,
+      "loss": 2.3562,
+      "step": 2860
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 5e-05,
+      "loss": 2.3754,
+      "step": 2865
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 5e-05,
+      "loss": 2.1568,
+      "step": 2870
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 5e-05,
+      "loss": 2.2053,
+      "step": 2875
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 5e-05,
+      "loss": 2.2557,
+      "step": 2880
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 5e-05,
+      "loss": 2.2292,
+      "step": 2885
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 5e-05,
+      "loss": 2.1612,
+      "step": 2890
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 5e-05,
+      "loss": 2.1222,
+      "step": 2895
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 5e-05,
+      "loss": 2.3077,
+      "step": 2900
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 5e-05,
+      "loss": 2.2526,
+      "step": 2905
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 5e-05,
+      "loss": 2.3626,
+      "step": 2910
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 5e-05,
+      "loss": 2.2218,
+      "step": 2915
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 5e-05,
+      "loss": 2.2083,
+      "step": 2920
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 5e-05,
+      "loss": 2.156,
+      "step": 2925
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 5e-05,
+      "loss": 2.2767,
+      "step": 2930
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 5e-05,
+      "loss": 2.289,
+      "step": 2935
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 5e-05,
+      "loss": 2.2848,
+      "step": 2940
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 5e-05,
+      "loss": 2.276,
+      "step": 2945
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 5e-05,
+      "loss": 2.2919,
+      "step": 2950
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 5e-05,
+      "loss": 2.2985,
+      "step": 2955
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 5e-05,
+      "loss": 2.2449,
+      "step": 2960
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 5e-05,
+      "loss": 2.2573,
+      "step": 2965
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 5e-05,
+      "loss": 2.2188,
+      "step": 2970
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 5e-05,
+      "loss": 2.3185,
+      "step": 2975
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 5e-05,
+      "loss": 2.1787,
+      "step": 2980
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 5e-05,
+      "loss": 2.311,
+      "step": 2985
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 5e-05,
+      "loss": 2.2696,
+      "step": 2990
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 5e-05,
+      "loss": 2.1472,
+      "step": 2995
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 5e-05,
+      "loss": 2.371,
+      "step": 3000
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 5e-05,
+      "loss": 2.2385,
+      "step": 3005
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 5e-05,
+      "loss": 2.213,
+      "step": 3010
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 5e-05,
+      "loss": 2.2276,
+      "step": 3015
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 5e-05,
+      "loss": 2.3147,
+      "step": 3020
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 5e-05,
+      "loss": 2.2818,
+      "step": 3025
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 5e-05,
+      "loss": 2.3763,
+      "step": 3030
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 5e-05,
+      "loss": 2.2653,
+      "step": 3035
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 5e-05,
+      "loss": 2.1061,
+      "step": 3040
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 5e-05,
+      "loss": 2.1532,
+      "step": 3045
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 5e-05,
+      "loss": 2.2831,
+      "step": 3050
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 5e-05,
+      "loss": 2.284,
+      "step": 3055
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 5e-05,
+      "loss": 2.316,
+      "step": 3060
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 5e-05,
+      "loss": 2.2093,
+      "step": 3065
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 5e-05,
+      "loss": 2.3092,
+      "step": 3070
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 5e-05,
+      "loss": 2.121,
+      "step": 3075
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 5e-05,
+      "loss": 2.1213,
+      "step": 3080
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 5e-05,
+      "loss": 2.2853,
+      "step": 3085
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 5e-05,
+      "loss": 2.2103,
+      "step": 3090
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 5e-05,
+      "loss": 2.1558,
+      "step": 3095
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 5e-05,
+      "loss": 2.3642,
+      "step": 3100
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 5e-05,
+      "loss": 2.1975,
+      "step": 3105
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 5e-05,
+      "loss": 2.3263,
+      "step": 3110
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 5e-05,
+      "loss": 2.2045,
+      "step": 3115
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 5e-05,
+      "loss": 2.2657,
+      "step": 3120
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 5e-05,
+      "loss": 2.1824,
+      "step": 3125
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 5e-05,
+      "loss": 2.2344,
+      "step": 3130
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 5e-05,
+      "loss": 2.3884,
+      "step": 3135
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 5e-05,
+      "loss": 2.1757,
+      "step": 3140
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 5e-05,
+      "loss": 2.2789,
+      "step": 3145
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 5e-05,
+      "loss": 2.3038,
+      "step": 3150
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 5e-05,
+      "loss": 2.2446,
+      "step": 3155
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 5e-05,
+      "loss": 2.1975,
+      "step": 3160
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 5e-05,
+      "loss": 2.4234,
+      "step": 3165
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 5e-05,
+      "loss": 2.2307,
+      "step": 3170
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 5e-05,
+      "loss": 2.1857,
+      "step": 3175
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 5e-05,
+      "loss": 2.2006,
+      "step": 3180
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 5e-05,
+      "loss": 2.2475,
+      "step": 3185
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 5e-05,
+      "loss": 2.3465,
+      "step": 3190
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 5e-05,
+      "loss": 2.1186,
+      "step": 3195
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 5e-05,
+      "loss": 2.2803,
+      "step": 3200
+    }
+  ],
+  "max_steps": 4942,
+  "num_train_epochs": 2,
+  "total_flos": 3.8026776806424576e+17,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09f5d04fb3754a0efdcfac1f31e2455f243c3bf11081d2bb3f0513c22a30ddef
+size 4271

zero_to_fp32.py

@@ -0,0 +1,484 @@
+#!/usr/bin/env python
+
+# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+import deepspeed
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION,
+                                            OPTIMIZER_STATE_DICT,
+                                            PARAM_SHAPES,
+                                            SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS,
+                                            ZERO_STAGE,
+                                            PARTITION_COUNT,
+                                            PARAM_SHAPES,
+                                            BUFFER_NAMES)
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage == 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_optim_files(checkpoint_dir):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    optim_files = sorted(glob.glob(os.path.join(checkpoint_dir,
+                                                "*_optim_states.pt")),
+                         key=natural_keys)
+
+    if len(optim_files) == 0:
+        raise FileNotFoundError(
+            f"can't find '*_optim_states.pt' files in directory '{checkpoint_dir}'")
+
+    return optim_files
+
+
+def parse_model_state(file):
+    state_dict = torch.load(file, map_location=device)
+
+    if BUFFER_NAMES not in state_dict:
+        raise ValueError(f"{file} is not a model state checkpoint")
+    buffer_names = state_dict[BUFFER_NAMES]
+    if debug:
+        print("Found buffers:", buffer_names)
+
+    # recover just the buffers while restoring them to fp32 if they were saved in fp16
+    buffers = {
+        k: v.float()
+        for k,
+        v in state_dict["module"].items() if k in buffer_names
+    }
+    param_shapes = state_dict[PARAM_SHAPES]
+
+    ds_version = state_dict.get(DS_VERSION, None)
+
+    return buffers, param_shapes, ds_version
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dicts.append(torch.load(f, map_location=device))
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage == 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage == 2:
+        fp32_flat_groups = [
+            state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key]
+            for i in range(len(state_dicts))
+        ]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key],
+                      0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(
+        f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_file = get_model_state_file(ds_checkpoint_dir, zero_stage)
+    buffers, param_shapes, ds_version = parse_model_state(model_file)
+    print(f'Parsing checkpoint created by deepspeed=={ds_version}')
+
+    if zero_stage == 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                                          param_shapes,
+                                                          fp32_flat_groups,
+                                                          buffers)
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(
+                    f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum([
+        full_single_fp32_vector.numel()
+        for full_single_fp32_vector in merged_single_partition_of_fp32_groups
+    ])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum(
+            [sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(
+                    f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} "
+                )
+            state_dict[name] = full_single_fp32_vector.narrow(
+                0,
+                offset,
+                unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(
+                f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size,
+                                               param_shapes,
+                                               fp32_flat_groups,
+                                               buffers):
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    state_dict = OrderedDict()
+
+    # buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"{total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0,
+                                             offset,
+                                             partitioned_numel)
+                  for i in range(world_size)),
+            0).narrow(0,
+                      0,
+                      unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(
+            f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(
+        f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements"
+    )
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "checkpoint_dir",
+        type=str,
+        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help=
+        "path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)"
+    )
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)