Initial commit with copied model

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

added_tokens.json

@@ -0,0 +1,29 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652,
+  "[PAD]": 151669
+}

config.json

@@ -0,0 +1,35 @@
+{
+  "architectures": [
+    "Qwen3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 28,
+  "model_type": "qwen3",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 8,
+  "parallel_config": {
+    "parallel_style": "none",
+    "pipeline_parallel": 1,
+    "tensor_parallel": 1
+  },
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "max_new_tokens": 2048,
+  "transformers_version": "4.51.3"
+}

global_step3290/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a624e108def6d9b6e5f79d65e016767c64279fa06340d33ce8c9aaac122b507
+size 894080304

global_step3290/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e07ed15aebd22a73ecc47b886681de6325383497b853fb19bd527d5a8632329
+size 1192188984

latest

@@ -0,0 +1 @@
+global_step3290

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf80bd478e6f9fa957d71bd9ffb37f7be681d3aff710158e1e7c65b64e847c34
+size 1503300328

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92cc13315f24c28015d695b6cde08bb1cd6fea4cbc435998485ed6fbe4c91285
+size 15024

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8594aab587aa802e1e2bcfff77510f83c4ab060c0fa208567fb10200d8e606f9
+size 1064

special_tokens_map.json

@@ -0,0 +1,25 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": "<|endoftext|>",
+  "pad_token": {
+    "content": "[PAD]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ea50a7b472ec8fecfd70e281c476f75a704a92a72cbeb4efbb880e1a9e3a67d
+size 11422836

tokenizer_config.json

@@ -0,0 +1,249 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151669": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{% for message in messages %}\n<|im_start|>{{ message.role }}\n{{ message.content }}<|im_end|>\n{% endfor %}\n{% if add_generation_prompt %}\n<|im_start|>assistant\n{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "[PAD]",
+  "padding_side": "right",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

trainer_state.json

@@ -0,0 +1,1222 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9999240179317681,
+  "eval_steps": 160,
+  "global_step": 3290,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0003039282729275891,
+      "grad_norm": 5.229409217834473,
+      "learning_rate": 0.0,
+      "loss": 1.2164,
+      "mean_token_accuracy": 0.7386507540941238,
+      "num_tokens": 87378.0,
+      "step": 1
+    },
+    {
+      "epoch": 0.007598206823189727,
+      "grad_norm": 4.077776908874512,
+      "learning_rate": 1.2121212121212122e-06,
+      "loss": 1.3951,
+      "mean_token_accuracy": 0.7043480854481459,
+      "num_tokens": 2555177.0,
+      "step": 25
+    },
+    {
+      "epoch": 0.015196413646379454,
+      "grad_norm": 3.2261855602264404,
+      "learning_rate": 2.474747474747475e-06,
+      "loss": 1.3716,
+      "mean_token_accuracy": 0.7009463596343994,
+      "num_tokens": 5171296.0,
+      "step": 50
+    },
+    {
+      "epoch": 0.02279462046956918,
+      "grad_norm": 1.221886396408081,
+      "learning_rate": 3.737373737373738e-06,
+      "loss": 1.2697,
+      "mean_token_accuracy": 0.7165025919675827,
+      "num_tokens": 7791520.0,
+      "step": 75
+    },
+    {
+      "epoch": 0.030392827292758908,
+      "grad_norm": 1.2025024890899658,
+      "learning_rate": 5e-06,
+      "loss": 1.2611,
+      "mean_token_accuracy": 0.7154420650005341,
+      "num_tokens": 10428214.0,
+      "step": 100
+    },
+    {
+      "epoch": 0.03799103411594864,
+      "grad_norm": 1.2315771579742432,
+      "learning_rate": 4.960827326856785e-06,
+      "loss": 1.2618,
+      "mean_token_accuracy": 0.7133645015954971,
+      "num_tokens": 13035315.0,
+      "step": 125
+    },
+    {
+      "epoch": 0.04558924093913836,
+      "grad_norm": 1.1814574003219604,
+      "learning_rate": 4.92165465371357e-06,
+      "loss": 1.2494,
+      "mean_token_accuracy": 0.7152758586406708,
+      "num_tokens": 15555229.0,
+      "step": 150
+    },
+    {
+      "epoch": 0.05318744776232809,
+      "grad_norm": 1.0967789888381958,
+      "learning_rate": 4.882481980570354e-06,
+      "loss": 1.1929,
+      "mean_token_accuracy": 0.7268460893630981,
+      "num_tokens": 18110892.0,
+      "step": 175
+    },
+    {
+      "epoch": 0.060785654585517816,
+      "grad_norm": 1.38316011428833,
+      "learning_rate": 4.843309307427139e-06,
+      "loss": 1.2274,
+      "mean_token_accuracy": 0.7177506709098815,
+      "num_tokens": 20638485.0,
+      "step": 200
+    },
+    {
+      "epoch": 0.06838386140870754,
+      "grad_norm": 1.1938471794128418,
+      "learning_rate": 4.804136634283924e-06,
+      "loss": 1.2036,
+      "mean_token_accuracy": 0.7217154312133789,
+      "num_tokens": 23186502.0,
+      "step": 225
+    },
+    {
+      "epoch": 0.07598206823189728,
+      "grad_norm": 1.2482621669769287,
+      "learning_rate": 4.764963961140708e-06,
+      "loss": 1.1962,
+      "mean_token_accuracy": 0.7237263369560242,
+      "num_tokens": 25737398.0,
+      "step": 250
+    },
+    {
+      "epoch": 0.083580275055087,
+      "grad_norm": 0.9959760308265686,
+      "learning_rate": 4.725791287997493e-06,
+      "loss": 1.1908,
+      "mean_token_accuracy": 0.7237645584344864,
+      "num_tokens": 28386635.0,
+      "step": 275
+    },
+    {
+      "epoch": 0.09117848187827672,
+      "grad_norm": 1.1848807334899902,
+      "learning_rate": 4.686618614854278e-06,
+      "loss": 1.1968,
+      "mean_token_accuracy": 0.7229974180459976,
+      "num_tokens": 30963803.0,
+      "step": 300
+    },
+    {
+      "epoch": 0.09877668870146646,
+      "grad_norm": 1.3056175708770752,
+      "learning_rate": 4.647445941711063e-06,
+      "loss": 1.1937,
+      "mean_token_accuracy": 0.7229531377553939,
+      "num_tokens": 33594219.0,
+      "step": 325
+    },
+    {
+      "epoch": 0.10637489552465618,
+      "grad_norm": 1.135485053062439,
+      "learning_rate": 4.608273268567847e-06,
+      "loss": 1.1815,
+      "mean_token_accuracy": 0.724581116437912,
+      "num_tokens": 36131516.0,
+      "step": 350
+    },
+    {
+      "epoch": 0.11397310234784591,
+      "grad_norm": 0.9885596036911011,
+      "learning_rate": 4.569100595424632e-06,
+      "loss": 1.1588,
+      "mean_token_accuracy": 0.7286039453744888,
+      "num_tokens": 38679243.0,
+      "step": 375
+    },
+    {
+      "epoch": 0.12157130917103563,
+      "grad_norm": 0.9995855093002319,
+      "learning_rate": 4.529927922281417e-06,
+      "loss": 1.1754,
+      "mean_token_accuracy": 0.7271614295244216,
+      "num_tokens": 41253914.0,
+      "step": 400
+    },
+    {
+      "epoch": 0.12916951599422535,
+      "grad_norm": 0.9095941185951233,
+      "learning_rate": 4.490755249138202e-06,
+      "loss": 1.2398,
+      "mean_token_accuracy": 0.7160615402460099,
+      "num_tokens": 43939124.0,
+      "step": 425
+    },
+    {
+      "epoch": 0.1367677228174151,
+      "grad_norm": 1.0696874856948853,
+      "learning_rate": 4.451582575994986e-06,
+      "loss": 1.183,
+      "mean_token_accuracy": 0.7253994250297546,
+      "num_tokens": 46572312.0,
+      "step": 450
+    },
+    {
+      "epoch": 0.14436592964060482,
+      "grad_norm": 1.0358718633651733,
+      "learning_rate": 4.41240990285177e-06,
+      "loss": 1.1931,
+      "mean_token_accuracy": 0.7237947028875351,
+      "num_tokens": 49064213.0,
+      "step": 475
+    },
+    {
+      "epoch": 0.15196413646379456,
+      "grad_norm": 0.9893081784248352,
+      "learning_rate": 4.373237229708556e-06,
+      "loss": 1.1951,
+      "mean_token_accuracy": 0.7238883310556412,
+      "num_tokens": 51663643.0,
+      "step": 500
+    },
+    {
+      "epoch": 0.15956234328698427,
+      "grad_norm": 1.0677660703659058,
+      "learning_rate": 4.334064556565341e-06,
+      "loss": 1.1721,
+      "mean_token_accuracy": 0.7265544033050537,
+      "num_tokens": 54161634.0,
+      "step": 525
+    },
+    {
+      "epoch": 0.167160550110174,
+      "grad_norm": 0.8927258849143982,
+      "learning_rate": 4.294891883422124e-06,
+      "loss": 1.1685,
+      "mean_token_accuracy": 0.7285602086782456,
+      "num_tokens": 56784150.0,
+      "step": 550
+    },
+    {
+      "epoch": 0.17475875693336373,
+      "grad_norm": 0.9111976027488708,
+      "learning_rate": 4.25571921027891e-06,
+      "loss": 1.1562,
+      "mean_token_accuracy": 0.7312729161977768,
+      "num_tokens": 59382430.0,
+      "step": 575
+    },
+    {
+      "epoch": 0.18235696375655344,
+      "grad_norm": 0.8939260244369507,
+      "learning_rate": 4.216546537135695e-06,
+      "loss": 1.1818,
+      "mean_token_accuracy": 0.7256826394796372,
+      "num_tokens": 61928694.0,
+      "step": 600
+    },
+    {
+      "epoch": 0.18995517057974318,
+      "grad_norm": 0.9611142873764038,
+      "learning_rate": 4.177373863992479e-06,
+      "loss": 1.1793,
+      "mean_token_accuracy": 0.7260627967119216,
+      "num_tokens": 64496365.0,
+      "step": 625
+    },
+    {
+      "epoch": 0.1975533774029329,
+      "grad_norm": 0.9116730690002441,
+      "learning_rate": 4.138201190849264e-06,
+      "loss": 1.149,
+      "mean_token_accuracy": 0.7322022247314454,
+      "num_tokens": 67108712.0,
+      "step": 650
+    },
+    {
+      "epoch": 0.20515158422612265,
+      "grad_norm": 0.9878592491149902,
+      "learning_rate": 4.099028517706049e-06,
+      "loss": 1.1509,
+      "mean_token_accuracy": 0.7325534474849701,
+      "num_tokens": 69702021.0,
+      "step": 675
+    },
+    {
+      "epoch": 0.21274979104931235,
+      "grad_norm": 0.9378741979598999,
+      "learning_rate": 4.059855844562833e-06,
+      "loss": 1.1801,
+      "mean_token_accuracy": 0.727109580039978,
+      "num_tokens": 72301699.0,
+      "step": 700
+    },
+    {
+      "epoch": 0.2203479978725021,
+      "grad_norm": 1.0023488998413086,
+      "learning_rate": 4.020683171419618e-06,
+      "loss": 1.1814,
+      "mean_token_accuracy": 0.7265605771541596,
+      "num_tokens": 74824515.0,
+      "step": 725
+    },
+    {
+      "epoch": 0.22794620469569182,
+      "grad_norm": 0.9925772547721863,
+      "learning_rate": 3.981510498276403e-06,
+      "loss": 1.1932,
+      "mean_token_accuracy": 0.7241325139999389,
+      "num_tokens": 77468439.0,
+      "step": 750
+    },
+    {
+      "epoch": 0.23554441151888153,
+      "grad_norm": 1.0404661893844604,
+      "learning_rate": 3.942337825133187e-06,
+      "loss": 1.1525,
+      "mean_token_accuracy": 0.7301458239555358,
+      "num_tokens": 80143332.0,
+      "step": 775
+    },
+    {
+      "epoch": 0.24314261834207127,
+      "grad_norm": 0.9634000062942505,
+      "learning_rate": 3.903165151989972e-06,
+      "loss": 1.1691,
+      "mean_token_accuracy": 0.7280007082223893,
+      "num_tokens": 82731678.0,
+      "step": 800
+    },
+    {
+      "epoch": 0.25074082516526097,
+      "grad_norm": 0.9083330631256104,
+      "learning_rate": 3.863992478846757e-06,
+      "loss": 1.1272,
+      "mean_token_accuracy": 0.7353595513105392,
+      "num_tokens": 85345821.0,
+      "step": 825
+    },
+    {
+      "epoch": 0.2583390319884507,
+      "grad_norm": 0.9275352358818054,
+      "learning_rate": 3.824819805703541e-06,
+      "loss": 1.1558,
+      "mean_token_accuracy": 0.7300055593252182,
+      "num_tokens": 87948235.0,
+      "step": 850
+    },
+    {
+      "epoch": 0.26593723881164044,
+      "grad_norm": 0.9232833385467529,
+      "learning_rate": 3.785647132560326e-06,
+      "loss": 1.162,
+      "mean_token_accuracy": 0.7284415501356125,
+      "num_tokens": 90574681.0,
+      "step": 875
+    },
+    {
+      "epoch": 0.2735354456348302,
+      "grad_norm": 1.131739854812622,
+      "learning_rate": 3.746474459417111e-06,
+      "loss": 1.1697,
+      "mean_token_accuracy": 0.7278369426727295,
+      "num_tokens": 93215297.0,
+      "step": 900
+    },
+    {
+      "epoch": 0.2811336524580199,
+      "grad_norm": 0.9522098898887634,
+      "learning_rate": 3.7073017862738957e-06,
+      "loss": 1.1846,
+      "mean_token_accuracy": 0.7259439510107041,
+      "num_tokens": 95791861.0,
+      "step": 925
+    },
+    {
+      "epoch": 0.28873185928120965,
+      "grad_norm": 1.0696117877960205,
+      "learning_rate": 3.66812911313068e-06,
+      "loss": 1.1394,
+      "mean_token_accuracy": 0.733702262043953,
+      "num_tokens": 98410562.0,
+      "step": 950
+    },
+    {
+      "epoch": 0.2963300661043994,
+      "grad_norm": 0.9875262379646301,
+      "learning_rate": 3.628956439987465e-06,
+      "loss": 1.1214,
+      "mean_token_accuracy": 0.7367280977964401,
+      "num_tokens": 101018834.0,
+      "step": 975
+    },
+    {
+      "epoch": 0.3039282729275891,
+      "grad_norm": 1.0046495199203491,
+      "learning_rate": 3.5897837668442497e-06,
+      "loss": 1.1471,
+      "mean_token_accuracy": 0.7323124688863755,
+      "num_tokens": 103548660.0,
+      "step": 1000
+    },
+    {
+      "epoch": 0.3115264797507788,
+      "grad_norm": 0.9220812320709229,
+      "learning_rate": 3.550611093701035e-06,
+      "loss": 1.1144,
+      "mean_token_accuracy": 0.7385808283090591,
+      "num_tokens": 106144886.0,
+      "step": 1025
+    },
+    {
+      "epoch": 0.31912468657396853,
+      "grad_norm": 0.84840327501297,
+      "learning_rate": 3.511438420557819e-06,
+      "loss": 1.1292,
+      "mean_token_accuracy": 0.7352547180652619,
+      "num_tokens": 108737841.0,
+      "step": 1050
+    },
+    {
+      "epoch": 0.32672289339715826,
+      "grad_norm": 0.9784463047981262,
+      "learning_rate": 3.4722657474146037e-06,
+      "loss": 1.1397,
+      "mean_token_accuracy": 0.7329702425003052,
+      "num_tokens": 111263726.0,
+      "step": 1075
+    },
+    {
+      "epoch": 0.334321100220348,
+      "grad_norm": 0.9851428270339966,
+      "learning_rate": 3.433093074271389e-06,
+      "loss": 1.1522,
+      "mean_token_accuracy": 0.7319200646877289,
+      "num_tokens": 113862710.0,
+      "step": 1100
+    },
+    {
+      "epoch": 0.34191930704353773,
+      "grad_norm": 1.0207332372665405,
+      "learning_rate": 3.3939204011281735e-06,
+      "loss": 1.1596,
+      "mean_token_accuracy": 0.7311873000860214,
+      "num_tokens": 116455029.0,
+      "step": 1125
+    },
+    {
+      "epoch": 0.34951751386672747,
+      "grad_norm": 1.057591199874878,
+      "learning_rate": 3.3547477279849577e-06,
+      "loss": 1.1469,
+      "mean_token_accuracy": 0.7317946165800094,
+      "num_tokens": 119016227.0,
+      "step": 1150
+    },
+    {
+      "epoch": 0.3571157206899172,
+      "grad_norm": 0.8993873596191406,
+      "learning_rate": 3.3155750548417424e-06,
+      "loss": 1.166,
+      "mean_token_accuracy": 0.7290427106618881,
+      "num_tokens": 121623537.0,
+      "step": 1175
+    },
+    {
+      "epoch": 0.3647139275131069,
+      "grad_norm": 0.9834045171737671,
+      "learning_rate": 3.2764023816985275e-06,
+      "loss": 1.1433,
+      "mean_token_accuracy": 0.7334721457958221,
+      "num_tokens": 124161994.0,
+      "step": 1200
+    },
+    {
+      "epoch": 0.3723121343362966,
+      "grad_norm": 0.9397407174110413,
+      "learning_rate": 3.237229708555312e-06,
+      "loss": 1.1613,
+      "mean_token_accuracy": 0.7288734668493271,
+      "num_tokens": 126771313.0,
+      "step": 1225
+    },
+    {
+      "epoch": 0.37991034115948635,
+      "grad_norm": 0.8688826560974121,
+      "learning_rate": 3.1980570354120964e-06,
+      "loss": 1.1314,
+      "mean_token_accuracy": 0.7348222607374191,
+      "num_tokens": 129372256.0,
+      "step": 1250
+    },
+    {
+      "epoch": 0.3875085479826761,
+      "grad_norm": 0.9596332311630249,
+      "learning_rate": 3.1588843622688815e-06,
+      "loss": 1.1771,
+      "mean_token_accuracy": 0.7289222145080566,
+      "num_tokens": 132004952.0,
+      "step": 1275
+    },
+    {
+      "epoch": 0.3951067548058658,
+      "grad_norm": 0.8962276577949524,
+      "learning_rate": 3.119711689125666e-06,
+      "loss": 1.13,
+      "mean_token_accuracy": 0.7361785328388214,
+      "num_tokens": 134597308.0,
+      "step": 1300
+    },
+    {
+      "epoch": 0.40270496162905556,
+      "grad_norm": 0.9526214599609375,
+      "learning_rate": 3.0805390159824512e-06,
+      "loss": 1.1279,
+      "mean_token_accuracy": 0.7370401775836944,
+      "num_tokens": 137154477.0,
+      "step": 1325
+    },
+    {
+      "epoch": 0.4103031684522453,
+      "grad_norm": 0.9959564208984375,
+      "learning_rate": 3.0413663428392355e-06,
+      "loss": 1.1501,
+      "mean_token_accuracy": 0.7321191453933715,
+      "num_tokens": 139811600.0,
+      "step": 1350
+    },
+    {
+      "epoch": 0.41790137527543497,
+      "grad_norm": 0.8501905798912048,
+      "learning_rate": 3.00219366969602e-06,
+      "loss": 1.1509,
+      "mean_token_accuracy": 0.7308077716827392,
+      "num_tokens": 142453993.0,
+      "step": 1375
+    },
+    {
+      "epoch": 0.4254995820986247,
+      "grad_norm": 0.913529098033905,
+      "learning_rate": 2.9630209965528052e-06,
+      "loss": 1.1338,
+      "mean_token_accuracy": 0.7351639837026596,
+      "num_tokens": 144957196.0,
+      "step": 1400
+    },
+    {
+      "epoch": 0.43309778892181444,
+      "grad_norm": 0.9400737285614014,
+      "learning_rate": 2.92384832340959e-06,
+      "loss": 1.1322,
+      "mean_token_accuracy": 0.7360527998209,
+      "num_tokens": 147574322.0,
+      "step": 1425
+    },
+    {
+      "epoch": 0.4406959957450042,
+      "grad_norm": 0.9140069484710693,
+      "learning_rate": 2.884675650266374e-06,
+      "loss": 1.1359,
+      "mean_token_accuracy": 0.7348436897993088,
+      "num_tokens": 150085697.0,
+      "step": 1450
+    },
+    {
+      "epoch": 0.4482942025681939,
+      "grad_norm": 0.9307470321655273,
+      "learning_rate": 2.8455029771231592e-06,
+      "loss": 1.1537,
+      "mean_token_accuracy": 0.7332108038663864,
+      "num_tokens": 152693678.0,
+      "step": 1475
+    },
+    {
+      "epoch": 0.45589240939138365,
+      "grad_norm": 0.9836713671684265,
+      "learning_rate": 2.806330303979944e-06,
+      "loss": 1.1398,
+      "mean_token_accuracy": 0.7350041055679322,
+      "num_tokens": 155307631.0,
+      "step": 1500
+    },
+    {
+      "epoch": 0.4634906162145734,
+      "grad_norm": 0.9581474661827087,
+      "learning_rate": 2.7671576308367286e-06,
+      "loss": 1.1213,
+      "mean_token_accuracy": 0.7367819517850875,
+      "num_tokens": 157931564.0,
+      "step": 1525
+    },
+    {
+      "epoch": 0.47108882303776306,
+      "grad_norm": 0.9196196794509888,
+      "learning_rate": 2.727984957693513e-06,
+      "loss": 1.1416,
+      "mean_token_accuracy": 0.7337223035097122,
+      "num_tokens": 160523557.0,
+      "step": 1550
+    },
+    {
+      "epoch": 0.4786870298609528,
+      "grad_norm": 1.0535234212875366,
+      "learning_rate": 2.688812284550298e-06,
+      "loss": 1.1376,
+      "mean_token_accuracy": 0.7336768537759781,
+      "num_tokens": 163117027.0,
+      "step": 1575
+    },
+    {
+      "epoch": 0.48628523668414253,
+      "grad_norm": 0.9108763337135315,
+      "learning_rate": 2.6496396114070826e-06,
+      "loss": 1.1186,
+      "mean_token_accuracy": 0.7363489520549774,
+      "num_tokens": 165577284.0,
+      "step": 1600
+    },
+    {
+      "epoch": 0.49388344350733226,
+      "grad_norm": 0.914046585559845,
+      "learning_rate": 2.6104669382638677e-06,
+      "loss": 1.1193,
+      "mean_token_accuracy": 0.7358357053995133,
+      "num_tokens": 168176206.0,
+      "step": 1625
+    },
+    {
+      "epoch": 0.5014816503305219,
+      "grad_norm": 0.9301220774650574,
+      "learning_rate": 2.571294265120652e-06,
+      "loss": 1.1359,
+      "mean_token_accuracy": 0.7337970972061157,
+      "num_tokens": 170860276.0,
+      "step": 1650
+    },
+    {
+      "epoch": 0.5090798571537117,
+      "grad_norm": 1.1479674577713013,
+      "learning_rate": 2.5321215919774366e-06,
+      "loss": 1.1339,
+      "mean_token_accuracy": 0.7356235873699188,
+      "num_tokens": 173397712.0,
+      "step": 1675
+    },
+    {
+      "epoch": 0.5166780639769014,
+      "grad_norm": 0.9349409937858582,
+      "learning_rate": 2.4929489188342217e-06,
+      "loss": 1.1381,
+      "mean_token_accuracy": 0.7338095015287399,
+      "num_tokens": 175907043.0,
+      "step": 1700
+    },
+    {
+      "epoch": 0.5242762708000912,
+      "grad_norm": 0.9135065078735352,
+      "learning_rate": 2.453776245691006e-06,
+      "loss": 1.1424,
+      "mean_token_accuracy": 0.7341409718990326,
+      "num_tokens": 178454221.0,
+      "step": 1725
+    },
+    {
+      "epoch": 0.5318744776232809,
+      "grad_norm": 0.8928455710411072,
+      "learning_rate": 2.414603572547791e-06,
+      "loss": 1.1486,
+      "mean_token_accuracy": 0.7311108547449112,
+      "num_tokens": 181045005.0,
+      "step": 1750
+    },
+    {
+      "epoch": 0.5394726844464707,
+      "grad_norm": 0.980609118938446,
+      "learning_rate": 2.3754308994045757e-06,
+      "loss": 1.1368,
+      "mean_token_accuracy": 0.7330597722530365,
+      "num_tokens": 183666329.0,
+      "step": 1775
+    },
+    {
+      "epoch": 0.5470708912696604,
+      "grad_norm": 0.9772244691848755,
+      "learning_rate": 2.3362582262613603e-06,
+      "loss": 1.1105,
+      "mean_token_accuracy": 0.7395724445581436,
+      "num_tokens": 186221096.0,
+      "step": 1800
+    },
+    {
+      "epoch": 0.55466909809285,
+      "grad_norm": 0.9453775882720947,
+      "learning_rate": 2.297085553118145e-06,
+      "loss": 1.1207,
+      "mean_token_accuracy": 0.737914999127388,
+      "num_tokens": 188781580.0,
+      "step": 1825
+    },
+    {
+      "epoch": 0.5622673049160398,
+      "grad_norm": 0.9014413952827454,
+      "learning_rate": 2.2579128799749297e-06,
+      "loss": 1.1436,
+      "mean_token_accuracy": 0.7317037135362625,
+      "num_tokens": 191380677.0,
+      "step": 1850
+    },
+    {
+      "epoch": 0.5698655117392295,
+      "grad_norm": 0.8946366310119629,
+      "learning_rate": 2.2187402068317143e-06,
+      "loss": 1.1548,
+      "mean_token_accuracy": 0.7314022815227509,
+      "num_tokens": 193996266.0,
+      "step": 1875
+    },
+    {
+      "epoch": 0.5774637185624193,
+      "grad_norm": 0.9767251014709473,
+      "learning_rate": 2.179567533688499e-06,
+      "loss": 1.1258,
+      "mean_token_accuracy": 0.737195520401001,
+      "num_tokens": 196548577.0,
+      "step": 1900
+    },
+    {
+      "epoch": 0.585061925385609,
+      "grad_norm": 1.0095189809799194,
+      "learning_rate": 2.1403948605452837e-06,
+      "loss": 1.1481,
+      "mean_token_accuracy": 0.7341015815734864,
+      "num_tokens": 199121126.0,
+      "step": 1925
+    },
+    {
+      "epoch": 0.5926601322087988,
+      "grad_norm": 0.9052937030792236,
+      "learning_rate": 2.1012221874020688e-06,
+      "loss": 1.1309,
+      "mean_token_accuracy": 0.7348120081424713,
+      "num_tokens": 201731098.0,
+      "step": 1950
+    },
+    {
+      "epoch": 0.6002583390319884,
+      "grad_norm": 0.9695770144462585,
+      "learning_rate": 2.062049514258853e-06,
+      "loss": 1.1248,
+      "mean_token_accuracy": 0.7372716355323792,
+      "num_tokens": 204324674.0,
+      "step": 1975
+    },
+    {
+      "epoch": 0.6078565458551782,
+      "grad_norm": 0.9187005758285522,
+      "learning_rate": 2.022876841115638e-06,
+      "loss": 1.1275,
+      "mean_token_accuracy": 0.7356339997053146,
+      "num_tokens": 206937271.0,
+      "step": 2000
+    },
+    {
+      "epoch": 0.6154547526783679,
+      "grad_norm": 0.9396357536315918,
+      "learning_rate": 1.9837041679724223e-06,
+      "loss": 1.1475,
+      "mean_token_accuracy": 0.7320456486940384,
+      "num_tokens": 209530276.0,
+      "step": 2025
+    },
+    {
+      "epoch": 0.6230529595015576,
+      "grad_norm": 0.9308034181594849,
+      "learning_rate": 1.9445314948292074e-06,
+      "loss": 1.107,
+      "mean_token_accuracy": 0.7407030069828033,
+      "num_tokens": 212127023.0,
+      "step": 2050
+    },
+    {
+      "epoch": 0.6306511663247474,
+      "grad_norm": 0.9127600193023682,
+      "learning_rate": 1.9053588216859919e-06,
+      "loss": 1.117,
+      "mean_token_accuracy": 0.7379663151502609,
+      "num_tokens": 214698253.0,
+      "step": 2075
+    },
+    {
+      "epoch": 0.6382493731479371,
+      "grad_norm": 1.0522152185440063,
+      "learning_rate": 1.8661861485427768e-06,
+      "loss": 1.1134,
+      "mean_token_accuracy": 0.738114013671875,
+      "num_tokens": 217256660.0,
+      "step": 2100
+    },
+    {
+      "epoch": 0.6458475799711269,
+      "grad_norm": 1.1676892042160034,
+      "learning_rate": 1.8270134753995614e-06,
+      "loss": 1.1501,
+      "mean_token_accuracy": 0.7336319923400879,
+      "num_tokens": 219876024.0,
+      "step": 2125
+    },
+    {
+      "epoch": 0.6534457867943165,
+      "grad_norm": 0.950326144695282,
+      "learning_rate": 1.7878408022563463e-06,
+      "loss": 1.1027,
+      "mean_token_accuracy": 0.7408947384357453,
+      "num_tokens": 222494174.0,
+      "step": 2150
+    },
+    {
+      "epoch": 0.6610439936175063,
+      "grad_norm": 0.9229258298873901,
+      "learning_rate": 1.7486681291131308e-06,
+      "loss": 1.149,
+      "mean_token_accuracy": 0.7321878397464752,
+      "num_tokens": 225084016.0,
+      "step": 2175
+    },
+    {
+      "epoch": 0.668642200440696,
+      "grad_norm": 0.8653574585914612,
+      "learning_rate": 1.7094954559699156e-06,
+      "loss": 1.1264,
+      "mean_token_accuracy": 0.7363163530826569,
+      "num_tokens": 227745579.0,
+      "step": 2200
+    },
+    {
+      "epoch": 0.6762404072638857,
+      "grad_norm": 0.8321031332015991,
+      "learning_rate": 1.6703227828267e-06,
+      "loss": 1.1262,
+      "mean_token_accuracy": 0.7373967486619949,
+      "num_tokens": 230365604.0,
+      "step": 2225
+    },
+    {
+      "epoch": 0.6838386140870755,
+      "grad_norm": 0.9620904326438904,
+      "learning_rate": 1.631150109683485e-06,
+      "loss": 1.104,
+      "mean_token_accuracy": 0.7394238811731338,
+      "num_tokens": 232943997.0,
+      "step": 2250
+    },
+    {
+      "epoch": 0.6914368209102651,
+      "grad_norm": 0.9633401036262512,
+      "learning_rate": 1.5919774365402697e-06,
+      "loss": 1.1312,
+      "mean_token_accuracy": 0.7345698297023773,
+      "num_tokens": 235514397.0,
+      "step": 2275
+    },
+    {
+      "epoch": 0.6990350277334549,
+      "grad_norm": 1.049621343612671,
+      "learning_rate": 1.5528047633970545e-06,
+      "loss": 1.1038,
+      "mean_token_accuracy": 0.7410222667455674,
+      "num_tokens": 237958000.0,
+      "step": 2300
+    },
+    {
+      "epoch": 0.7066332345566446,
+      "grad_norm": 0.8942002654075623,
+      "learning_rate": 1.513632090253839e-06,
+      "loss": 1.128,
+      "mean_token_accuracy": 0.7355525487661362,
+      "num_tokens": 240563561.0,
+      "step": 2325
+    },
+    {
+      "epoch": 0.7142314413798344,
+      "grad_norm": 1.0083601474761963,
+      "learning_rate": 1.4744594171106239e-06,
+      "loss": 1.1763,
+      "mean_token_accuracy": 0.7265212672948838,
+      "num_tokens": 243194983.0,
+      "step": 2350
+    },
+    {
+      "epoch": 0.7218296482030241,
+      "grad_norm": 0.8960743546485901,
+      "learning_rate": 1.4352867439674083e-06,
+      "loss": 1.1586,
+      "mean_token_accuracy": 0.7288067770004273,
+      "num_tokens": 245824173.0,
+      "step": 2375
+    },
+    {
+      "epoch": 0.7294278550262138,
+      "grad_norm": 0.9686855673789978,
+      "learning_rate": 1.3961140708241932e-06,
+      "loss": 1.1404,
+      "mean_token_accuracy": 0.732091948390007,
+      "num_tokens": 248428880.0,
+      "step": 2400
+    },
+    {
+      "epoch": 0.7370260618494036,
+      "grad_norm": 0.8258799910545349,
+      "learning_rate": 1.3569413976809779e-06,
+      "loss": 1.1119,
+      "mean_token_accuracy": 0.7375982666015625,
+      "num_tokens": 251034901.0,
+      "step": 2425
+    },
+    {
+      "epoch": 0.7446242686725932,
+      "grad_norm": 0.9364578127861023,
+      "learning_rate": 1.3177687245377627e-06,
+      "loss": 1.162,
+      "mean_token_accuracy": 0.7280535787343979,
+      "num_tokens": 253656324.0,
+      "step": 2450
+    },
+    {
+      "epoch": 0.752222475495783,
+      "grad_norm": 0.8600585460662842,
+      "learning_rate": 1.2785960513945472e-06,
+      "loss": 1.0984,
+      "mean_token_accuracy": 0.7421051144599915,
+      "num_tokens": 256255282.0,
+      "step": 2475
+    },
+    {
+      "epoch": 0.7598206823189727,
+      "grad_norm": 0.955594003200531,
+      "learning_rate": 1.2394233782513319e-06,
+      "loss": 1.1248,
+      "mean_token_accuracy": 0.7350970155000687,
+      "num_tokens": 258805575.0,
+      "step": 2500
+    },
+    {
+      "epoch": 0.7674188891421625,
+      "grad_norm": 0.8654617667198181,
+      "learning_rate": 1.2002507051081167e-06,
+      "loss": 1.139,
+      "mean_token_accuracy": 0.7342792183160782,
+      "num_tokens": 261394940.0,
+      "step": 2525
+    },
+    {
+      "epoch": 0.7750170959653522,
+      "grad_norm": 0.9037804007530212,
+      "learning_rate": 1.1610780319649014e-06,
+      "loss": 1.1507,
+      "mean_token_accuracy": 0.731409004330635,
+      "num_tokens": 264013243.0,
+      "step": 2550
+    },
+    {
+      "epoch": 0.7826153027885419,
+      "grad_norm": 0.9133325815200806,
+      "learning_rate": 1.121905358821686e-06,
+      "loss": 1.097,
+      "mean_token_accuracy": 0.7419029778242111,
+      "num_tokens": 266628788.0,
+      "step": 2575
+    },
+    {
+      "epoch": 0.7902135096117316,
+      "grad_norm": 0.960132360458374,
+      "learning_rate": 1.0827326856784708e-06,
+      "loss": 1.1012,
+      "mean_token_accuracy": 0.7402530944347382,
+      "num_tokens": 269174466.0,
+      "step": 2600
+    },
+    {
+      "epoch": 0.7978117164349213,
+      "grad_norm": 1.2616750001907349,
+      "learning_rate": 1.0435600125352554e-06,
+      "loss": 1.1035,
+      "mean_token_accuracy": 0.7417422020435334,
+      "num_tokens": 271735962.0,
+      "step": 2625
+    },
+    {
+      "epoch": 0.8054099232581111,
+      "grad_norm": 0.9597173929214478,
+      "learning_rate": 1.00438733939204e-06,
+      "loss": 1.0966,
+      "mean_token_accuracy": 0.7435225594043732,
+      "num_tokens": 274392874.0,
+      "step": 2650
+    },
+    {
+      "epoch": 0.8130081300813008,
+      "grad_norm": 0.8417394161224365,
+      "learning_rate": 9.65214666248825e-07,
+      "loss": 1.1347,
+      "mean_token_accuracy": 0.7362286591529846,
+      "num_tokens": 276982423.0,
+      "step": 2675
+    },
+    {
+      "epoch": 0.8206063369044906,
+      "grad_norm": 0.8907487392425537,
+      "learning_rate": 9.260419931056095e-07,
+      "loss": 1.1112,
+      "mean_token_accuracy": 0.7379048210382462,
+      "num_tokens": 279629729.0,
+      "step": 2700
+    },
+    {
+      "epoch": 0.8282045437276803,
+      "grad_norm": 0.9732270240783691,
+      "learning_rate": 8.868693199623943e-07,
+      "loss": 1.1219,
+      "mean_token_accuracy": 0.7383491581678391,
+      "num_tokens": 282200981.0,
+      "step": 2725
+    },
+    {
+      "epoch": 0.8358027505508699,
+      "grad_norm": 0.8799415230751038,
+      "learning_rate": 8.47696646819179e-07,
+      "loss": 1.1059,
+      "mean_token_accuracy": 0.7389830541610718,
+      "num_tokens": 284699697.0,
+      "step": 2750
+    },
+    {
+      "epoch": 0.8434009573740597,
+      "grad_norm": 0.8372652530670166,
+      "learning_rate": 8.085239736759637e-07,
+      "loss": 1.123,
+      "mean_token_accuracy": 0.7350180590152741,
+      "num_tokens": 287274170.0,
+      "step": 2775
+    },
+    {
+      "epoch": 0.8509991641972494,
+      "grad_norm": 0.8474750518798828,
+      "learning_rate": 7.693513005327484e-07,
+      "loss": 1.1316,
+      "mean_token_accuracy": 0.735204011797905,
+      "num_tokens": 289905267.0,
+      "step": 2800
+    },
+    {
+      "epoch": 0.8585973710204392,
+      "grad_norm": 0.9228349924087524,
+      "learning_rate": 7.301786273895331e-07,
+      "loss": 1.1276,
+      "mean_token_accuracy": 0.7360332900285721,
+      "num_tokens": 292531690.0,
+      "step": 2825
+    },
+    {
+      "epoch": 0.8661955778436289,
+      "grad_norm": 0.9425334334373474,
+      "learning_rate": 6.910059542463178e-07,
+      "loss": 1.1446,
+      "mean_token_accuracy": 0.7327738231420518,
+      "num_tokens": 295102204.0,
+      "step": 2850
+    },
+    {
+      "epoch": 0.8737937846668187,
+      "grad_norm": 0.8812386393547058,
+      "learning_rate": 6.518332811031025e-07,
+      "loss": 1.1197,
+      "mean_token_accuracy": 0.7385855436325073,
+      "num_tokens": 297696866.0,
+      "step": 2875
+    },
+    {
+      "epoch": 0.8813919914900084,
+      "grad_norm": 0.8696539998054504,
+      "learning_rate": 6.126606079598872e-07,
+      "loss": 1.1084,
+      "mean_token_accuracy": 0.740631007552147,
+      "num_tokens": 300294510.0,
+      "step": 2900
+    },
+    {
+      "epoch": 0.888990198313198,
+      "grad_norm": 0.9522122144699097,
+      "learning_rate": 5.73487934816672e-07,
+      "loss": 1.1582,
+      "mean_token_accuracy": 0.7323483002185821,
+      "num_tokens": 302898731.0,
+      "step": 2925
+    },
+    {
+      "epoch": 0.8965884051363878,
+      "grad_norm": 0.9082636833190918,
+      "learning_rate": 5.343152616734566e-07,
+      "loss": 1.1264,
+      "mean_token_accuracy": 0.7359338957071304,
+      "num_tokens": 305410011.0,
+      "step": 2950
+    },
+    {
+      "epoch": 0.9041866119595775,
+      "grad_norm": 0.9275860786437988,
+      "learning_rate": 4.951425885302413e-07,
+      "loss": 1.1528,
+      "mean_token_accuracy": 0.7339082890748978,
+      "num_tokens": 308110984.0,
+      "step": 2975
+    },
+    {
+      "epoch": 0.9117848187827673,
+      "grad_norm": 1.0664831399917603,
+      "learning_rate": 4.55969915387026e-07,
+      "loss": 1.1404,
+      "mean_token_accuracy": 0.7343163812160491,
+      "num_tokens": 310767815.0,
+      "step": 3000
+    },
+    {
+      "epoch": 0.919383025605957,
+      "grad_norm": 0.9530277252197266,
+      "learning_rate": 4.1679724224381073e-07,
+      "loss": 1.1293,
+      "mean_token_accuracy": 0.7350982189178467,
+      "num_tokens": 313289189.0,
+      "step": 3025
+    },
+    {
+      "epoch": 0.9269812324291468,
+      "grad_norm": 0.8777811527252197,
+      "learning_rate": 3.7762456910059545e-07,
+      "loss": 1.1222,
+      "mean_token_accuracy": 0.7371740919351578,
+      "num_tokens": 315885578.0,
+      "step": 3050
+    },
+    {
+      "epoch": 0.9345794392523364,
+      "grad_norm": 0.9783887267112732,
+      "learning_rate": 3.3845189595738017e-07,
+      "loss": 1.1018,
+      "mean_token_accuracy": 0.7426222825050354,
+      "num_tokens": 318443917.0,
+      "step": 3075
+    },
+    {
+      "epoch": 0.9421776460755261,
+      "grad_norm": 0.8978294730186462,
+      "learning_rate": 2.9927922281416484e-07,
+      "loss": 1.1463,
+      "mean_token_accuracy": 0.7316457951068878,
+      "num_tokens": 321002620.0,
+      "step": 3100
+    },
+    {
+      "epoch": 0.9497758528987159,
+      "grad_norm": 0.9735873937606812,
+      "learning_rate": 2.6010654967094956e-07,
+      "loss": 1.1119,
+      "mean_token_accuracy": 0.7383992117643357,
+      "num_tokens": 323561038.0,
+      "step": 3125
+    },
+    {
+      "epoch": 0.9573740597219056,
+      "grad_norm": 0.8766360282897949,
+      "learning_rate": 2.2093387652773425e-07,
+      "loss": 1.0943,
+      "mean_token_accuracy": 0.7416167676448822,
+      "num_tokens": 326152504.0,
+      "step": 3150
+    },
+    {
+      "epoch": 0.9649722665450954,
+      "grad_norm": 0.8869988322257996,
+      "learning_rate": 1.8176120338451897e-07,
+      "loss": 1.1099,
+      "mean_token_accuracy": 0.7376412642002106,
+      "num_tokens": 328738142.0,
+      "step": 3175
+    },
+    {
+      "epoch": 0.9725704733682851,
+      "grad_norm": 0.9266097545623779,
+      "learning_rate": 1.4258853024130367e-07,
+      "loss": 1.1165,
+      "mean_token_accuracy": 0.7394632256031036,
+      "num_tokens": 331295732.0,
+      "step": 3200
+    },
+    {
+      "epoch": 0.9801686801914749,
+      "grad_norm": 0.9964390993118286,
+      "learning_rate": 1.0341585709808838e-07,
+      "loss": 1.1191,
+      "mean_token_accuracy": 0.7373811560869217,
+      "num_tokens": 333891414.0,
+      "step": 3225
+    },
+    {
+      "epoch": 0.9877668870146645,
+      "grad_norm": 0.9817653298377991,
+      "learning_rate": 6.424318395487308e-08,
+      "loss": 1.1356,
+      "mean_token_accuracy": 0.7354251599311828,
+      "num_tokens": 336538563.0,
+      "step": 3250
+    },
+    {
+      "epoch": 0.9953650938378543,
+      "grad_norm": 0.8675805330276489,
+      "learning_rate": 2.507051081165779e-08,
+      "loss": 1.1482,
+      "mean_token_accuracy": 0.7319550043344498,
+      "num_tokens": 339216210.0,
+      "step": 3275
+    }
+  ],
+  "logging_steps": 25,
+  "max_steps": 3290,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.8698621709430292e+18,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56ed8a20d75838e68b4b9c64cec17c2d7ad2d9597671b5bf4a16124c4f707494
+size 7352

zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 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 . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# 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.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+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_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device, weights_only=False)
+
+        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]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+    total_files = len(files)
+    state_dicts = []
+    for f in tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    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}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] 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, exclude_frozen_parameters):
+    """
+    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_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # 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.")
+
+    # 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() if _has_callable(shape, 'numel') else math.prod(shape)
+            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")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    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 _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
+
+    # 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
+
+    # 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
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # 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
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+        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"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        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 Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    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.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    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")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    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_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                                          tag,
+                                                          exclude_frozen_parameters,
+                                                          lazy_mode=True)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+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_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    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_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)