Upload folder using huggingface_hub

.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
+ckpt-80/tokenizer.json filter=lfs diff=lfs merge=lfs -text

ckpt-80/config.json

@@ -0,0 +1,30 @@
+{
+  "_name_or_path": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 151646,
+  "eos_token_id": 151643,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 8960,
+  "max_position_embeddings": 131072,
+  "max_window_layers": 21,
+  "model_type": "qwen2",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.48.1",
+  "use_cache": false,
+  "use_mrope": false,
+  "use_sliding_window": false,
+  "vocab_size": 151665
+}

ckpt-80/generation_config.json

@@ -0,0 +1,9 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 151646,
+  "do_sample": true,
+  "eos_token_id": 151643,
+  "temperature": 0.6,
+  "top_p": 0.95,
+  "transformers_version": "4.48.1"
+}

ckpt-80/latest

@@ -0,0 +1 @@
+global_step80

ckpt-80/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d931e9e5b724a25ccf55888fb8e07a49bdd22545eaae34e42b1c50e5a1fdbdc3
+size 3552549728

ckpt-80/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6baf06104b269181ea5d4120ca1ef39775b39f1a989deefdffedb75191849e7c
+size 14960

ckpt-80/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53621d02c8b7b14481ebb6d202cd4f6bf26d8a2753884ba86ed4f8a8edc9f3ab
+size 14960

ckpt-80/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3e6cc02a4108cccb70308c26b649ff4c1a93250a6252d7dd31fda6f995317d9
+size 14960

ckpt-80/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad58e4beae534f7dc5df6c05453c33a74665be802c015bbe4dabaa92dae6f70d
+size 14960

ckpt-80/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f108534be0dc23a9323e6fef6fe315e081f6150b4ca4cdadba69918370b4584
+size 1064

ckpt-80/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<｜begin▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<｜end▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

ckpt-80/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e20ddafc659ba90242154b55275402edeca0715e5dbb30f56815a4ce081f4893
+size 11422778

ckpt-80/tokenizer_config.json

@@ -0,0 +1,195 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<｜end▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<｜User｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151645": {
+      "content": "<｜Assistant｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151646": {
+      "content": "<｜begin▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|EOT|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151648": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151649": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "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
+    }
+  },
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}\n    {%- endif %}\n    {{- \"\\n\\n# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "extra_special_tokens": {},
+  "legacy": true,
+  "model_max_length": 16384,
+  "pad_token": "<｜end▁of▁sentence｜>",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": null,
+  "use_default_system_prompt": false
+}

ckpt-80/trainer_state.json

@@ -0,0 +1,1473 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.33350703491401773,
+  "eval_steps": 500,
+  "global_step": 80,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.004168837936425221,
+      "grad_norm": 2.006458271799073,
+      "learning_rate": 2.0000000000000003e-06,
+      "log_odds_chosen": -0.375732421875,
+      "log_odds_ratio": -0.9052734375,
+      "logits/chosen": 0.669158935546875,
+      "logits/rejected": 0.4799842834472656,
+      "logps/chosen": -0.84521484375,
+      "logps/rejected": -0.636474609375,
+      "loss": 16.0332,
+      "nll_loss": 1.072998046875,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0845794677734375,
+      "rewards/margins": -0.0209197998046875,
+      "rewards/rejected": -0.06365966796875,
+      "step": 1
+    },
+    {
+      "epoch": 0.008337675872850442,
+      "grad_norm": 2.0306125193054982,
+      "learning_rate": 4.000000000000001e-06,
+      "log_odds_chosen": -0.347900390625,
+      "log_odds_ratio": -0.88427734375,
+      "logits/chosen": 0.18726348876953125,
+      "logits/rejected": 0.063751220703125,
+      "logps/chosen": -0.598388671875,
+      "logps/rejected": -0.456298828125,
+      "loss": 15.582,
+      "nll_loss": 0.7652587890625,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0598602294921875,
+      "rewards/margins": -0.0142364501953125,
+      "rewards/rejected": -0.045623779296875,
+      "step": 2
+    },
+    {
+      "epoch": 0.012506513809275664,
+      "grad_norm": 1.9051609121742616,
+      "learning_rate": 6e-06,
+      "log_odds_chosen": -0.397216796875,
+      "log_odds_ratio": -0.9169921875,
+      "logits/chosen": 0.2111358642578125,
+      "logits/rejected": 0.04972648620605469,
+      "logps/chosen": -0.76123046875,
+      "logps/rejected": -0.55908203125,
+      "loss": 14.3369,
+      "nll_loss": 0.921142578125,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.076141357421875,
+      "rewards/margins": -0.0202484130859375,
+      "rewards/rejected": -0.0558929443359375,
+      "step": 3
+    },
+    {
+      "epoch": 0.016675351745700884,
+      "grad_norm": 1.78874596409052,
+      "learning_rate": 8.000000000000001e-06,
+      "log_odds_chosen": -0.355712890625,
+      "log_odds_ratio": -0.8896484375,
+      "logits/chosen": 0.1468048095703125,
+      "logits/rejected": -0.00798797607421875,
+      "logps/chosen": -0.6734619140625,
+      "logps/rejected": -0.51214599609375,
+      "loss": 15.6787,
+      "nll_loss": 0.810302734375,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0673370361328125,
+      "rewards/margins": -0.01611328125,
+      "rewards/rejected": -0.0512237548828125,
+      "step": 4
+    },
+    {
+      "epoch": 0.020844189682126108,
+      "grad_norm": 1.5688561088336885,
+      "learning_rate": 1e-05,
+      "log_odds_chosen": -0.29296875,
+      "log_odds_ratio": -0.853271484375,
+      "logits/chosen": 0.534515380859375,
+      "logits/rejected": 0.4119873046875,
+      "logps/chosen": -0.6806640625,
+      "logps/rejected": -0.544921875,
+      "loss": 15.1973,
+      "nll_loss": 0.8408203125,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0680389404296875,
+      "rewards/margins": -0.0135498046875,
+      "rewards/rejected": -0.0544891357421875,
+      "step": 5
+    },
+    {
+      "epoch": 0.02501302761855133,
+      "grad_norm": 1.3268998143676978,
+      "learning_rate": 1.2e-05,
+      "log_odds_chosen": -0.41455078125,
+      "log_odds_ratio": -0.9267578125,
+      "logits/chosen": 0.428436279296875,
+      "logits/rejected": 0.2169036865234375,
+      "logps/chosen": -0.7254638671875,
+      "logps/rejected": -0.5296630859375,
+      "loss": 15.9805,
+      "nll_loss": 0.88427734375,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0725555419921875,
+      "rewards/margins": -0.01958465576171875,
+      "rewards/rejected": -0.05297088623046875,
+      "step": 6
+    },
+    {
+      "epoch": 0.02918186555497655,
+      "grad_norm": 1.3269961863715432,
+      "learning_rate": 1.4e-05,
+      "log_odds_chosen": -0.3531494140625,
+      "log_odds_ratio": -0.888427734375,
+      "logits/chosen": 0.6148757934570312,
+      "logits/rejected": 0.463531494140625,
+      "logps/chosen": -0.7569580078125,
+      "logps/rejected": -0.5888671875,
+      "loss": 15.6689,
+      "nll_loss": 0.8975830078125,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0756683349609375,
+      "rewards/margins": -0.016815185546875,
+      "rewards/rejected": -0.0588531494140625,
+      "step": 7
+    },
+    {
+      "epoch": 0.03335070349140177,
+      "grad_norm": 0.9711343808401325,
+      "learning_rate": 1.6000000000000003e-05,
+      "log_odds_chosen": -0.35888671875,
+      "log_odds_ratio": -0.892578125,
+      "logits/chosen": 0.6737632751464844,
+      "logits/rejected": 0.534698486328125,
+      "logps/chosen": -0.7044677734375,
+      "logps/rejected": -0.548828125,
+      "loss": 14.3604,
+      "nll_loss": 0.825439453125,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0704498291015625,
+      "rewards/margins": -0.01556396484375,
+      "rewards/rejected": -0.0548858642578125,
+      "step": 8
+    },
+    {
+      "epoch": 0.03751954142782699,
+      "grad_norm": 0.9237098587195779,
+      "learning_rate": 1.8e-05,
+      "log_odds_chosen": -0.34375,
+      "log_odds_ratio": -0.882080078125,
+      "logits/chosen": 0.6319732666015625,
+      "logits/rejected": 0.4871063232421875,
+      "logps/chosen": -0.7308349609375,
+      "logps/rejected": -0.5621337890625,
+      "loss": 14.1436,
+      "nll_loss": 0.8466796875,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0731201171875,
+      "rewards/margins": -0.01691436767578125,
+      "rewards/rejected": -0.05620574951171875,
+      "step": 9
+    },
+    {
+      "epoch": 0.041688379364252216,
+      "grad_norm": 0.8445328349436582,
+      "learning_rate": 2e-05,
+      "log_odds_chosen": -0.3358154296875,
+      "log_odds_ratio": -0.879150390625,
+      "logits/chosen": 0.501800537109375,
+      "logits/rejected": 0.3776397705078125,
+      "logps/chosen": -0.6666259765625,
+      "logps/rejected": -0.51416015625,
+      "loss": 12.7095,
+      "nll_loss": 0.76318359375,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0666351318359375,
+      "rewards/margins": -0.01522064208984375,
+      "rewards/rejected": -0.05141448974609375,
+      "step": 10
+    },
+    {
+      "epoch": 0.04585721730067743,
+      "grad_norm": 0.8396125768364633,
+      "learning_rate": 1.9999067159645222e-05,
+      "log_odds_chosen": -0.384765625,
+      "log_odds_ratio": -0.908447265625,
+      "logits/chosen": 0.563262939453125,
+      "logits/rejected": 0.3618621826171875,
+      "logps/chosen": -0.813720703125,
+      "logps/rejected": -0.6082763671875,
+      "loss": 14.0229,
+      "nll_loss": 0.898193359375,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0813140869140625,
+      "rewards/margins": -0.0204925537109375,
+      "rewards/rejected": -0.060821533203125,
+      "step": 11
+    },
+    {
+      "epoch": 0.05002605523710266,
+      "grad_norm": 0.8291225655579284,
+      "learning_rate": 1.999626881261911e-05,
+      "log_odds_chosen": -0.33319091796875,
+      "log_odds_ratio": -0.877197265625,
+      "logits/chosen": 0.5009918212890625,
+      "logits/rejected": 0.38329315185546875,
+      "logps/chosen": -0.680419921875,
+      "logps/rejected": -0.53778076171875,
+      "loss": 13.0962,
+      "nll_loss": 0.78662109375,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.06806182861328125,
+      "rewards/margins": -0.01427459716796875,
+      "rewards/rejected": -0.0537872314453125,
+      "step": 12
+    },
+    {
+      "epoch": 0.05419489317352788,
+      "grad_norm": 0.6447061073782809,
+      "learning_rate": 1.9991605481003865e-05,
+      "log_odds_chosen": -0.2783203125,
+      "log_odds_ratio": -0.846923828125,
+      "logits/chosen": 0.17709732055664062,
+      "logits/rejected": 0.054500579833984375,
+      "logps/chosen": -0.6295166015625,
+      "logps/rejected": -0.5006103515625,
+      "loss": 12.0044,
+      "nll_loss": 0.6942138671875,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.062957763671875,
+      "rewards/margins": -0.0128936767578125,
+      "rewards/rejected": -0.0500640869140625,
+      "step": 13
+    },
+    {
+      "epoch": 0.0583637311099531,
+      "grad_norm": 0.7554659875392797,
+      "learning_rate": 1.998507803482828e-05,
+      "log_odds_chosen": -0.246826171875,
+      "log_odds_ratio": -0.829345703125,
+      "logits/chosen": -0.043701171875,
+      "logits/rejected": -0.208831787109375,
+      "logps/chosen": -0.7122802734375,
+      "logps/rejected": -0.58056640625,
+      "loss": 11.9097,
+      "nll_loss": 0.7593994140625,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": -0.0711822509765625,
+      "rewards/margins": -0.0131378173828125,
+      "rewards/rejected": -0.05804443359375,
+      "step": 14
+    },
+    {
+      "epoch": 0.06253256904637833,
+      "grad_norm": 0.7898038988103401,
+      "learning_rate": 1.9976687691905394e-05,
+      "log_odds_chosen": -0.0479736328125,
+      "log_odds_ratio": -0.71923828125,
+      "logits/chosen": -0.0294342041015625,
+      "logits/rejected": -0.11614990234375,
+      "logps/chosen": -0.5894775390625,
+      "logps/rejected": -0.5662841796875,
+      "loss": 12.1235,
+      "nll_loss": 0.68035888671875,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.05896759033203125,
+      "rewards/margins": -0.002349853515625,
+      "rewards/rejected": -0.05661773681640625,
+      "step": 15
+    },
+    {
+      "epoch": 0.06670140698280354,
+      "grad_norm": 0.6995568607969461,
+      "learning_rate": 1.9966436017605296e-05,
+      "log_odds_chosen": 0.0357666015625,
+      "log_odds_ratio": -0.694580078125,
+      "logits/chosen": 0.13065338134765625,
+      "logits/rejected": -0.0316619873046875,
+      "logps/chosen": -0.6864013671875,
+      "logps/rejected": -0.66455078125,
+      "loss": 11.8877,
+      "nll_loss": 0.76220703125,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.0686492919921875,
+      "rewards/margins": -0.002166748046875,
+      "rewards/rejected": -0.0664825439453125,
+      "step": 16
+    },
+    {
+      "epoch": 0.07087024491922876,
+      "grad_norm": 0.6249952276948901,
+      "learning_rate": 1.9954324924563088e-05,
+      "log_odds_chosen": 0.3499755859375,
+      "log_odds_ratio": -0.5440673828125,
+      "logits/chosen": -0.125030517578125,
+      "logits/rejected": -0.2731895446777344,
+      "logps/chosen": -0.5921630859375,
+      "logps/rejected": -0.7261962890625,
+      "loss": 11.9478,
+      "nll_loss": 0.6683349609375,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -0.05919647216796875,
+      "rewards/margins": 0.01343536376953125,
+      "rewards/rejected": -0.0726318359375,
+      "step": 17
+    },
+    {
+      "epoch": 0.07503908285565399,
+      "grad_norm": 0.5476240540355588,
+      "learning_rate": 1.9940356672322037e-05,
+      "log_odds_chosen": 0.38232421875,
+      "log_odds_ratio": -0.5316162109375,
+      "logits/chosen": 0.2089080810546875,
+      "logits/rejected": 0.069793701171875,
+      "logps/chosen": -0.5899658203125,
+      "logps/rejected": -0.7808837890625,
+      "loss": 11.967,
+      "nll_loss": 0.6422119140625,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.0589752197265625,
+      "rewards/margins": 0.0191497802734375,
+      "rewards/rejected": -0.078125,
+      "step": 18
+    },
+    {
+      "epoch": 0.07920792079207921,
+      "grad_norm": 0.49418444669956524,
+      "learning_rate": 1.9924533866912017e-05,
+      "log_odds_chosen": 0.64111328125,
+      "log_odds_ratio": -0.435546875,
+      "logits/chosen": 0.16015625,
+      "logits/rejected": 0.032745361328125,
+      "logps/chosen": -0.5224609375,
+      "logps/rejected": -0.823486328125,
+      "loss": 10.4492,
+      "nll_loss": 0.5810546875,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.05222320556640625,
+      "rewards/margins": 0.03015899658203125,
+      "rewards/rejected": -0.0823822021484375,
+      "step": 19
+    },
+    {
+      "epoch": 0.08337675872850443,
+      "grad_norm": 0.4118961781762262,
+      "learning_rate": 1.9906859460363307e-05,
+      "log_odds_chosen": 0.6845703125,
+      "log_odds_ratio": -0.42779541015625,
+      "logits/chosen": -0.066619873046875,
+      "logits/rejected": -0.18597412109375,
+      "logps/chosen": -0.5198974609375,
+      "logps/rejected": -0.866455078125,
+      "loss": 10.5518,
+      "nll_loss": 0.5537109375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.052001953125,
+      "rewards/margins": 0.03460693359375,
+      "rewards/rejected": -0.08660888671875,
+      "step": 20
+    },
+    {
+      "epoch": 0.08754559666492966,
+      "grad_norm": 0.44268649674883015,
+      "learning_rate": 1.988733675015585e-05,
+      "log_odds_chosen": 0.911376953125,
+      "log_odds_ratio": -0.35015869140625,
+      "logits/chosen": 0.190277099609375,
+      "logits/rejected": -0.041168212890625,
+      "logps/chosen": -0.8385009765625,
+      "logps/rejected": -1.418212890625,
+      "loss": 12.3477,
+      "nll_loss": 0.89599609375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0838623046875,
+      "rewards/margins": 0.05804443359375,
+      "rewards/rejected": -0.14190673828125,
+      "step": 21
+    },
+    {
+      "epoch": 0.09171443460135487,
+      "grad_norm": 0.4486387895857316,
+      "learning_rate": 1.9865969378604023e-05,
+      "log_odds_chosen": 0.932373046875,
+      "log_odds_ratio": -0.340576171875,
+      "logits/chosen": -0.33026123046875,
+      "logits/rejected": -0.42425537109375,
+      "logps/chosen": -0.4442138671875,
+      "logps/rejected": -0.883544921875,
+      "loss": 10.5381,
+      "nll_loss": 0.4703369140625,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04442596435546875,
+      "rewards/margins": 0.0439300537109375,
+      "rewards/rejected": -0.088348388671875,
+      "step": 22
+    },
+    {
+      "epoch": 0.09588327253778009,
+      "grad_norm": 0.4080983278603557,
+      "learning_rate": 1.9842761332177115e-05,
+      "log_odds_chosen": 0.9261474609375,
+      "log_odds_ratio": -0.35150146484375,
+      "logits/chosen": -0.09943103790283203,
+      "logits/rejected": -0.194610595703125,
+      "logps/chosen": -0.5177001953125,
+      "logps/rejected": -0.9964599609375,
+      "loss": 10.3979,
+      "nll_loss": 0.560546875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0517425537109375,
+      "rewards/margins": 0.04787445068359375,
+      "rewards/rejected": -0.0996246337890625,
+      "step": 23
+    },
+    {
+      "epoch": 0.10005211047420531,
+      "grad_norm": 0.46777770854029893,
+      "learning_rate": 1.9817716940755586e-05,
+      "log_odds_chosen": 1.640869140625,
+      "log_odds_ratio": -0.2108612060546875,
+      "logits/chosen": -0.008392333984375,
+      "logits/rejected": -0.141082763671875,
+      "logps/chosen": -0.7047119140625,
+      "logps/rejected": -1.828857421875,
+      "loss": 11.5957,
+      "nll_loss": 0.774169921875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0703887939453125,
+      "rewards/margins": 0.11248779296875,
+      "rewards/rejected": -0.182952880859375,
+      "step": 24
+    },
+    {
+      "epoch": 0.10422094841063054,
+      "grad_norm": 0.3738919624381692,
+      "learning_rate": 1.979084087682323e-05,
+      "log_odds_chosen": 1.23486328125,
+      "log_odds_ratio": -0.2798309326171875,
+      "logits/chosen": -0.10467529296875,
+      "logits/rejected": -0.171875,
+      "logps/chosen": -0.482421875,
+      "logps/rejected": -1.177490234375,
+      "loss": 9.4116,
+      "nll_loss": 0.53662109375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0482177734375,
+      "rewards/margins": 0.0694732666015625,
+      "rewards/rejected": -0.1176910400390625,
+      "step": 25
+    },
+    {
+      "epoch": 0.10838978634705576,
+      "grad_norm": 0.37340979877433916,
+      "learning_rate": 1.9762138154595448e-05,
+      "log_odds_chosen": 1.1141357421875,
+      "log_odds_ratio": -0.31060791015625,
+      "logits/chosen": -0.01920318603515625,
+      "logits/rejected": -0.038970947265625,
+      "logps/chosen": -0.52880859375,
+      "logps/rejected": -1.1468505859375,
+      "loss": 10.0625,
+      "nll_loss": 0.5792236328125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.05289459228515625,
+      "rewards/margins": 0.06182098388671875,
+      "rewards/rejected": -0.1146697998046875,
+      "step": 26
+    },
+    {
+      "epoch": 0.11255862428348098,
+      "grad_norm": 0.341385013718231,
+      "learning_rate": 1.9731614129083756e-05,
+      "log_odds_chosen": 1.3974609375,
+      "log_odds_ratio": -0.2440338134765625,
+      "logits/chosen": 0.07257080078125,
+      "logits/rejected": 0.07470703125,
+      "logps/chosen": -0.56982421875,
+      "logps/rejected": -1.440185546875,
+      "loss": 9.9248,
+      "nll_loss": 0.64013671875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0569915771484375,
+      "rewards/margins": 0.0870208740234375,
+      "rewards/rejected": -0.14398193359375,
+      "step": 27
+    },
+    {
+      "epoch": 0.1167274622199062,
+      "grad_norm": 0.30885132002114984,
+      "learning_rate": 1.9699274495096712e-05,
+      "log_odds_chosen": 1.874755859375,
+      "log_odds_ratio": -0.1857452392578125,
+      "logits/chosen": 0.2138671875,
+      "logits/rejected": 0.263427734375,
+      "logps/chosen": -0.588134765625,
+      "logps/rejected": -1.86572265625,
+      "loss": 9.9067,
+      "nll_loss": 0.6661376953125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.058837890625,
+      "rewards/margins": 0.1278228759765625,
+      "rewards/rejected": -0.18658447265625,
+      "step": 28
+    },
+    {
+      "epoch": 0.12089630015633142,
+      "grad_norm": 0.2949038286121744,
+      "learning_rate": 1.9665125286177448e-05,
+      "log_odds_chosen": 1.40380859375,
+      "log_odds_ratio": -0.23876953125,
+      "logits/chosen": 0.0499420166015625,
+      "logits/rejected": 0.10356712341308594,
+      "logps/chosen": -0.5357666015625,
+      "logps/rejected": -1.36376953125,
+      "loss": 10.1011,
+      "nll_loss": 0.5960693359375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0535430908203125,
+      "rewards/margins": 0.0828399658203125,
+      "rewards/rejected": -0.136383056640625,
+      "step": 29
+    },
+    {
+      "epoch": 0.12506513809275666,
+      "grad_norm": 0.3151066899258469,
+      "learning_rate": 1.9629172873477995e-05,
+      "log_odds_chosen": 1.58837890625,
+      "log_odds_ratio": -0.21865081787109375,
+      "logits/chosen": -0.019866943359375,
+      "logits/rejected": 0.0438232421875,
+      "logps/chosen": -0.55517578125,
+      "logps/rejected": -1.543212890625,
+      "loss": 10.2285,
+      "nll_loss": 0.6005859375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.05548858642578125,
+      "rewards/margins": 0.0986175537109375,
+      "rewards/rejected": -0.154266357421875,
+      "step": 30
+    },
+    {
+      "epoch": 0.12923397602918185,
+      "grad_norm": 0.3455023253633968,
+      "learning_rate": 1.9591423964570634e-05,
+      "log_odds_chosen": 1.602294921875,
+      "log_odds_ratio": -0.215545654296875,
+      "logits/chosen": -0.253387451171875,
+      "logits/rejected": -0.179046630859375,
+      "logps/chosen": -0.44189453125,
+      "logps/rejected": -1.34130859375,
+      "loss": 9.7915,
+      "nll_loss": 0.476318359375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.044219970703125,
+      "rewards/margins": 0.089874267578125,
+      "rewards/rejected": -0.134063720703125,
+      "step": 31
+    },
+    {
+      "epoch": 0.13340281396560708,
+      "grad_norm": 0.32074624128845997,
+      "learning_rate": 1.9551885602196482e-05,
+      "log_odds_chosen": 1.82177734375,
+      "log_odds_ratio": -0.20027446746826172,
+      "logits/chosen": -0.19781494140625,
+      "logits/rejected": -0.0693511962890625,
+      "logps/chosen": -0.6165771484375,
+      "logps/rejected": -1.843017578125,
+      "loss": 9.4944,
+      "nll_loss": 0.66357421875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0616455078125,
+      "rewards/margins": 0.1227874755859375,
+      "rewards/rejected": -0.1843719482421875,
+      "step": 32
+    },
+    {
+      "epoch": 0.1375716519020323,
+      "grad_norm": 0.294138620125773,
+      "learning_rate": 1.9510565162951538e-05,
+      "log_odds_chosen": 1.5264892578125,
+      "log_odds_ratio": -0.214874267578125,
+      "logits/chosen": -0.17253494262695312,
+      "logits/rejected": -0.06893157958984375,
+      "logps/chosen": -0.4769287109375,
+      "logps/rejected": -1.3394775390625,
+      "loss": 8.6384,
+      "nll_loss": 0.505126953125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04773712158203125,
+      "rewards/margins": 0.0861053466796875,
+      "rewards/rejected": -0.1338653564453125,
+      "step": 33
+    },
+    {
+      "epoch": 0.14174048983845752,
+      "grad_norm": 0.3001891139691023,
+      "learning_rate": 1.9467470355910438e-05,
+      "log_odds_chosen": 1.650146484375,
+      "log_odds_ratio": -0.1990966796875,
+      "logits/chosen": 0.2593860626220703,
+      "logits/rejected": 0.3832855224609375,
+      "logps/chosen": -0.5384521484375,
+      "logps/rejected": -1.548828125,
+      "loss": 9.6108,
+      "nll_loss": 0.5833740234375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0538482666015625,
+      "rewards/margins": 0.1009674072265625,
+      "rewards/rejected": -0.15478515625,
+      "step": 34
+    },
+    {
+      "epoch": 0.14590932777488275,
+      "grad_norm": 0.3087581556467376,
+      "learning_rate": 1.9422609221188208e-05,
+      "log_odds_chosen": 2.18505859375,
+      "log_odds_ratio": -0.1383209228515625,
+      "logits/chosen": 0.0203094482421875,
+      "logits/rejected": 0.2121429443359375,
+      "logps/chosen": -0.51885986328125,
+      "logps/rejected": -1.948486328125,
+      "loss": 10.4194,
+      "nll_loss": 0.5579833984375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0518798828125,
+      "rewards/margins": 0.1428070068359375,
+      "rewards/rejected": -0.19476318359375,
+      "step": 35
+    },
+    {
+      "epoch": 0.15007816571130797,
+      "grad_norm": 0.2787737237763669,
+      "learning_rate": 1.9375990128440205e-05,
+      "log_odds_chosen": 1.763427734375,
+      "log_odds_ratio": -0.20301055908203125,
+      "logits/chosen": 0.231658935546875,
+      "logits/rejected": 0.39324188232421875,
+      "logps/chosen": -0.5516357421875,
+      "logps/rejected": -1.69140625,
+      "loss": 9.2695,
+      "nll_loss": 0.5987548828125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0550994873046875,
+      "rewards/margins": 0.1139068603515625,
+      "rewards/rejected": -0.169036865234375,
+      "step": 36
+    },
+    {
+      "epoch": 0.1542470036477332,
+      "grad_norm": 0.2729863818054811,
+      "learning_rate": 1.932762177530064e-05,
+      "log_odds_chosen": 1.658447265625,
+      "log_odds_ratio": -0.193603515625,
+      "logits/chosen": -0.1316375732421875,
+      "logits/rejected": 0.02693939208984375,
+      "logps/chosen": -0.43603515625,
+      "logps/rejected": -1.3701171875,
+      "loss": 8.771,
+      "nll_loss": 0.4583740234375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0435943603515625,
+      "rewards/margins": 0.093505859375,
+      "rewards/rejected": -0.137054443359375,
+      "step": 37
+    },
+    {
+      "epoch": 0.15841584158415842,
+      "grad_norm": 0.2498280060687234,
+      "learning_rate": 1.9277513185759847e-05,
+      "log_odds_chosen": 1.498291015625,
+      "log_odds_ratio": -0.2294921875,
+      "logits/chosen": 0.28093433380126953,
+      "logits/rejected": 0.437744140625,
+      "logps/chosen": -0.53216552734375,
+      "logps/rejected": -1.40478515625,
+      "loss": 8.7964,
+      "nll_loss": 0.5870361328125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.053192138671875,
+      "rewards/margins": 0.0873870849609375,
+      "rewards/rejected": -0.14056396484375,
+      "step": 38
+    },
+    {
+      "epoch": 0.16258467952058364,
+      "grad_norm": 0.24874511073857605,
+      "learning_rate": 1.922567370848072e-05,
+      "log_odds_chosen": 1.70361328125,
+      "log_odds_ratio": -0.1970672607421875,
+      "logits/chosen": 0.152435302734375,
+      "logits/rejected": 0.3475341796875,
+      "logps/chosen": -0.517578125,
+      "logps/rejected": -1.5634765625,
+      "loss": 9.2043,
+      "nll_loss": 0.55792236328125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.05176544189453125,
+      "rewards/margins": 0.10449981689453125,
+      "rewards/rejected": -0.156280517578125,
+      "step": 39
+    },
+    {
+      "epoch": 0.16675351745700887,
+      "grad_norm": 0.25505276051189985,
+      "learning_rate": 1.917211301505453e-05,
+      "log_odds_chosen": 1.87109375,
+      "log_odds_ratio": -0.1639862060546875,
+      "logits/chosen": -0.01544189453125,
+      "logits/rejected": 0.23333740234375,
+      "logps/chosen": -0.6492919921875,
+      "logps/rejected": -1.87548828125,
+      "loss": 8.9302,
+      "nll_loss": 0.6806640625,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.06491851806640625,
+      "rewards/margins": 0.122833251953125,
+      "rewards/rejected": -0.18768310546875,
+      "step": 40
+    },
+    {
+      "epoch": 0.1709223553934341,
+      "grad_norm": 0.25038865681527556,
+      "learning_rate": 1.9116841098196538e-05,
+      "log_odds_chosen": 1.571533203125,
+      "log_odds_ratio": -0.2160186767578125,
+      "logits/chosen": 0.20674896240234375,
+      "logits/rejected": 0.411865234375,
+      "logps/chosen": -0.6446533203125,
+      "logps/rejected": -1.655517578125,
+      "loss": 10.1494,
+      "nll_loss": 0.683837890625,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.064483642578125,
+      "rewards/margins": 0.1011505126953125,
+      "rewards/rejected": -0.1656494140625,
+      "step": 41
+    },
+    {
+      "epoch": 0.1750911933298593,
+      "grad_norm": 0.25353387868938165,
+      "learning_rate": 1.9059868269881637e-05,
+      "log_odds_chosen": 2.16650390625,
+      "log_odds_ratio": -0.152008056640625,
+      "logits/chosen": 0.08073043823242188,
+      "logits/rejected": 0.36627197265625,
+      "logps/chosen": -0.5087890625,
+      "logps/rejected": -1.926513671875,
+      "loss": 8.4038,
+      "nll_loss": 0.5565185546875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.050872802734375,
+      "rewards/margins": 0.141632080078125,
+      "rewards/rejected": -0.192596435546875,
+      "step": 42
+    },
+    {
+      "epoch": 0.17926003126628454,
+      "grad_norm": 0.2688107571543341,
+      "learning_rate": 1.9001205159420512e-05,
+      "log_odds_chosen": 1.97216796875,
+      "log_odds_ratio": -0.1601715087890625,
+      "logits/chosen": 0.13031005859375,
+      "logits/rejected": 0.38970947265625,
+      "logps/chosen": -0.494384765625,
+      "logps/rejected": -1.752685546875,
+      "loss": 9.4678,
+      "nll_loss": 0.529296875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0494537353515625,
+      "rewards/margins": 0.12591552734375,
+      "rewards/rejected": -0.175323486328125,
+      "step": 43
+    },
+    {
+      "epoch": 0.18342886920270973,
+      "grad_norm": 0.23905117537105064,
+      "learning_rate": 1.8940862711476515e-05,
+      "log_odds_chosen": 1.85693359375,
+      "log_odds_ratio": -0.168853759765625,
+      "logits/chosen": 0.04300975799560547,
+      "logits/rejected": 0.3031005859375,
+      "logps/chosen": -0.48065185546875,
+      "logps/rejected": -1.60107421875,
+      "loss": 8.9258,
+      "nll_loss": 0.512451171875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04808807373046875,
+      "rewards/margins": 0.1120452880859375,
+      "rewards/rejected": -0.16015625,
+      "step": 44
+    },
+    {
+      "epoch": 0.18759770713913496,
+      "grad_norm": 0.24465566014470783,
+      "learning_rate": 1.8878852184023754e-05,
+      "log_odds_chosen": 1.964111328125,
+      "log_odds_ratio": -0.16321563720703125,
+      "logits/chosen": 0.3105430603027344,
+      "logits/rejected": 0.6311187744140625,
+      "logps/chosen": -0.600830078125,
+      "logps/rejected": -1.91357421875,
+      "loss": 9.2246,
+      "nll_loss": 0.6397705078125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0601043701171875,
+      "rewards/margins": 0.1313323974609375,
+      "rewards/rejected": -0.19140625,
+      "step": 45
+    },
+    {
+      "epoch": 0.19176654507556018,
+      "grad_norm": 0.24634616718189936,
+      "learning_rate": 1.8815185146246718e-05,
+      "log_odds_chosen": 1.959228515625,
+      "log_odds_ratio": -0.1479034423828125,
+      "logits/chosen": 0.133026123046875,
+      "logits/rejected": 0.42559814453125,
+      "logps/chosen": -0.4908447265625,
+      "logps/rejected": -1.70263671875,
+      "loss": 8.5979,
+      "nll_loss": 0.5269775390625,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04906463623046875,
+      "rewards/margins": 0.1209716796875,
+      "rewards/rejected": -0.170166015625,
+      "step": 46
+    },
+    {
+      "epoch": 0.1959353830119854,
+      "grad_norm": 0.24351532815561816,
+      "learning_rate": 1.8749873476381827e-05,
+      "log_odds_chosen": 1.909423828125,
+      "log_odds_ratio": -0.160247802734375,
+      "logits/chosen": 0.156036376953125,
+      "logits/rejected": 0.4642486572265625,
+      "logps/chosen": -0.5460205078125,
+      "logps/rejected": -1.7294921875,
+      "loss": 9.5645,
+      "nll_loss": 0.58837890625,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.05460357666015625,
+      "rewards/margins": 0.1183624267578125,
+      "rewards/rejected": -0.173065185546875,
+      "step": 47
+    },
+    {
+      "epoch": 0.20010422094841063,
+      "grad_norm": 0.2494867438799568,
+      "learning_rate": 1.8682929359501338e-05,
+      "log_odds_chosen": 2.22119140625,
+      "log_odds_ratio": -0.14777088165283203,
+      "logits/chosen": 0.09519004821777344,
+      "logits/rejected": 0.4602165222167969,
+      "logps/chosen": -0.4898681640625,
+      "logps/rejected": -1.955322265625,
+      "loss": 8.2217,
+      "nll_loss": 0.5205078125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.048980712890625,
+      "rewards/margins": 0.1465911865234375,
+      "rewards/rejected": -0.19561767578125,
+      "step": 48
+    },
+    {
+      "epoch": 0.20427305888483585,
+      "grad_norm": 0.2613632942085652,
+      "learning_rate": 1.8614365285240002e-05,
+      "log_odds_chosen": 2.54736328125,
+      "log_odds_ratio": -0.12019157409667969,
+      "logits/chosen": 0.262115478515625,
+      "logits/rejected": 0.7015838623046875,
+      "logps/chosen": -0.5364990234375,
+      "logps/rejected": -2.32861328125,
+      "loss": 10.1479,
+      "nll_loss": 0.58154296875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.053680419921875,
+      "rewards/margins": 0.1791839599609375,
+      "rewards/rejected": -0.232757568359375,
+      "step": 49
+    },
+    {
+      "epoch": 0.20844189682126107,
+      "grad_norm": 0.2512225056309303,
+      "learning_rate": 1.8544194045464888e-05,
+      "log_odds_chosen": 2.5048828125,
+      "log_odds_ratio": -0.11281204223632812,
+      "logits/chosen": 0.14321136474609375,
+      "logits/rejected": 0.6065673828125,
+      "logps/chosen": -0.578369140625,
+      "logps/rejected": -2.333984375,
+      "loss": 8.9175,
+      "nll_loss": 0.6080322265625,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.057830810546875,
+      "rewards/margins": 0.175628662109375,
+      "rewards/rejected": -0.23345947265625,
+      "step": 50
+    },
+    {
+      "epoch": 0.2126107347576863,
+      "grad_norm": 0.24838000958008843,
+      "learning_rate": 1.8472428731888836e-05,
+      "log_odds_chosen": 2.25341796875,
+      "log_odds_ratio": -0.14614105224609375,
+      "logits/chosen": 0.20198440551757812,
+      "logits/rejected": 0.6221237182617188,
+      "logps/chosen": -0.55615234375,
+      "logps/rejected": -2.099609375,
+      "loss": 9.3623,
+      "nll_loss": 0.6011962890625,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.055572509765625,
+      "rewards/margins": 0.154205322265625,
+      "rewards/rejected": -0.209808349609375,
+      "step": 51
+    },
+    {
+      "epoch": 0.21677957269411152,
+      "grad_norm": 0.2819144258147841,
+      "learning_rate": 1.8399082733627967e-05,
+      "log_odds_chosen": 2.265869140625,
+      "log_odds_ratio": -0.13389968872070312,
+      "logits/chosen": 0.2725982666015625,
+      "logits/rejected": 0.713623046875,
+      "logps/chosen": -0.59423828125,
+      "logps/rejected": -2.17724609375,
+      "loss": 9.584,
+      "nll_loss": 0.6329345703125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0594482421875,
+      "rewards/margins": 0.158233642578125,
+      "rewards/rejected": -0.217864990234375,
+      "step": 52
+    },
+    {
+      "epoch": 0.22094841063053675,
+      "grad_norm": 0.23567493182733973,
+      "learning_rate": 1.832416973470368e-05,
+      "log_odds_chosen": 2.381591796875,
+      "log_odds_ratio": -0.1349315643310547,
+      "logits/chosen": 0.10595703125,
+      "logits/rejected": 0.584869384765625,
+      "logps/chosen": -0.5697021484375,
+      "logps/rejected": -2.246826171875,
+      "loss": 9.1006,
+      "nll_loss": 0.5938720703125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.05701446533203125,
+      "rewards/margins": 0.167572021484375,
+      "rewards/rejected": -0.224456787109375,
+      "step": 53
+    },
+    {
+      "epoch": 0.22511724856696197,
+      "grad_norm": 0.22661409894490514,
+      "learning_rate": 1.8247703711489684e-05,
+      "log_odds_chosen": 2.53515625,
+      "log_odds_ratio": -0.11096572875976562,
+      "logits/chosen": 0.1064910888671875,
+      "logits/rejected": 0.5858306884765625,
+      "logps/chosen": -0.457763671875,
+      "logps/rejected": -2.11865234375,
+      "loss": 8.2666,
+      "nll_loss": 0.489501953125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04578399658203125,
+      "rewards/margins": 0.1662139892578125,
+      "rewards/rejected": -0.212005615234375,
+      "step": 54
+    },
+    {
+      "epoch": 0.2292860865033872,
+      "grad_norm": 0.2122635369524212,
+      "learning_rate": 1.816969893010442e-05,
+      "log_odds_chosen": 2.40380859375,
+      "log_odds_ratio": -0.1234893798828125,
+      "logits/chosen": 0.23822784423828125,
+      "logits/rejected": 0.7010650634765625,
+      "logps/chosen": -0.4842529296875,
+      "logps/rejected": -2.067626953125,
+      "loss": 8.4883,
+      "nll_loss": 0.5263671875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04840087890625,
+      "rewards/margins": 0.158416748046875,
+      "rewards/rejected": -0.206787109375,
+      "step": 55
+    },
+    {
+      "epoch": 0.2334549244398124,
+      "grad_norm": 0.24092973363237136,
+      "learning_rate": 1.8090169943749477e-05,
+      "log_odds_chosen": 2.4140625,
+      "log_odds_ratio": -0.11363983154296875,
+      "logits/chosen": 0.22986769676208496,
+      "logits/rejected": 0.75823974609375,
+      "logps/chosen": -0.5335693359375,
+      "logps/rejected": -2.19140625,
+      "loss": 9.4546,
+      "nll_loss": 0.58251953125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0533447265625,
+      "rewards/margins": 0.1658935546875,
+      "rewards/rejected": -0.21905517578125,
+      "step": 56
+    },
+    {
+      "epoch": 0.2376237623762376,
+      "grad_norm": 0.24128229489592115,
+      "learning_rate": 1.8009131589994418e-05,
+      "log_odds_chosen": 2.52001953125,
+      "log_odds_ratio": -0.10130882263183594,
+      "logits/chosen": 0.3608417510986328,
+      "logits/rejected": 0.9401397705078125,
+      "logps/chosen": -0.5968017578125,
+      "logps/rejected": -2.3515625,
+      "loss": 9.0918,
+      "nll_loss": 0.6282958984375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.05966949462890625,
+      "rewards/margins": 0.175506591796875,
+      "rewards/rejected": -0.235137939453125,
+      "step": 57
+    },
+    {
+      "epoch": 0.24179260031266284,
+      "grad_norm": 0.21850524678110875,
+      "learning_rate": 1.7926598988008584e-05,
+      "log_odds_chosen": 2.6856689453125,
+      "log_odds_ratio": -0.14227867126464844,
+      "logits/chosen": 0.4098968505859375,
+      "logits/rejected": 1.03839111328125,
+      "logps/chosen": -0.536865234375,
+      "logps/rejected": -2.50439453125,
+      "loss": 9.1926,
+      "nll_loss": 0.564697265625,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.05364990234375,
+      "rewards/margins": 0.1970672607421875,
+      "rewards/rejected": -0.25042724609375,
+      "step": 58
+    },
+    {
+      "epoch": 0.24596143824908806,
+      "grad_norm": 0.2280813073435613,
+      "learning_rate": 1.7842587535740315e-05,
+      "log_odds_chosen": 2.0107421875,
+      "log_odds_ratio": -0.14935302734375,
+      "logits/chosen": 0.4387807846069336,
+      "logits/rejected": 0.90997314453125,
+      "logps/chosen": -0.5528564453125,
+      "logps/rejected": -1.870361328125,
+      "loss": 8.6011,
+      "nll_loss": 0.5877685546875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.05525970458984375,
+      "rewards/margins": 0.131805419921875,
+      "rewards/rejected": -0.18701171875,
+      "step": 59
+    },
+    {
+      "epoch": 0.2501302761855133,
+      "grad_norm": 0.22058979865253361,
+      "learning_rate": 1.77571129070442e-05,
+      "log_odds_chosen": 2.25244140625,
+      "log_odds_ratio": -0.1284637451171875,
+      "logits/chosen": 0.28824806213378906,
+      "logits/rejected": 0.82415771484375,
+      "logps/chosen": -0.5008544921875,
+      "logps/rejected": -1.96728515625,
+      "loss": 8.9312,
+      "nll_loss": 0.5223388671875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.05010986328125,
+      "rewards/margins": 0.1463623046875,
+      "rewards/rejected": -0.19659423828125,
+      "step": 60
+    },
+    {
+      "epoch": 0.2542991141219385,
+      "grad_norm": 0.21889441829668788,
+      "learning_rate": 1.7670191048756827e-05,
+      "log_odds_chosen": 2.78369140625,
+      "log_odds_ratio": -0.07579803466796875,
+      "logits/chosen": 0.07476806640625,
+      "logits/rejected": 0.78863525390625,
+      "logps/chosen": -0.48797607421875,
+      "logps/rejected": -2.36572265625,
+      "loss": 8.9883,
+      "nll_loss": 0.5145263671875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0487823486328125,
+      "rewards/margins": 0.18756103515625,
+      "rewards/rejected": -0.2364501953125,
+      "step": 61
+    },
+    {
+      "epoch": 0.2584679520583637,
+      "grad_norm": 0.24736221036261355,
+      "learning_rate": 1.758183817772163e-05,
+      "log_odds_chosen": 3.00390625,
+      "log_odds_ratio": -0.09209632873535156,
+      "logits/chosen": 0.5449905395507812,
+      "logits/rejected": 1.349365234375,
+      "logps/chosen": -0.6527099609375,
+      "logps/rejected": -2.9609375,
+      "loss": 9.061,
+      "nll_loss": 0.70263671875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0652313232421875,
+      "rewards/margins": 0.230743408203125,
+      "rewards/rejected": -0.29608154296875,
+      "step": 62
+    },
+    {
+      "epoch": 0.26263678999478896,
+      "grad_norm": 0.2374825351022171,
+      "learning_rate": 1.749207077776331e-05,
+      "log_odds_chosen": 3.13818359375,
+      "log_odds_ratio": -0.06859254837036133,
+      "logits/chosen": 0.263458251953125,
+      "logits/rejected": 1.0475082397460938,
+      "logps/chosen": -0.48260498046875,
+      "logps/rejected": -2.7265625,
+      "loss": 8.6455,
+      "nll_loss": 0.51739501953125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04827880859375,
+      "rewards/margins": 0.2242431640625,
+      "rewards/rejected": -0.27252197265625,
+      "step": 63
+    },
+    {
+      "epoch": 0.26680562793121415,
+      "grad_norm": 0.23724721805422275,
+      "learning_rate": 1.740090559661252e-05,
+      "log_odds_chosen": 2.47119140625,
+      "log_odds_ratio": -0.134368896484375,
+      "logits/chosen": 0.35262298583984375,
+      "logits/rejected": 1.022674560546875,
+      "logps/chosen": -0.5977783203125,
+      "logps/rejected": -2.363525390625,
+      "loss": 9.0049,
+      "nll_loss": 0.6239013671875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0598297119140625,
+      "rewards/margins": 0.1763916015625,
+      "rewards/rejected": -0.23626708984375,
+      "step": 64
+    },
+    {
+      "epoch": 0.2709744658676394,
+      "grad_norm": 0.21699208752080323,
+      "learning_rate": 1.730835964278124e-05,
+      "log_odds_chosen": 2.75830078125,
+      "log_odds_ratio": -0.10488510131835938,
+      "logits/chosen": 0.1995849609375,
+      "logits/rejected": 0.919189453125,
+      "logps/chosen": -0.4716796875,
+      "logps/rejected": -2.36962890625,
+      "loss": 8.4927,
+      "nll_loss": 0.5042724609375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0471649169921875,
+      "rewards/margins": 0.189697265625,
+      "rewards/rejected": -0.2369384765625,
+      "step": 65
+    },
+    {
+      "epoch": 0.2751433038040646,
+      "grad_norm": 0.22482452188091642,
+      "learning_rate": 1.721445018238956e-05,
+      "log_odds_chosen": 2.879638671875,
+      "log_odds_ratio": -0.09115219116210938,
+      "logits/chosen": 0.1223907470703125,
+      "logits/rejected": 0.9041290283203125,
+      "logps/chosen": -0.4810791015625,
+      "logps/rejected": -2.525634765625,
+      "loss": 8.51,
+      "nll_loss": 0.5155029296875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0481109619140625,
+      "rewards/margins": 0.2045440673828125,
+      "rewards/rejected": -0.252655029296875,
+      "step": 66
+    },
+    {
+      "epoch": 0.27931214174048985,
+      "grad_norm": 0.21139538168727146,
+      "learning_rate": 1.7119194735944336e-05,
+      "log_odds_chosen": 2.35986328125,
+      "log_odds_ratio": -0.12477874755859375,
+      "logits/chosen": 0.51458740234375,
+      "logits/rejected": 1.1568603515625,
+      "logps/chosen": -0.528564453125,
+      "logps/rejected": -2.12548828125,
+      "loss": 8.5718,
+      "nll_loss": 0.5594482421875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0528717041015625,
+      "rewards/margins": 0.1595611572265625,
+      "rewards/rejected": -0.212554931640625,
+      "step": 67
+    },
+    {
+      "epoch": 0.28348097967691505,
+      "grad_norm": 0.2293374844558361,
+      "learning_rate": 1.7022611075070476e-05,
+      "log_odds_chosen": 2.974609375,
+      "log_odds_ratio": -0.0671539306640625,
+      "logits/chosen": 0.3210601806640625,
+      "logits/rejected": 1.1292877197265625,
+      "logps/chosen": -0.477783203125,
+      "logps/rejected": -2.55126953125,
+      "loss": 8.9819,
+      "nll_loss": 0.50531005859375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04778289794921875,
+      "rewards/margins": 0.207275390625,
+      "rewards/rejected": -0.25506591796875,
+      "step": 68
+    },
+    {
+      "epoch": 0.2876498176133403,
+      "grad_norm": 0.2398973971579909,
+      "learning_rate": 1.6924717219195258e-05,
+      "log_odds_chosen": 2.865234375,
+      "log_odds_ratio": -0.09108352661132812,
+      "logits/chosen": 0.3566474914550781,
+      "logits/rejected": 1.15185546875,
+      "logps/chosen": -0.5130615234375,
+      "logps/rejected": -2.56396484375,
+      "loss": 8.7573,
+      "nll_loss": 0.5433349609375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.051300048828125,
+      "rewards/margins": 0.20513916015625,
+      "rewards/rejected": -0.2564697265625,
+      "step": 69
+    },
+    {
+      "epoch": 0.2918186555497655,
+      "grad_norm": 0.2380817503932237,
+      "learning_rate": 1.6825531432186545e-05,
+      "log_odds_chosen": 3.68310546875,
+      "log_odds_ratio": -0.06094849109649658,
+      "logits/chosen": 0.10516357421875,
+      "logits/rejected": 1.127197265625,
+      "logps/chosen": -0.48614501953125,
+      "logps/rejected": -3.21044921875,
+      "loss": 8.9338,
+      "nll_loss": 0.51641845703125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04863739013671875,
+      "rewards/margins": 0.2725067138671875,
+      "rewards/rejected": -0.321075439453125,
+      "step": 70
+    },
+    {
+      "epoch": 0.29598749348619074,
+      "grad_norm": 0.2159429700257839,
+      "learning_rate": 1.6725072218945274e-05,
+      "log_odds_chosen": 3.00537109375,
+      "log_odds_ratio": -0.062255859375,
+      "logits/chosen": 0.08046150207519531,
+      "logits/rejected": 0.93292236328125,
+      "logps/chosen": -0.4757080078125,
+      "logps/rejected": -2.52734375,
+      "loss": 8.4172,
+      "nll_loss": 0.5010986328125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04759979248046875,
+      "rewards/margins": 0.2049560546875,
+      "rewards/rejected": -0.25262451171875,
+      "step": 71
+    },
+    {
+      "epoch": 0.30015633142261594,
+      "grad_norm": 0.21370278036618626,
+      "learning_rate": 1.662335832195308e-05,
+      "log_odds_chosen": 3.220703125,
+      "log_odds_ratio": -0.07026481628417969,
+      "logits/chosen": 0.4130706787109375,
+      "logits/rejected": 1.337158203125,
+      "logps/chosen": -0.494140625,
+      "logps/rejected": -2.8369140625,
+      "loss": 8.4102,
+      "nll_loss": 0.5313720703125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04938507080078125,
+      "rewards/margins": 0.234222412109375,
+      "rewards/rejected": -0.28363037109375,
+      "step": 72
+    },
+    {
+      "epoch": 0.3043251693590412,
+      "grad_norm": 0.21445462010402655,
+      "learning_rate": 1.6520408717775507e-05,
+      "log_odds_chosen": 3.030029296875,
+      "log_odds_ratio": -0.09325134754180908,
+      "logits/chosen": 0.291961669921875,
+      "logits/rejected": 1.124176025390625,
+      "logps/chosen": -0.46539306640625,
+      "logps/rejected": -2.61962890625,
+      "loss": 8.7749,
+      "nll_loss": 0.52032470703125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04654693603515625,
+      "rewards/margins": 0.215423583984375,
+      "rewards/rejected": -0.261932373046875,
+      "step": 73
+    },
+    {
+      "epoch": 0.3084940072954664,
+      "grad_norm": 0.2305921085093174,
+      "learning_rate": 1.6416242613521612e-05,
+      "log_odds_chosen": 3.40966796875,
+      "log_odds_ratio": -0.051792144775390625,
+      "logits/chosen": 0.22701263427734375,
+      "logits/rejected": 1.232421875,
+      "logps/chosen": -0.4923095703125,
+      "logps/rejected": -3.0087890625,
+      "loss": 8.9971,
+      "nll_loss": 0.532958984375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.04920196533203125,
+      "rewards/margins": 0.2520751953125,
+      "rewards/rejected": -0.3011474609375,
+      "step": 74
+    },
+    {
+      "epoch": 0.3126628452318916,
+      "grad_norm": 0.21171111932668515,
+      "learning_rate": 1.631087944326053e-05,
+      "log_odds_chosen": 2.77880859375,
+      "log_odds_ratio": -0.11410236358642578,
+      "logits/chosen": 0.3912620544433594,
+      "logits/rejected": 1.1761474609375,
+      "logps/chosen": -0.5250244140625,
+      "logps/rejected": -2.53466796875,
+      "loss": 9.2783,
+      "nll_loss": 0.5582275390625,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.052490234375,
+      "rewards/margins": 0.200958251953125,
+      "rewards/rejected": -0.25347900390625,
+      "step": 75
+    },
+    {
+      "epoch": 0.31683168316831684,
+      "grad_norm": 0.21327287216080953,
+      "learning_rate": 1.6204338864395683e-05,
+      "log_odds_chosen": 3.2021484375,
+      "log_odds_ratio": -0.09289395809173584,
+      "logits/chosen": 0.41394615173339844,
+      "logits/rejected": 1.296875,
+      "logps/chosen": -0.5283203125,
+      "logps/rejected": -2.9228515625,
+      "loss": 8.793,
+      "nll_loss": 0.5673828125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0528564453125,
+      "rewards/margins": 0.2394561767578125,
+      "rewards/rejected": -0.29241943359375,
+      "step": 76
+    },
+    {
+      "epoch": 0.32100052110474203,
+      "grad_norm": 0.22081572582819636,
+      "learning_rate": 1.609664075399735e-05,
+      "log_odds_chosen": 3.8564453125,
+      "log_odds_ratio": -0.06345272064208984,
+      "logits/chosen": 0.2453155517578125,
+      "logits/rejected": 1.3897705078125,
+      "logps/chosen": -0.47216796875,
+      "logps/rejected": -3.40869140625,
+      "loss": 8.6531,
+      "nll_loss": 0.510498046875,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0472259521484375,
+      "rewards/margins": 0.293609619140625,
+      "rewards/rejected": -0.34075927734375,
+      "step": 77
+    },
+    {
+      "epoch": 0.3251693590411673,
+      "grad_norm": 0.21379165530281527,
+      "learning_rate": 1.5987805205094225e-05,
+      "log_odds_chosen": 3.5078125,
+      "log_odds_ratio": -0.060007452964782715,
+      "logits/chosen": 0.2599372863769531,
+      "logits/rejected": 1.30712890625,
+      "logps/chosen": -0.5233154296875,
+      "logps/rejected": -3.1806640625,
+      "loss": 8.3716,
+      "nll_loss": 0.5660400390625,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0523529052734375,
+      "rewards/margins": 0.2657470703125,
+      "rewards/rejected": -0.318115234375,
+      "step": 78
+    },
+    {
+      "epoch": 0.3293381969775925,
+      "grad_norm": 0.2285233153975496,
+      "learning_rate": 1.5877852522924733e-05,
+      "log_odds_chosen": 3.60546875,
+      "log_odds_ratio": -0.041787147521972656,
+      "logits/chosen": 0.21383285522460938,
+      "logits/rejected": 1.2878341674804688,
+      "logps/chosen": -0.51556396484375,
+      "logps/rejected": -3.1748046875,
+      "loss": 8.3945,
+      "nll_loss": 0.5489501953125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0515289306640625,
+      "rewards/margins": 0.266082763671875,
+      "rewards/rejected": -0.3175048828125,
+      "step": 79
+    },
+    {
+      "epoch": 0.33350703491401773,
+      "grad_norm": 0.217365661040775,
+      "learning_rate": 1.5766803221148676e-05,
+      "log_odds_chosen": 3.79150390625,
+      "log_odds_ratio": -0.03936195373535156,
+      "logits/chosen": -0.07393646240234375,
+      "logits/rejected": 1.0196533203125,
+      "logps/chosen": -0.41058349609375,
+      "logps/rejected": -3.138671875,
+      "loss": 8.1189,
+      "nll_loss": 0.43505859375,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.0410614013671875,
+      "rewards/margins": 0.272613525390625,
+      "rewards/rejected": -0.31378173828125,
+      "step": 80
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 240,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 80,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 0.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

ckpt-80/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d663160fa42d2a5dc56e444c10033529d0fa4a28b7c6e5e99d32a5399f9c1057
+size 8440

ckpt-80/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)