Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,7 @@ 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
+checkpoint-107/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+checkpoint-214/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+checkpoint-321/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,175 @@
+---
+library_name: transformers
+license: apache-2.0
+base_model: Qwen/Qwen2.5-Math-1.5B-Instruct
+tags:
+- generated_from_trainer
+datasets:
+- train.jsonl
+model-index:
+- name: outputs/out
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
+<details><summary>See axolotl config</summary>
+
+axolotl version: `0.6.0`
+```yaml
+base_model: Qwen/Qwen2.5-Math-1.5B-Instruct
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
+trust_remote_code: false
+
+load_in_8bit: false
+load_in_4bit: false
+strict: false
+
+output_dir: ./outputs/out
+remove_unused_columns: false
+
+chat_template: qwen_25
+# chat_template: qwen_25
+datasets:
+  - path: train.jsonl
+    type: chat_template
+    field_messages: messages
+    message_field_role: role
+    message_field_content: content
+    roles:
+      system:
+        -system
+      user:
+        - user
+      assistant:
+        - assistant
+
+dataset_prepared_path: mr1-sft-1
+# dataset_prepared_path: ko_r1
+val_set_size: 0.005
+eval_sample_packing: False
+
+overrides_of_model_config:
+  # RoPE Scaling https://github.com/huggingface/transformers/pull/24653
+  rope_scaling:
+    type: linear
+    factor: 8.0
+    
+sequence_len: 32768
+sample_packing: False
+pad_to_sequence_len: False
+
+wandb_project: MR1
+wandb_entity:
+wandb_watch:
+wandb_name:
+wandb_log_model:
+
+plugins:
+  - axolotl.integrations.liger.LigerPlugin
+liger_rope: true
+liger_rms_norm: true
+liger_swiglu: true
+liger_fused_linear_cross_entropy: true
+
+gradient_accumulation_steps: 32
+micro_batch_size: 2
+eval_batch_size: 1
+num_epochs: 3
+optimizer: paged_adamw_8bit
+lr_scheduler: cosine
+learning_rate: 2e-5
+
+train_on_inputs: false
+group_by_length: false
+bf16: auto
+fp16: 
+tf32: false
+
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+early_stopping_patience:
+resume_from_checkpoint:
+logging_steps: 1
+xformers_attention:
+flash_attention: true
+
+warmup_ratio: 0.05
+evals_per_epoch: 3
+eval_max_new_tokens: 128
+eval_table_size:
+saves_per_epoch: 1
+debug:
+deepspeed: deepspeed_configs/zero3_bf16.json
+weight_decay: 0.01
+fsdp:
+fsdp_config:
+special_tokens:
+  eos_token: 
+
+```
+
+</details><br>
+
+# outputs/out
+
+This model is a fine-tuned version of [Qwen/Qwen2.5-Math-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Math-1.5B-Instruct) on the train.jsonl dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.6320
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 2e-05
+- train_batch_size: 2
+- eval_batch_size: 1
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 8
+- gradient_accumulation_steps: 32
+- total_train_batch_size: 512
+- total_eval_batch_size: 8
+- optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_steps: 16
+- num_epochs: 3.0
+
+### Training results
+
+| Training Loss | Epoch  | Step | Validation Loss |
+|:-------------:|:------:|:----:|:---------------:|
+| 4.5461        | 0.0093 | 1    | 4.5535          |
+| 1.4397        | 0.3362 | 36   | 1.3349          |
+| 0.8795        | 0.6723 | 72   | 0.8389          |
+| 0.7726        | 1.0    | 108  | 0.7298          |
+| 0.7374        | 1.3362 | 144  | 0.6811          |
+| 0.6928        | 1.6723 | 180  | 0.6554          |
+| 0.6742        | 2.0    | 216  | 0.6418          |
+| 0.691         | 2.3362 | 252  | 0.6349          |
+| 0.6656        | 2.6723 | 288  | 0.6320          |
+
+
+### Framework versions
+
+- Transformers 4.48.3
+- Pytorch 2.5.1+cu121
+- Datasets 3.2.0
+- Tokenizers 0.21.0

added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

checkpoint-107/added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

checkpoint-107/config.json

@@ -0,0 +1,31 @@
+{
+  "_name_or_path": "Qwen/Qwen2.5-Math-1.5B-Instruct",
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 8960,
+  "max_position_embeddings": 32768,
+  "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": {
+    "factor": 8.0,
+    "type": "linear"
+  },
+  "rope_theta": 10000.0,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.48.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151665
+}

checkpoint-107/generation_config.json

@@ -0,0 +1,10 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "transformers_version": "4.48.3"
+}

checkpoint-107/latest

@@ -0,0 +1 @@
+global_step107

checkpoint-107/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c1a094ea230a47098ca36b3fb49d7978a92b4db7164e1a3d7dfb476de21bc1a
+size 3086634632

checkpoint-107/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36d2a2034ebb05cb71c510897f2795b31164e50f17b270bc25d2be3ad9a17b22
+size 15984

checkpoint-107/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:060dfdb1c49102cbdc8868a6031e68787601b4ccd782f3fb9b137e20c1fd2c7a
+size 15984

checkpoint-107/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af01895cb66e616591f2e4baa8dcd8151530eab133c73571ccb31c74f35422ce
+size 15984

checkpoint-107/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:677921992b1e0cef3aee776f245975003d22f51d9bd6ed20f248ded1deb72fa9
+size 15984

checkpoint-107/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d69353c629541c690c5471f8ec05fdab2bfecf3d37afaa436bc45939da6db68f
+size 15984

checkpoint-107/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e40ba6668cc03c9162c68a933d164bf38ae2d196a9a6fec03ae615491201185
+size 15984

checkpoint-107/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:870968fea834e24b2e099cf3e4fe1e3fb8caf38d8f8e5b790d7d47386d4d05f5
+size 15984

checkpoint-107/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9e19618bee7c6ef43256fea25abe19bca88535eb1e7dc213cde8929ae4e8180
+size 15984

checkpoint-107/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42ee68d968ce6cc761e6d17993c44cf45babb9f2ad94a6958480df12ec57fe36
+size 1064

checkpoint-107/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-107/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c5ae00e602b8860cbd784ba82a8aa14e8feecec692e7076590d014d7b7fdafa
+size 11421896

checkpoint-107/tokenizer_config.json

@@ -0,0 +1,208 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{%- 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": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

checkpoint-107/trainer_state.json

@@ -0,0 +1,806 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9991245987744383,
+  "eval_steps": 36,
+  "global_step": 107,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.009337613072658301,
+      "grad_norm": 0.6103967193596562,
+      "learning_rate": 1.25e-06,
+      "loss": 4.5461,
+      "step": 1
+    },
+    {
+      "epoch": 0.009337613072658301,
+      "eval_loss": 4.553500175476074,
+      "eval_runtime": 16.4134,
+      "eval_samples_per_second": 16.816,
+      "eval_steps_per_second": 2.132,
+      "step": 1
+    },
+    {
+      "epoch": 0.018675226145316602,
+      "grad_norm": 0.6058263693172029,
+      "learning_rate": 2.5e-06,
+      "loss": 4.5365,
+      "step": 2
+    },
+    {
+      "epoch": 0.028012839217974907,
+      "grad_norm": 0.6061325195841676,
+      "learning_rate": 3.7500000000000005e-06,
+      "loss": 4.5438,
+      "step": 3
+    },
+    {
+      "epoch": 0.037350452290633204,
+      "grad_norm": 0.5966679918377138,
+      "learning_rate": 5e-06,
+      "loss": 4.5495,
+      "step": 4
+    },
+    {
+      "epoch": 0.04668806536329151,
+      "grad_norm": 0.5637877145545853,
+      "learning_rate": 6.25e-06,
+      "loss": 4.4744,
+      "step": 5
+    },
+    {
+      "epoch": 0.05602567843594981,
+      "grad_norm": 0.4955765028851229,
+      "learning_rate": 7.500000000000001e-06,
+      "loss": 4.3581,
+      "step": 6
+    },
+    {
+      "epoch": 0.06536329150860812,
+      "grad_norm": 0.33161457169025776,
+      "learning_rate": 8.750000000000001e-06,
+      "loss": 3.9768,
+      "step": 7
+    },
+    {
+      "epoch": 0.07470090458126641,
+      "grad_norm": 0.30498618054818755,
+      "learning_rate": 1e-05,
+      "loss": 3.9306,
+      "step": 8
+    },
+    {
+      "epoch": 0.08403851765392471,
+      "grad_norm": 0.3418457910893006,
+      "learning_rate": 1.125e-05,
+      "loss": 3.6729,
+      "step": 9
+    },
+    {
+      "epoch": 0.09337613072658302,
+      "grad_norm": 0.3051650090830193,
+      "learning_rate": 1.25e-05,
+      "loss": 3.5839,
+      "step": 10
+    },
+    {
+      "epoch": 0.10271374379924132,
+      "grad_norm": 0.2540841556589586,
+      "learning_rate": 1.375e-05,
+      "loss": 3.4916,
+      "step": 11
+    },
+    {
+      "epoch": 0.11205135687189963,
+      "grad_norm": 0.2523065620988122,
+      "learning_rate": 1.5000000000000002e-05,
+      "loss": 3.3087,
+      "step": 12
+    },
+    {
+      "epoch": 0.12138896994455792,
+      "grad_norm": 0.20081817150865616,
+      "learning_rate": 1.6250000000000002e-05,
+      "loss": 3.1931,
+      "step": 13
+    },
+    {
+      "epoch": 0.13072658301721624,
+      "grad_norm": 0.1823149251708444,
+      "learning_rate": 1.7500000000000002e-05,
+      "loss": 3.1415,
+      "step": 14
+    },
+    {
+      "epoch": 0.14006419608987453,
+      "grad_norm": 0.15949645404615334,
+      "learning_rate": 1.8750000000000002e-05,
+      "loss": 3.0393,
+      "step": 15
+    },
+    {
+      "epoch": 0.14940180916253282,
+      "grad_norm": 0.1325611844716812,
+      "learning_rate": 2e-05,
+      "loss": 2.8927,
+      "step": 16
+    },
+    {
+      "epoch": 0.15873942223519114,
+      "grad_norm": 0.11464751241192053,
+      "learning_rate": 1.9999469523400122e-05,
+      "loss": 2.7378,
+      "step": 17
+    },
+    {
+      "epoch": 0.16807703530784943,
+      "grad_norm": 0.1070431164754716,
+      "learning_rate": 1.9997878149881576e-05,
+      "loss": 2.6558,
+      "step": 18
+    },
+    {
+      "epoch": 0.17741464838050774,
+      "grad_norm": 0.10433083170549438,
+      "learning_rate": 1.999522604828164e-05,
+      "loss": 2.534,
+      "step": 19
+    },
+    {
+      "epoch": 0.18675226145316604,
+      "grad_norm": 0.10117296773389844,
+      "learning_rate": 1.9991513499975883e-05,
+      "loss": 2.4259,
+      "step": 20
+    },
+    {
+      "epoch": 0.19608987452582433,
+      "grad_norm": 0.09323265692759455,
+      "learning_rate": 1.9986740898848306e-05,
+      "loss": 2.3488,
+      "step": 21
+    },
+    {
+      "epoch": 0.20542748759848264,
+      "grad_norm": 0.07868283241281675,
+      "learning_rate": 1.9980908751249556e-05,
+      "loss": 2.2658,
+      "step": 22
+    },
+    {
+      "epoch": 0.21476510067114093,
+      "grad_norm": 0.07548690533550786,
+      "learning_rate": 1.997401767594319e-05,
+      "loss": 2.1874,
+      "step": 23
+    },
+    {
+      "epoch": 0.22410271374379925,
+      "grad_norm": 0.07720457793760661,
+      "learning_rate": 1.996606840404006e-05,
+      "loss": 2.0998,
+      "step": 24
+    },
+    {
+      "epoch": 0.23344032681645754,
+      "grad_norm": 0.07639897047393754,
+      "learning_rate": 1.9957061778920703e-05,
+      "loss": 2.0469,
+      "step": 25
+    },
+    {
+      "epoch": 0.24277793988911583,
+      "grad_norm": 0.07016046755367683,
+      "learning_rate": 1.9946998756145894e-05,
+      "loss": 1.9562,
+      "step": 26
+    },
+    {
+      "epoch": 0.2521155529617741,
+      "grad_norm": 0.06084064670530258,
+      "learning_rate": 1.9935880403355255e-05,
+      "loss": 1.8943,
+      "step": 27
+    },
+    {
+      "epoch": 0.26145316603443247,
+      "grad_norm": 0.05530834717342951,
+      "learning_rate": 1.9923707900153984e-05,
+      "loss": 1.7771,
+      "step": 28
+    },
+    {
+      "epoch": 0.27079077910709076,
+      "grad_norm": 0.05296398833585214,
+      "learning_rate": 1.9910482537987704e-05,
+      "loss": 1.7624,
+      "step": 29
+    },
+    {
+      "epoch": 0.28012839217974905,
+      "grad_norm": 0.05206178996458072,
+      "learning_rate": 1.989620572000544e-05,
+      "loss": 1.6848,
+      "step": 30
+    },
+    {
+      "epoch": 0.28946600525240734,
+      "grad_norm": 0.050431830710485716,
+      "learning_rate": 1.9880878960910772e-05,
+      "loss": 1.6577,
+      "step": 31
+    },
+    {
+      "epoch": 0.29880361832506563,
+      "grad_norm": 0.04760195929852826,
+      "learning_rate": 1.9864503886801108e-05,
+      "loss": 1.6128,
+      "step": 32
+    },
+    {
+      "epoch": 0.308141231397724,
+      "grad_norm": 0.04482797539629353,
+      "learning_rate": 1.9847082234995172e-05,
+      "loss": 1.5791,
+      "step": 33
+    },
+    {
+      "epoch": 0.31747884447038227,
+      "grad_norm": 0.04130893521590142,
+      "learning_rate": 1.982861585384869e-05,
+      "loss": 1.5063,
+      "step": 34
+    },
+    {
+      "epoch": 0.32681645754304056,
+      "grad_norm": 0.04057758728502848,
+      "learning_rate": 1.9809106702558277e-05,
+      "loss": 1.4877,
+      "step": 35
+    },
+    {
+      "epoch": 0.33615407061569885,
+      "grad_norm": 0.038180219331760724,
+      "learning_rate": 1.978855685095358e-05,
+      "loss": 1.4397,
+      "step": 36
+    },
+    {
+      "epoch": 0.33615407061569885,
+      "eval_loss": 1.3349355459213257,
+      "eval_runtime": 16.413,
+      "eval_samples_per_second": 16.816,
+      "eval_steps_per_second": 2.132,
+      "step": 36
+    },
+    {
+      "epoch": 0.34549168368835714,
+      "grad_norm": 0.0360323358530499,
+      "learning_rate": 1.9766968479277684e-05,
+      "loss": 1.3977,
+      "step": 37
+    },
+    {
+      "epoch": 0.3548292967610155,
+      "grad_norm": 0.035146043601200425,
+      "learning_rate": 1.974434387795579e-05,
+      "loss": 1.418,
+      "step": 38
+    },
+    {
+      "epoch": 0.3641669098336738,
+      "grad_norm": 0.033121913235728655,
+      "learning_rate": 1.972068544735221e-05,
+      "loss": 1.3359,
+      "step": 39
+    },
+    {
+      "epoch": 0.37350452290633207,
+      "grad_norm": 0.031036722451915032,
+      "learning_rate": 1.969599569751571e-05,
+      "loss": 1.3401,
+      "step": 40
+    },
+    {
+      "epoch": 0.38284213597899036,
+      "grad_norm": 0.030669047831503912,
+      "learning_rate": 1.9670277247913205e-05,
+      "loss": 1.2638,
+      "step": 41
+    },
+    {
+      "epoch": 0.39217974905164865,
+      "grad_norm": 0.03002785386211285,
+      "learning_rate": 1.964353282715183e-05,
+      "loss": 1.2649,
+      "step": 42
+    },
+    {
+      "epoch": 0.401517362124307,
+      "grad_norm": 0.02817025139456513,
+      "learning_rate": 1.961576527268946e-05,
+      "loss": 1.2476,
+      "step": 43
+    },
+    {
+      "epoch": 0.4108549751969653,
+      "grad_norm": 0.03110381051330927,
+      "learning_rate": 1.9586977530533677e-05,
+      "loss": 1.1923,
+      "step": 44
+    },
+    {
+      "epoch": 0.4201925882696236,
+      "grad_norm": 0.031162812150372473,
+      "learning_rate": 1.95571726549292e-05,
+      "loss": 1.2044,
+      "step": 45
+    },
+    {
+      "epoch": 0.42953020134228187,
+      "grad_norm": 0.029584239693380046,
+      "learning_rate": 1.9526353808033827e-05,
+      "loss": 1.1764,
+      "step": 46
+    },
+    {
+      "epoch": 0.43886781441494016,
+      "grad_norm": 0.027540684267770742,
+      "learning_rate": 1.9494524259582994e-05,
+      "loss": 1.1634,
+      "step": 47
+    },
+    {
+      "epoch": 0.4482054274875985,
+      "grad_norm": 0.028821860863637565,
+      "learning_rate": 1.9461687386542826e-05,
+      "loss": 1.1426,
+      "step": 48
+    },
+    {
+      "epoch": 0.4575430405602568,
+      "grad_norm": 0.027945951918964533,
+      "learning_rate": 1.9427846672751873e-05,
+      "loss": 1.1154,
+      "step": 49
+    },
+    {
+      "epoch": 0.4668806536329151,
+      "grad_norm": 0.026881160663849425,
+      "learning_rate": 1.93930057085515e-05,
+      "loss": 1.0981,
+      "step": 50
+    },
+    {
+      "epoch": 0.4762182667055734,
+      "grad_norm": 0.025002230522002903,
+      "learning_rate": 1.9357168190404937e-05,
+      "loss": 1.0873,
+      "step": 51
+    },
+    {
+      "epoch": 0.48555587977823167,
+      "grad_norm": 0.02318894097448886,
+      "learning_rate": 1.932033792050515e-05,
+      "loss": 1.0701,
+      "step": 52
+    },
+    {
+      "epoch": 0.49489349285089,
+      "grad_norm": 0.021872991266037325,
+      "learning_rate": 1.928251880637141e-05,
+      "loss": 1.0439,
+      "step": 53
+    },
+    {
+      "epoch": 0.5042311059235483,
+      "grad_norm": 0.021369213964442125,
+      "learning_rate": 1.924371486043473e-05,
+      "loss": 1.062,
+      "step": 54
+    },
+    {
+      "epoch": 0.5135687189962066,
+      "grad_norm": 0.02074830989778219,
+      "learning_rate": 1.920393019961217e-05,
+      "loss": 1.0392,
+      "step": 55
+    },
+    {
+      "epoch": 0.5229063320688649,
+      "grad_norm": 0.020345018132100934,
+      "learning_rate": 1.916316904487005e-05,
+      "loss": 1.02,
+      "step": 56
+    },
+    {
+      "epoch": 0.5322439451415232,
+      "grad_norm": 0.01970957719141645,
+      "learning_rate": 1.9121435720776122e-05,
+      "loss": 0.9923,
+      "step": 57
+    },
+    {
+      "epoch": 0.5415815582141815,
+      "grad_norm": 0.019741562276937644,
+      "learning_rate": 1.9078734655040763e-05,
+      "loss": 0.9993,
+      "step": 58
+    },
+    {
+      "epoch": 0.5509191712868398,
+      "grad_norm": 0.0195142739708107,
+      "learning_rate": 1.9035070378047204e-05,
+      "loss": 0.9943,
+      "step": 59
+    },
+    {
+      "epoch": 0.5602567843594981,
+      "grad_norm": 0.017297498916208055,
+      "learning_rate": 1.8990447522370886e-05,
+      "loss": 0.9753,
+      "step": 60
+    },
+    {
+      "epoch": 0.5695943974321565,
+      "grad_norm": 0.01658779952035472,
+      "learning_rate": 1.8944870822287957e-05,
+      "loss": 0.9883,
+      "step": 61
+    },
+    {
+      "epoch": 0.5789320105048147,
+      "grad_norm": 0.018530815854782102,
+      "learning_rate": 1.8898345113273e-05,
+      "loss": 0.9482,
+      "step": 62
+    },
+    {
+      "epoch": 0.588269623577473,
+      "grad_norm": 0.017748842064851857,
+      "learning_rate": 1.8850875331485996e-05,
+      "loss": 0.9533,
+      "step": 63
+    },
+    {
+      "epoch": 0.5976072366501313,
+      "grad_norm": 0.016871546397150952,
+      "learning_rate": 1.8802466513248635e-05,
+      "loss": 0.948,
+      "step": 64
+    },
+    {
+      "epoch": 0.6069448497227896,
+      "grad_norm": 0.01608669807726683,
+      "learning_rate": 1.8753123794509974e-05,
+      "loss": 0.9295,
+      "step": 65
+    },
+    {
+      "epoch": 0.616282462795448,
+      "grad_norm": 0.017453769141315367,
+      "learning_rate": 1.8702852410301556e-05,
+      "loss": 0.9198,
+      "step": 66
+    },
+    {
+      "epoch": 0.6256200758681062,
+      "grad_norm": 0.017655286026566987,
+      "learning_rate": 1.865165769418196e-05,
+      "loss": 0.9535,
+      "step": 67
+    },
+    {
+      "epoch": 0.6349576889407645,
+      "grad_norm": 0.017498254303415152,
+      "learning_rate": 1.8599545077670983e-05,
+      "loss": 0.9278,
+      "step": 68
+    },
+    {
+      "epoch": 0.6442953020134228,
+      "grad_norm": 0.01573094552466445,
+      "learning_rate": 1.854652008967335e-05,
+      "loss": 0.912,
+      "step": 69
+    },
+    {
+      "epoch": 0.6536329150860811,
+      "grad_norm": 0.015035283908585596,
+      "learning_rate": 1.8492588355892125e-05,
+      "loss": 0.9182,
+      "step": 70
+    },
+    {
+      "epoch": 0.6629705281587395,
+      "grad_norm": 0.01593275185300073,
+      "learning_rate": 1.8437755598231857e-05,
+      "loss": 0.9109,
+      "step": 71
+    },
+    {
+      "epoch": 0.6723081412313977,
+      "grad_norm": 0.014842299347889509,
+      "learning_rate": 1.8382027634191523e-05,
+      "loss": 0.8795,
+      "step": 72
+    },
+    {
+      "epoch": 0.6723081412313977,
+      "eval_loss": 0.8388988375663757,
+      "eval_runtime": 16.5273,
+      "eval_samples_per_second": 16.7,
+      "eval_steps_per_second": 2.118,
+      "step": 72
+    },
+    {
+      "epoch": 0.681645754304056,
+      "grad_norm": 0.015086756882680958,
+      "learning_rate": 1.8325410376247295e-05,
+      "loss": 0.8793,
+      "step": 73
+    },
+    {
+      "epoch": 0.6909833673767143,
+      "grad_norm": 0.014400642274755383,
+      "learning_rate": 1.826790983122527e-05,
+      "loss": 0.9139,
+      "step": 74
+    },
+    {
+      "epoch": 0.7003209804493726,
+      "grad_norm": 0.013490998034899438,
+      "learning_rate": 1.8209532099664177e-05,
+      "loss": 0.8767,
+      "step": 75
+    },
+    {
+      "epoch": 0.709658593522031,
+      "grad_norm": 0.011444201595399665,
+      "learning_rate": 1.8150283375168112e-05,
+      "loss": 0.8525,
+      "step": 76
+    },
+    {
+      "epoch": 0.7189962065946892,
+      "grad_norm": 0.01133323463508391,
+      "learning_rate": 1.8090169943749477e-05,
+      "loss": 0.8658,
+      "step": 77
+    },
+    {
+      "epoch": 0.7283338196673476,
+      "grad_norm": 0.011984830212373445,
+      "learning_rate": 1.8029198183162e-05,
+      "loss": 0.856,
+      "step": 78
+    },
+    {
+      "epoch": 0.7376714327400058,
+      "grad_norm": 0.01217011221195066,
+      "learning_rate": 1.796737456222413e-05,
+      "loss": 0.8892,
+      "step": 79
+    },
+    {
+      "epoch": 0.7470090458126641,
+      "grad_norm": 0.012510369655517745,
+      "learning_rate": 1.7904705640132717e-05,
+      "loss": 0.8562,
+      "step": 80
+    },
+    {
+      "epoch": 0.7563466588853225,
+      "grad_norm": 0.011806935544189488,
+      "learning_rate": 1.7841198065767107e-05,
+      "loss": 0.8644,
+      "step": 81
+    },
+    {
+      "epoch": 0.7656842719579807,
+      "grad_norm": 0.011433541340869584,
+      "learning_rate": 1.7776858576983713e-05,
+      "loss": 0.8692,
+      "step": 82
+    },
+    {
+      "epoch": 0.7750218850306391,
+      "grad_norm": 0.012566348663586087,
+      "learning_rate": 1.771169399990119e-05,
+      "loss": 0.8541,
+      "step": 83
+    },
+    {
+      "epoch": 0.7843594981032973,
+      "grad_norm": 0.01072107470914378,
+      "learning_rate": 1.7645711248176198e-05,
+      "loss": 0.8403,
+      "step": 84
+    },
+    {
+      "epoch": 0.7936971111759556,
+      "grad_norm": 0.010617575681982286,
+      "learning_rate": 1.7578917322269885e-05,
+      "loss": 0.81,
+      "step": 85
+    },
+    {
+      "epoch": 0.803034724248614,
+      "grad_norm": 0.010466741230697605,
+      "learning_rate": 1.7511319308705198e-05,
+      "loss": 0.8253,
+      "step": 86
+    },
+    {
+      "epoch": 0.8123723373212722,
+      "grad_norm": 0.010184306281966645,
+      "learning_rate": 1.744292437931502e-05,
+      "loss": 0.8526,
+      "step": 87
+    },
+    {
+      "epoch": 0.8217099503939306,
+      "grad_norm": 0.009850665553519504,
+      "learning_rate": 1.7373739790481263e-05,
+      "loss": 0.8304,
+      "step": 88
+    },
+    {
+      "epoch": 0.8310475634665888,
+      "grad_norm": 0.009236577038055978,
+      "learning_rate": 1.7303772882365018e-05,
+      "loss": 0.7908,
+      "step": 89
+    },
+    {
+      "epoch": 0.8403851765392472,
+      "grad_norm": 0.03711865800548727,
+      "learning_rate": 1.723303107812779e-05,
+      "loss": 0.8085,
+      "step": 90
+    },
+    {
+      "epoch": 0.8497227896119055,
+      "grad_norm": 0.010517236470326855,
+      "learning_rate": 1.7161521883143936e-05,
+      "loss": 0.8313,
+      "step": 91
+    },
+    {
+      "epoch": 0.8590604026845637,
+      "grad_norm": 0.011238580368575812,
+      "learning_rate": 1.7089252884204376e-05,
+      "loss": 0.814,
+      "step": 92
+    },
+    {
+      "epoch": 0.8683980157572221,
+      "grad_norm": 0.009935862606933948,
+      "learning_rate": 1.701623174871168e-05,
+      "loss": 0.8032,
+      "step": 93
+    },
+    {
+      "epoch": 0.8777356288298803,
+      "grad_norm": 0.009655128505771573,
+      "learning_rate": 1.6942466223866582e-05,
+      "loss": 0.8013,
+      "step": 94
+    },
+    {
+      "epoch": 0.8870732419025387,
+      "grad_norm": 0.010463930618313586,
+      "learning_rate": 1.6867964135846043e-05,
+      "loss": 0.8116,
+      "step": 95
+    },
+    {
+      "epoch": 0.896410854975197,
+      "grad_norm": 0.009360797680160574,
+      "learning_rate": 1.679273338897293e-05,
+      "loss": 0.8064,
+      "step": 96
+    },
+    {
+      "epoch": 0.9057484680478552,
+      "grad_norm": 0.00968251573081083,
+      "learning_rate": 1.6716781964877413e-05,
+      "loss": 0.8208,
+      "step": 97
+    },
+    {
+      "epoch": 0.9150860811205136,
+      "grad_norm": 0.009894122849502883,
+      "learning_rate": 1.664011792165012e-05,
+      "loss": 0.8042,
+      "step": 98
+    },
+    {
+      "epoch": 0.9244236941931718,
+      "grad_norm": 0.010626027391586145,
+      "learning_rate": 1.6562749392987255e-05,
+      "loss": 0.8037,
+      "step": 99
+    },
+    {
+      "epoch": 0.9337613072658302,
+      "grad_norm": 0.01083144628829605,
+      "learning_rate": 1.648468458732762e-05,
+      "loss": 0.8079,
+      "step": 100
+    },
+    {
+      "epoch": 0.9430989203384885,
+      "grad_norm": 0.012651590456989211,
+      "learning_rate": 1.6405931786981753e-05,
+      "loss": 0.7903,
+      "step": 101
+    },
+    {
+      "epoch": 0.9524365334111468,
+      "grad_norm": 0.010619166674062264,
+      "learning_rate": 1.6326499347253206e-05,
+      "loss": 0.7783,
+      "step": 102
+    },
+    {
+      "epoch": 0.9617741464838051,
+      "grad_norm": 0.009320303165881526,
+      "learning_rate": 1.6246395695552086e-05,
+      "loss": 0.8035,
+      "step": 103
+    },
+    {
+      "epoch": 0.9711117595564633,
+      "grad_norm": 0.009094019045947338,
+      "learning_rate": 1.6165629330500952e-05,
+      "loss": 0.7839,
+      "step": 104
+    },
+    {
+      "epoch": 0.9804493726291217,
+      "grad_norm": 0.009044004451355656,
+      "learning_rate": 1.6084208821033152e-05,
+      "loss": 0.7641,
+      "step": 105
+    },
+    {
+      "epoch": 0.98978698570178,
+      "grad_norm": 0.009202270165387296,
+      "learning_rate": 1.6002142805483686e-05,
+      "loss": 0.7666,
+      "step": 106
+    },
+    {
+      "epoch": 0.9991245987744383,
+      "grad_norm": 0.009577150303950122,
+      "learning_rate": 1.591943999067273e-05,
+      "loss": 0.7869,
+      "step": 107
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 321,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 107,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.037539103106138e+18,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-107/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c22096e6a9c9dea84f12ad6313d09723c960a987f46ad9af04dc6c1de407c91
+size 10872

checkpoint-107/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)

checkpoint-214/added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

checkpoint-214/config.json

@@ -0,0 +1,31 @@
+{
+  "_name_or_path": "Qwen/Qwen2.5-Math-1.5B-Instruct",
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 8960,
+  "max_position_embeddings": 32768,
+  "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": {
+    "factor": 8.0,
+    "type": "linear"
+  },
+  "rope_theta": 10000.0,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.48.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151665
+}

checkpoint-214/generation_config.json

@@ -0,0 +1,10 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "transformers_version": "4.48.3"
+}

checkpoint-214/latest

@@ -0,0 +1 @@
+global_step213

checkpoint-214/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfdd9e83239574e821a0b2e3a7b4de9e30eb846275fcb8c03ee376106f1210b3
+size 3086634632

checkpoint-214/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ee195ebde9bf012f945f068f133e7fe22fef5450c496607e3ef11cc2034a186
+size 15984

checkpoint-214/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf0fe1a3315d60b197207c5cb249d0ce4f9ce6d7585e696276d9ffbcb5379893
+size 15984

checkpoint-214/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01c5bd6eae04542162b3e94245555bd81312524066bc01d0ebbfc4fd8554240e
+size 15984

checkpoint-214/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45b74942c68b00d657cfce186b0eeb4aa8f52efa04b114803b605fee8de45972
+size 15984

checkpoint-214/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cd66dd2ba958fc9929441817d8154abbd929c0aa9cd66ff3171965bdaaf5d78
+size 15984

checkpoint-214/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89eeedefdd62514d0130acc330a5c08e9774c95d38c60997905cfd65fc54b710
+size 15984

checkpoint-214/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f43ced939100082608f57561a10e1888e69210c80675068db530c5815889910e
+size 15984

checkpoint-214/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d8d6ee244d99525e7004ae3f02d44ae63082d81fbbab7306f641ac6aeeb736f
+size 15984

checkpoint-214/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d2f0db282a9cd753eddd7ae2d5f2dc0a710bfada7c6671a4b3d7125f8773e81
+size 1064

checkpoint-214/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-214/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c5ae00e602b8860cbd784ba82a8aa14e8feecec692e7076590d014d7b7fdafa
+size 11421896

checkpoint-214/tokenizer_config.json

@@ -0,0 +1,208 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{%- 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": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

checkpoint-214/trainer_state.json

@@ -0,0 +1,1579 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.98978698570178,
+  "eval_steps": 36,
+  "global_step": 214,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.009337613072658301,
+      "grad_norm": 0.6103967193596562,
+      "learning_rate": 1.25e-06,
+      "loss": 4.5461,
+      "step": 1
+    },
+    {
+      "epoch": 0.009337613072658301,
+      "eval_loss": 4.553500175476074,
+      "eval_runtime": 16.4134,
+      "eval_samples_per_second": 16.816,
+      "eval_steps_per_second": 2.132,
+      "step": 1
+    },
+    {
+      "epoch": 0.018675226145316602,
+      "grad_norm": 0.6058263693172029,
+      "learning_rate": 2.5e-06,
+      "loss": 4.5365,
+      "step": 2
+    },
+    {
+      "epoch": 0.028012839217974907,
+      "grad_norm": 0.6061325195841676,
+      "learning_rate": 3.7500000000000005e-06,
+      "loss": 4.5438,
+      "step": 3
+    },
+    {
+      "epoch": 0.037350452290633204,
+      "grad_norm": 0.5966679918377138,
+      "learning_rate": 5e-06,
+      "loss": 4.5495,
+      "step": 4
+    },
+    {
+      "epoch": 0.04668806536329151,
+      "grad_norm": 0.5637877145545853,
+      "learning_rate": 6.25e-06,
+      "loss": 4.4744,
+      "step": 5
+    },
+    {
+      "epoch": 0.05602567843594981,
+      "grad_norm": 0.4955765028851229,
+      "learning_rate": 7.500000000000001e-06,
+      "loss": 4.3581,
+      "step": 6
+    },
+    {
+      "epoch": 0.06536329150860812,
+      "grad_norm": 0.33161457169025776,
+      "learning_rate": 8.750000000000001e-06,
+      "loss": 3.9768,
+      "step": 7
+    },
+    {
+      "epoch": 0.07470090458126641,
+      "grad_norm": 0.30498618054818755,
+      "learning_rate": 1e-05,
+      "loss": 3.9306,
+      "step": 8
+    },
+    {
+      "epoch": 0.08403851765392471,
+      "grad_norm": 0.3418457910893006,
+      "learning_rate": 1.125e-05,
+      "loss": 3.6729,
+      "step": 9
+    },
+    {
+      "epoch": 0.09337613072658302,
+      "grad_norm": 0.3051650090830193,
+      "learning_rate": 1.25e-05,
+      "loss": 3.5839,
+      "step": 10
+    },
+    {
+      "epoch": 0.10271374379924132,
+      "grad_norm": 0.2540841556589586,
+      "learning_rate": 1.375e-05,
+      "loss": 3.4916,
+      "step": 11
+    },
+    {
+      "epoch": 0.11205135687189963,
+      "grad_norm": 0.2523065620988122,
+      "learning_rate": 1.5000000000000002e-05,
+      "loss": 3.3087,
+      "step": 12
+    },
+    {
+      "epoch": 0.12138896994455792,
+      "grad_norm": 0.20081817150865616,
+      "learning_rate": 1.6250000000000002e-05,
+      "loss": 3.1931,
+      "step": 13
+    },
+    {
+      "epoch": 0.13072658301721624,
+      "grad_norm": 0.1823149251708444,
+      "learning_rate": 1.7500000000000002e-05,
+      "loss": 3.1415,
+      "step": 14
+    },
+    {
+      "epoch": 0.14006419608987453,
+      "grad_norm": 0.15949645404615334,
+      "learning_rate": 1.8750000000000002e-05,
+      "loss": 3.0393,
+      "step": 15
+    },
+    {
+      "epoch": 0.14940180916253282,
+      "grad_norm": 0.1325611844716812,
+      "learning_rate": 2e-05,
+      "loss": 2.8927,
+      "step": 16
+    },
+    {
+      "epoch": 0.15873942223519114,
+      "grad_norm": 0.11464751241192053,
+      "learning_rate": 1.9999469523400122e-05,
+      "loss": 2.7378,
+      "step": 17
+    },
+    {
+      "epoch": 0.16807703530784943,
+      "grad_norm": 0.1070431164754716,
+      "learning_rate": 1.9997878149881576e-05,
+      "loss": 2.6558,
+      "step": 18
+    },
+    {
+      "epoch": 0.17741464838050774,
+      "grad_norm": 0.10433083170549438,
+      "learning_rate": 1.999522604828164e-05,
+      "loss": 2.534,
+      "step": 19
+    },
+    {
+      "epoch": 0.18675226145316604,
+      "grad_norm": 0.10117296773389844,
+      "learning_rate": 1.9991513499975883e-05,
+      "loss": 2.4259,
+      "step": 20
+    },
+    {
+      "epoch": 0.19608987452582433,
+      "grad_norm": 0.09323265692759455,
+      "learning_rate": 1.9986740898848306e-05,
+      "loss": 2.3488,
+      "step": 21
+    },
+    {
+      "epoch": 0.20542748759848264,
+      "grad_norm": 0.07868283241281675,
+      "learning_rate": 1.9980908751249556e-05,
+      "loss": 2.2658,
+      "step": 22
+    },
+    {
+      "epoch": 0.21476510067114093,
+      "grad_norm": 0.07548690533550786,
+      "learning_rate": 1.997401767594319e-05,
+      "loss": 2.1874,
+      "step": 23
+    },
+    {
+      "epoch": 0.22410271374379925,
+      "grad_norm": 0.07720457793760661,
+      "learning_rate": 1.996606840404006e-05,
+      "loss": 2.0998,
+      "step": 24
+    },
+    {
+      "epoch": 0.23344032681645754,
+      "grad_norm": 0.07639897047393754,
+      "learning_rate": 1.9957061778920703e-05,
+      "loss": 2.0469,
+      "step": 25
+    },
+    {
+      "epoch": 0.24277793988911583,
+      "grad_norm": 0.07016046755367683,
+      "learning_rate": 1.9946998756145894e-05,
+      "loss": 1.9562,
+      "step": 26
+    },
+    {
+      "epoch": 0.2521155529617741,
+      "grad_norm": 0.06084064670530258,
+      "learning_rate": 1.9935880403355255e-05,
+      "loss": 1.8943,
+      "step": 27
+    },
+    {
+      "epoch": 0.26145316603443247,
+      "grad_norm": 0.05530834717342951,
+      "learning_rate": 1.9923707900153984e-05,
+      "loss": 1.7771,
+      "step": 28
+    },
+    {
+      "epoch": 0.27079077910709076,
+      "grad_norm": 0.05296398833585214,
+      "learning_rate": 1.9910482537987704e-05,
+      "loss": 1.7624,
+      "step": 29
+    },
+    {
+      "epoch": 0.28012839217974905,
+      "grad_norm": 0.05206178996458072,
+      "learning_rate": 1.989620572000544e-05,
+      "loss": 1.6848,
+      "step": 30
+    },
+    {
+      "epoch": 0.28946600525240734,
+      "grad_norm": 0.050431830710485716,
+      "learning_rate": 1.9880878960910772e-05,
+      "loss": 1.6577,
+      "step": 31
+    },
+    {
+      "epoch": 0.29880361832506563,
+      "grad_norm": 0.04760195929852826,
+      "learning_rate": 1.9864503886801108e-05,
+      "loss": 1.6128,
+      "step": 32
+    },
+    {
+      "epoch": 0.308141231397724,
+      "grad_norm": 0.04482797539629353,
+      "learning_rate": 1.9847082234995172e-05,
+      "loss": 1.5791,
+      "step": 33
+    },
+    {
+      "epoch": 0.31747884447038227,
+      "grad_norm": 0.04130893521590142,
+      "learning_rate": 1.982861585384869e-05,
+      "loss": 1.5063,
+      "step": 34
+    },
+    {
+      "epoch": 0.32681645754304056,
+      "grad_norm": 0.04057758728502848,
+      "learning_rate": 1.9809106702558277e-05,
+      "loss": 1.4877,
+      "step": 35
+    },
+    {
+      "epoch": 0.33615407061569885,
+      "grad_norm": 0.038180219331760724,
+      "learning_rate": 1.978855685095358e-05,
+      "loss": 1.4397,
+      "step": 36
+    },
+    {
+      "epoch": 0.33615407061569885,
+      "eval_loss": 1.3349355459213257,
+      "eval_runtime": 16.413,
+      "eval_samples_per_second": 16.816,
+      "eval_steps_per_second": 2.132,
+      "step": 36
+    },
+    {
+      "epoch": 0.34549168368835714,
+      "grad_norm": 0.0360323358530499,
+      "learning_rate": 1.9766968479277684e-05,
+      "loss": 1.3977,
+      "step": 37
+    },
+    {
+      "epoch": 0.3548292967610155,
+      "grad_norm": 0.035146043601200425,
+      "learning_rate": 1.974434387795579e-05,
+      "loss": 1.418,
+      "step": 38
+    },
+    {
+      "epoch": 0.3641669098336738,
+      "grad_norm": 0.033121913235728655,
+      "learning_rate": 1.972068544735221e-05,
+      "loss": 1.3359,
+      "step": 39
+    },
+    {
+      "epoch": 0.37350452290633207,
+      "grad_norm": 0.031036722451915032,
+      "learning_rate": 1.969599569751571e-05,
+      "loss": 1.3401,
+      "step": 40
+    },
+    {
+      "epoch": 0.38284213597899036,
+      "grad_norm": 0.030669047831503912,
+      "learning_rate": 1.9670277247913205e-05,
+      "loss": 1.2638,
+      "step": 41
+    },
+    {
+      "epoch": 0.39217974905164865,
+      "grad_norm": 0.03002785386211285,
+      "learning_rate": 1.964353282715183e-05,
+      "loss": 1.2649,
+      "step": 42
+    },
+    {
+      "epoch": 0.401517362124307,
+      "grad_norm": 0.02817025139456513,
+      "learning_rate": 1.961576527268946e-05,
+      "loss": 1.2476,
+      "step": 43
+    },
+    {
+      "epoch": 0.4108549751969653,
+      "grad_norm": 0.03110381051330927,
+      "learning_rate": 1.9586977530533677e-05,
+      "loss": 1.1923,
+      "step": 44
+    },
+    {
+      "epoch": 0.4201925882696236,
+      "grad_norm": 0.031162812150372473,
+      "learning_rate": 1.95571726549292e-05,
+      "loss": 1.2044,
+      "step": 45
+    },
+    {
+      "epoch": 0.42953020134228187,
+      "grad_norm": 0.029584239693380046,
+      "learning_rate": 1.9526353808033827e-05,
+      "loss": 1.1764,
+      "step": 46
+    },
+    {
+      "epoch": 0.43886781441494016,
+      "grad_norm": 0.027540684267770742,
+      "learning_rate": 1.9494524259582994e-05,
+      "loss": 1.1634,
+      "step": 47
+    },
+    {
+      "epoch": 0.4482054274875985,
+      "grad_norm": 0.028821860863637565,
+      "learning_rate": 1.9461687386542826e-05,
+      "loss": 1.1426,
+      "step": 48
+    },
+    {
+      "epoch": 0.4575430405602568,
+      "grad_norm": 0.027945951918964533,
+      "learning_rate": 1.9427846672751873e-05,
+      "loss": 1.1154,
+      "step": 49
+    },
+    {
+      "epoch": 0.4668806536329151,
+      "grad_norm": 0.026881160663849425,
+      "learning_rate": 1.93930057085515e-05,
+      "loss": 1.0981,
+      "step": 50
+    },
+    {
+      "epoch": 0.4762182667055734,
+      "grad_norm": 0.025002230522002903,
+      "learning_rate": 1.9357168190404937e-05,
+      "loss": 1.0873,
+      "step": 51
+    },
+    {
+      "epoch": 0.48555587977823167,
+      "grad_norm": 0.02318894097448886,
+      "learning_rate": 1.932033792050515e-05,
+      "loss": 1.0701,
+      "step": 52
+    },
+    {
+      "epoch": 0.49489349285089,
+      "grad_norm": 0.021872991266037325,
+      "learning_rate": 1.928251880637141e-05,
+      "loss": 1.0439,
+      "step": 53
+    },
+    {
+      "epoch": 0.5042311059235483,
+      "grad_norm": 0.021369213964442125,
+      "learning_rate": 1.924371486043473e-05,
+      "loss": 1.062,
+      "step": 54
+    },
+    {
+      "epoch": 0.5135687189962066,
+      "grad_norm": 0.02074830989778219,
+      "learning_rate": 1.920393019961217e-05,
+      "loss": 1.0392,
+      "step": 55
+    },
+    {
+      "epoch": 0.5229063320688649,
+      "grad_norm": 0.020345018132100934,
+      "learning_rate": 1.916316904487005e-05,
+      "loss": 1.02,
+      "step": 56
+    },
+    {
+      "epoch": 0.5322439451415232,
+      "grad_norm": 0.01970957719141645,
+      "learning_rate": 1.9121435720776122e-05,
+      "loss": 0.9923,
+      "step": 57
+    },
+    {
+      "epoch": 0.5415815582141815,
+      "grad_norm": 0.019741562276937644,
+      "learning_rate": 1.9078734655040763e-05,
+      "loss": 0.9993,
+      "step": 58
+    },
+    {
+      "epoch": 0.5509191712868398,
+      "grad_norm": 0.0195142739708107,
+      "learning_rate": 1.9035070378047204e-05,
+      "loss": 0.9943,
+      "step": 59
+    },
+    {
+      "epoch": 0.5602567843594981,
+      "grad_norm": 0.017297498916208055,
+      "learning_rate": 1.8990447522370886e-05,
+      "loss": 0.9753,
+      "step": 60
+    },
+    {
+      "epoch": 0.5695943974321565,
+      "grad_norm": 0.01658779952035472,
+      "learning_rate": 1.8944870822287957e-05,
+      "loss": 0.9883,
+      "step": 61
+    },
+    {
+      "epoch": 0.5789320105048147,
+      "grad_norm": 0.018530815854782102,
+      "learning_rate": 1.8898345113273e-05,
+      "loss": 0.9482,
+      "step": 62
+    },
+    {
+      "epoch": 0.588269623577473,
+      "grad_norm": 0.017748842064851857,
+      "learning_rate": 1.8850875331485996e-05,
+      "loss": 0.9533,
+      "step": 63
+    },
+    {
+      "epoch": 0.5976072366501313,
+      "grad_norm": 0.016871546397150952,
+      "learning_rate": 1.8802466513248635e-05,
+      "loss": 0.948,
+      "step": 64
+    },
+    {
+      "epoch": 0.6069448497227896,
+      "grad_norm": 0.01608669807726683,
+      "learning_rate": 1.8753123794509974e-05,
+      "loss": 0.9295,
+      "step": 65
+    },
+    {
+      "epoch": 0.616282462795448,
+      "grad_norm": 0.017453769141315367,
+      "learning_rate": 1.8702852410301556e-05,
+      "loss": 0.9198,
+      "step": 66
+    },
+    {
+      "epoch": 0.6256200758681062,
+      "grad_norm": 0.017655286026566987,
+      "learning_rate": 1.865165769418196e-05,
+      "loss": 0.9535,
+      "step": 67
+    },
+    {
+      "epoch": 0.6349576889407645,
+      "grad_norm": 0.017498254303415152,
+      "learning_rate": 1.8599545077670983e-05,
+      "loss": 0.9278,
+      "step": 68
+    },
+    {
+      "epoch": 0.6442953020134228,
+      "grad_norm": 0.01573094552466445,
+      "learning_rate": 1.854652008967335e-05,
+      "loss": 0.912,
+      "step": 69
+    },
+    {
+      "epoch": 0.6536329150860811,
+      "grad_norm": 0.015035283908585596,
+      "learning_rate": 1.8492588355892125e-05,
+      "loss": 0.9182,
+      "step": 70
+    },
+    {
+      "epoch": 0.6629705281587395,
+      "grad_norm": 0.01593275185300073,
+      "learning_rate": 1.8437755598231857e-05,
+      "loss": 0.9109,
+      "step": 71
+    },
+    {
+      "epoch": 0.6723081412313977,
+      "grad_norm": 0.014842299347889509,
+      "learning_rate": 1.8382027634191523e-05,
+      "loss": 0.8795,
+      "step": 72
+    },
+    {
+      "epoch": 0.6723081412313977,
+      "eval_loss": 0.8388988375663757,
+      "eval_runtime": 16.5273,
+      "eval_samples_per_second": 16.7,
+      "eval_steps_per_second": 2.118,
+      "step": 72
+    },
+    {
+      "epoch": 0.681645754304056,
+      "grad_norm": 0.015086756882680958,
+      "learning_rate": 1.8325410376247295e-05,
+      "loss": 0.8793,
+      "step": 73
+    },
+    {
+      "epoch": 0.6909833673767143,
+      "grad_norm": 0.014400642274755383,
+      "learning_rate": 1.826790983122527e-05,
+      "loss": 0.9139,
+      "step": 74
+    },
+    {
+      "epoch": 0.7003209804493726,
+      "grad_norm": 0.013490998034899438,
+      "learning_rate": 1.8209532099664177e-05,
+      "loss": 0.8767,
+      "step": 75
+    },
+    {
+      "epoch": 0.709658593522031,
+      "grad_norm": 0.011444201595399665,
+      "learning_rate": 1.8150283375168112e-05,
+      "loss": 0.8525,
+      "step": 76
+    },
+    {
+      "epoch": 0.7189962065946892,
+      "grad_norm": 0.01133323463508391,
+      "learning_rate": 1.8090169943749477e-05,
+      "loss": 0.8658,
+      "step": 77
+    },
+    {
+      "epoch": 0.7283338196673476,
+      "grad_norm": 0.011984830212373445,
+      "learning_rate": 1.8029198183162e-05,
+      "loss": 0.856,
+      "step": 78
+    },
+    {
+      "epoch": 0.7376714327400058,
+      "grad_norm": 0.01217011221195066,
+      "learning_rate": 1.796737456222413e-05,
+      "loss": 0.8892,
+      "step": 79
+    },
+    {
+      "epoch": 0.7470090458126641,
+      "grad_norm": 0.012510369655517745,
+      "learning_rate": 1.7904705640132717e-05,
+      "loss": 0.8562,
+      "step": 80
+    },
+    {
+      "epoch": 0.7563466588853225,
+      "grad_norm": 0.011806935544189488,
+      "learning_rate": 1.7841198065767107e-05,
+      "loss": 0.8644,
+      "step": 81
+    },
+    {
+      "epoch": 0.7656842719579807,
+      "grad_norm": 0.011433541340869584,
+      "learning_rate": 1.7776858576983713e-05,
+      "loss": 0.8692,
+      "step": 82
+    },
+    {
+      "epoch": 0.7750218850306391,
+      "grad_norm": 0.012566348663586087,
+      "learning_rate": 1.771169399990119e-05,
+      "loss": 0.8541,
+      "step": 83
+    },
+    {
+      "epoch": 0.7843594981032973,
+      "grad_norm": 0.01072107470914378,
+      "learning_rate": 1.7645711248176198e-05,
+      "loss": 0.8403,
+      "step": 84
+    },
+    {
+      "epoch": 0.7936971111759556,
+      "grad_norm": 0.010617575681982286,
+      "learning_rate": 1.7578917322269885e-05,
+      "loss": 0.81,
+      "step": 85
+    },
+    {
+      "epoch": 0.803034724248614,
+      "grad_norm": 0.010466741230697605,
+      "learning_rate": 1.7511319308705198e-05,
+      "loss": 0.8253,
+      "step": 86
+    },
+    {
+      "epoch": 0.8123723373212722,
+      "grad_norm": 0.010184306281966645,
+      "learning_rate": 1.744292437931502e-05,
+      "loss": 0.8526,
+      "step": 87
+    },
+    {
+      "epoch": 0.8217099503939306,
+      "grad_norm": 0.009850665553519504,
+      "learning_rate": 1.7373739790481263e-05,
+      "loss": 0.8304,
+      "step": 88
+    },
+    {
+      "epoch": 0.8310475634665888,
+      "grad_norm": 0.009236577038055978,
+      "learning_rate": 1.7303772882365018e-05,
+      "loss": 0.7908,
+      "step": 89
+    },
+    {
+      "epoch": 0.8403851765392472,
+      "grad_norm": 0.03711865800548727,
+      "learning_rate": 1.723303107812779e-05,
+      "loss": 0.8085,
+      "step": 90
+    },
+    {
+      "epoch": 0.8497227896119055,
+      "grad_norm": 0.010517236470326855,
+      "learning_rate": 1.7161521883143936e-05,
+      "loss": 0.8313,
+      "step": 91
+    },
+    {
+      "epoch": 0.8590604026845637,
+      "grad_norm": 0.011238580368575812,
+      "learning_rate": 1.7089252884204376e-05,
+      "loss": 0.814,
+      "step": 92
+    },
+    {
+      "epoch": 0.8683980157572221,
+      "grad_norm": 0.009935862606933948,
+      "learning_rate": 1.701623174871168e-05,
+      "loss": 0.8032,
+      "step": 93
+    },
+    {
+      "epoch": 0.8777356288298803,
+      "grad_norm": 0.009655128505771573,
+      "learning_rate": 1.6942466223866582e-05,
+      "loss": 0.8013,
+      "step": 94
+    },
+    {
+      "epoch": 0.8870732419025387,
+      "grad_norm": 0.010463930618313586,
+      "learning_rate": 1.6867964135846043e-05,
+      "loss": 0.8116,
+      "step": 95
+    },
+    {
+      "epoch": 0.896410854975197,
+      "grad_norm": 0.009360797680160574,
+      "learning_rate": 1.679273338897293e-05,
+      "loss": 0.8064,
+      "step": 96
+    },
+    {
+      "epoch": 0.9057484680478552,
+      "grad_norm": 0.00968251573081083,
+      "learning_rate": 1.6716781964877413e-05,
+      "loss": 0.8208,
+      "step": 97
+    },
+    {
+      "epoch": 0.9150860811205136,
+      "grad_norm": 0.009894122849502883,
+      "learning_rate": 1.664011792165012e-05,
+      "loss": 0.8042,
+      "step": 98
+    },
+    {
+      "epoch": 0.9244236941931718,
+      "grad_norm": 0.010626027391586145,
+      "learning_rate": 1.6562749392987255e-05,
+      "loss": 0.8037,
+      "step": 99
+    },
+    {
+      "epoch": 0.9337613072658302,
+      "grad_norm": 0.01083144628829605,
+      "learning_rate": 1.648468458732762e-05,
+      "loss": 0.8079,
+      "step": 100
+    },
+    {
+      "epoch": 0.9430989203384885,
+      "grad_norm": 0.012651590456989211,
+      "learning_rate": 1.6405931786981753e-05,
+      "loss": 0.7903,
+      "step": 101
+    },
+    {
+      "epoch": 0.9524365334111468,
+      "grad_norm": 0.010619166674062264,
+      "learning_rate": 1.6326499347253206e-05,
+      "loss": 0.7783,
+      "step": 102
+    },
+    {
+      "epoch": 0.9617741464838051,
+      "grad_norm": 0.009320303165881526,
+      "learning_rate": 1.6246395695552086e-05,
+      "loss": 0.8035,
+      "step": 103
+    },
+    {
+      "epoch": 0.9711117595564633,
+      "grad_norm": 0.009094019045947338,
+      "learning_rate": 1.6165629330500952e-05,
+      "loss": 0.7839,
+      "step": 104
+    },
+    {
+      "epoch": 0.9804493726291217,
+      "grad_norm": 0.009044004451355656,
+      "learning_rate": 1.6084208821033152e-05,
+      "loss": 0.7641,
+      "step": 105
+    },
+    {
+      "epoch": 0.98978698570178,
+      "grad_norm": 0.009202270165387296,
+      "learning_rate": 1.6002142805483686e-05,
+      "loss": 0.7666,
+      "step": 106
+    },
+    {
+      "epoch": 0.9991245987744383,
+      "grad_norm": 0.009577150303950122,
+      "learning_rate": 1.591943999067273e-05,
+      "loss": 0.7869,
+      "step": 107
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.009577150303950122,
+      "learning_rate": 1.5836109150981885e-05,
+      "loss": 0.7726,
+      "step": 108
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 0.729849100112915,
+      "eval_runtime": 16.4211,
+      "eval_samples_per_second": 16.808,
+      "eval_steps_per_second": 2.131,
+      "step": 108
+    },
+    {
+      "epoch": 1.0093376130726583,
+      "grad_norm": 0.025397375893953918,
+      "learning_rate": 1.5752159127423262e-05,
+      "loss": 0.7934,
+      "step": 109
+    },
+    {
+      "epoch": 1.0186752261453167,
+      "grad_norm": 0.010902737901808093,
+      "learning_rate": 1.5667598826701463e-05,
+      "loss": 0.7627,
+      "step": 110
+    },
+    {
+      "epoch": 1.0280128392179748,
+      "grad_norm": 0.010800327585788388,
+      "learning_rate": 1.5582437220268648e-05,
+      "loss": 0.7578,
+      "step": 111
+    },
+    {
+      "epoch": 1.0373504522906332,
+      "grad_norm": 0.009732201305595227,
+      "learning_rate": 1.549668334337271e-05,
+      "loss": 0.7786,
+      "step": 112
+    },
+    {
+      "epoch": 1.0466880653632915,
+      "grad_norm": 0.009646638742060199,
+      "learning_rate": 1.541034629409865e-05,
+      "loss": 0.7867,
+      "step": 113
+    },
+    {
+      "epoch": 1.0560256784359499,
+      "grad_norm": 0.008481219102834343,
+      "learning_rate": 1.532343523240334e-05,
+      "loss": 0.7728,
+      "step": 114
+    },
+    {
+      "epoch": 1.0653632915086082,
+      "grad_norm": 0.007896528546841378,
+      "learning_rate": 1.523595937914368e-05,
+      "loss": 0.7478,
+      "step": 115
+    },
+    {
+      "epoch": 1.0747009045812663,
+      "grad_norm": 0.008638298804510322,
+      "learning_rate": 1.5147928015098309e-05,
+      "loss": 0.7479,
+      "step": 116
+    },
+    {
+      "epoch": 1.0840385176539247,
+      "grad_norm": 0.009113972298077106,
+      "learning_rate": 1.5059350479982966e-05,
+      "loss": 0.765,
+      "step": 117
+    },
+    {
+      "epoch": 1.093376130726583,
+      "grad_norm": 0.007570635714068737,
+      "learning_rate": 1.497023617145958e-05,
+      "loss": 0.7401,
+      "step": 118
+    },
+    {
+      "epoch": 1.1027137437992414,
+      "grad_norm": 0.007964839741138431,
+      "learning_rate": 1.488059454413923e-05,
+      "loss": 0.7397,
+      "step": 119
+    },
+    {
+      "epoch": 1.1120513568718997,
+      "grad_norm": 0.008981954045478064,
+      "learning_rate": 1.4790435108579048e-05,
+      "loss": 0.7497,
+      "step": 120
+    },
+    {
+      "epoch": 1.1213889699445578,
+      "grad_norm": 0.010253710038366608,
+      "learning_rate": 1.4699767430273202e-05,
+      "loss": 0.747,
+      "step": 121
+    },
+    {
+      "epoch": 1.1307265830172162,
+      "grad_norm": 0.009127204104812056,
+      "learning_rate": 1.4608601128638027e-05,
+      "loss": 0.7288,
+      "step": 122
+    },
+    {
+      "epoch": 1.1400641960898745,
+      "grad_norm": 0.008391872101434881,
+      "learning_rate": 1.4516945875991472e-05,
+      "loss": 0.7509,
+      "step": 123
+    },
+    {
+      "epoch": 1.1494018091625329,
+      "grad_norm": 0.007969062908818022,
+      "learning_rate": 1.4424811396526892e-05,
+      "loss": 0.7475,
+      "step": 124
+    },
+    {
+      "epoch": 1.1587394222351912,
+      "grad_norm": 0.011220499027460716,
+      "learning_rate": 1.4332207465281365e-05,
+      "loss": 0.7553,
+      "step": 125
+    },
+    {
+      "epoch": 1.1680770353078493,
+      "grad_norm": 0.008438223962809686,
+      "learning_rate": 1.423914390709861e-05,
+      "loss": 0.7417,
+      "step": 126
+    },
+    {
+      "epoch": 1.1774146483805077,
+      "grad_norm": 0.008124062455599669,
+      "learning_rate": 1.4145630595586607e-05,
+      "loss": 0.7315,
+      "step": 127
+    },
+    {
+      "epoch": 1.186752261453166,
+      "grad_norm": 0.008212207725829243,
+      "learning_rate": 1.4051677452070064e-05,
+      "loss": 0.7358,
+      "step": 128
+    },
+    {
+      "epoch": 1.1960898745258244,
+      "grad_norm": 0.007992807301141594,
+      "learning_rate": 1.3957294444537808e-05,
+      "loss": 0.7211,
+      "step": 129
+    },
+    {
+      "epoch": 1.2054274875984827,
+      "grad_norm": 0.006862436811624155,
+      "learning_rate": 1.386249158658522e-05,
+      "loss": 0.7328,
+      "step": 130
+    },
+    {
+      "epoch": 1.2147651006711409,
+      "grad_norm": 0.007484363453260225,
+      "learning_rate": 1.3767278936351853e-05,
+      "loss": 0.7409,
+      "step": 131
+    },
+    {
+      "epoch": 1.2241027137437992,
+      "grad_norm": 0.0076572272628448925,
+      "learning_rate": 1.3671666595454296e-05,
+      "loss": 0.7486,
+      "step": 132
+    },
+    {
+      "epoch": 1.2334403268164575,
+      "grad_norm": 0.008131184001251386,
+      "learning_rate": 1.357566470791445e-05,
+      "loss": 0.7376,
+      "step": 133
+    },
+    {
+      "epoch": 1.242777939889116,
+      "grad_norm": 0.008560166789615074,
+      "learning_rate": 1.347928345908329e-05,
+      "loss": 0.7506,
+      "step": 134
+    },
+    {
+      "epoch": 1.252115552961774,
+      "grad_norm": 0.007528447805555166,
+      "learning_rate": 1.3382533074560256e-05,
+      "loss": 0.7376,
+      "step": 135
+    },
+    {
+      "epoch": 1.2614531660344324,
+      "grad_norm": 0.007242664476183943,
+      "learning_rate": 1.3285423819108349e-05,
+      "loss": 0.7294,
+      "step": 136
+    },
+    {
+      "epoch": 1.2707907791070907,
+      "grad_norm": 0.007533824122445911,
+      "learning_rate": 1.3187965995565098e-05,
+      "loss": 0.7266,
+      "step": 137
+    },
+    {
+      "epoch": 1.280128392179749,
+      "grad_norm": 0.007327965197210789,
+      "learning_rate": 1.3090169943749475e-05,
+      "loss": 0.7303,
+      "step": 138
+    },
+    {
+      "epoch": 1.2894660052524074,
+      "grad_norm": 0.006703326795271757,
+      "learning_rate": 1.2992046039364893e-05,
+      "loss": 0.7157,
+      "step": 139
+    },
+    {
+      "epoch": 1.2988036183250657,
+      "grad_norm": 0.006779822330163705,
+      "learning_rate": 1.2893604692898381e-05,
+      "loss": 0.7244,
+      "step": 140
+    },
+    {
+      "epoch": 1.308141231397724,
+      "grad_norm": 0.006809286300573549,
+      "learning_rate": 1.2794856348516095e-05,
+      "loss": 0.738,
+      "step": 141
+    },
+    {
+      "epoch": 1.3174788444703822,
+      "grad_norm": 0.006995815937146753,
+      "learning_rate": 1.2695811482955227e-05,
+      "loss": 0.7257,
+      "step": 142
+    },
+    {
+      "epoch": 1.3268164575430406,
+      "grad_norm": 0.007022225511576764,
+      "learning_rate": 1.2596480604412485e-05,
+      "loss": 0.7344,
+      "step": 143
+    },
+    {
+      "epoch": 1.336154070615699,
+      "grad_norm": 0.006750693986284064,
+      "learning_rate": 1.24968742514292e-05,
+      "loss": 0.7374,
+      "step": 144
+    },
+    {
+      "epoch": 1.336154070615699,
+      "eval_loss": 0.6810739636421204,
+      "eval_runtime": 16.3456,
+      "eval_samples_per_second": 16.885,
+      "eval_steps_per_second": 2.141,
+      "step": 144
+    },
+    {
+      "epoch": 1.345491683688357,
+      "grad_norm": 0.00747295934805212,
+      "learning_rate": 1.2397002991773277e-05,
+      "loss": 0.7236,
+      "step": 145
+    },
+    {
+      "epoch": 1.3548292967610154,
+      "grad_norm": 0.006404283125929898,
+      "learning_rate": 1.2296877421317958e-05,
+      "loss": 0.7047,
+      "step": 146
+    },
+    {
+      "epoch": 1.3641669098336737,
+      "grad_norm": 0.007306315558814404,
+      "learning_rate": 1.2196508162917678e-05,
+      "loss": 0.7195,
+      "step": 147
+    },
+    {
+      "epoch": 1.373504522906332,
+      "grad_norm": 0.007020070609447835,
+      "learning_rate": 1.2095905865281026e-05,
+      "loss": 0.7154,
+      "step": 148
+    },
+    {
+      "epoch": 1.3828421359789904,
+      "grad_norm": 0.007513917861345975,
+      "learning_rate": 1.1995081201840958e-05,
+      "loss": 0.7175,
+      "step": 149
+    },
+    {
+      "epoch": 1.3921797490516488,
+      "grad_norm": 0.006157056101055128,
+      "learning_rate": 1.1894044869622403e-05,
+      "loss": 0.711,
+      "step": 150
+    },
+    {
+      "epoch": 1.401517362124307,
+      "grad_norm": 0.006769156388392959,
+      "learning_rate": 1.1792807588107358e-05,
+      "loss": 0.7162,
+      "step": 151
+    },
+    {
+      "epoch": 1.4108549751969652,
+      "grad_norm": 0.007625740849364638,
+      "learning_rate": 1.1691380098097598e-05,
+      "loss": 0.7074,
+      "step": 152
+    },
+    {
+      "epoch": 1.4201925882696236,
+      "grad_norm": 0.006751113454553622,
+      "learning_rate": 1.158977316057513e-05,
+      "loss": 0.7139,
+      "step": 153
+    },
+    {
+      "epoch": 1.429530201342282,
+      "grad_norm": 0.006999228085598098,
+      "learning_rate": 1.1487997555560503e-05,
+      "loss": 0.7129,
+      "step": 154
+    },
+    {
+      "epoch": 1.43886781441494,
+      "grad_norm": 0.007159925361221997,
+      "learning_rate": 1.1386064080969095e-05,
+      "loss": 0.7198,
+      "step": 155
+    },
+    {
+      "epoch": 1.4482054274875984,
+      "grad_norm": 0.006542656052835091,
+      "learning_rate": 1.1283983551465512e-05,
+      "loss": 0.6975,
+      "step": 156
+    },
+    {
+      "epoch": 1.4575430405602567,
+      "grad_norm": 0.007362347745191795,
+      "learning_rate": 1.118176679731619e-05,
+      "loss": 0.7233,
+      "step": 157
+    },
+    {
+      "epoch": 1.466880653632915,
+      "grad_norm": 0.007507454634951982,
+      "learning_rate": 1.1079424663240372e-05,
+      "loss": 0.7013,
+      "step": 158
+    },
+    {
+      "epoch": 1.4762182667055734,
+      "grad_norm": 0.006924288571642269,
+      "learning_rate": 1.0976968007259519e-05,
+      "loss": 0.7169,
+      "step": 159
+    },
+    {
+      "epoch": 1.4855558797782318,
+      "grad_norm": 0.007327013436583258,
+      "learning_rate": 1.0874407699545329e-05,
+      "loss": 0.7045,
+      "step": 160
+    },
+    {
+      "epoch": 1.4948934928508901,
+      "grad_norm": 0.0067947511037762314,
+      "learning_rate": 1.0771754621266466e-05,
+      "loss": 0.7114,
+      "step": 161
+    },
+    {
+      "epoch": 1.5042311059235483,
+      "grad_norm": 0.007479624734064453,
+      "learning_rate": 1.0669019663434117e-05,
+      "loss": 0.7081,
+      "step": 162
+    },
+    {
+      "epoch": 1.5135687189962066,
+      "grad_norm": 0.0071184716193534595,
+      "learning_rate": 1.0566213725746506e-05,
+      "loss": 0.7277,
+      "step": 163
+    },
+    {
+      "epoch": 1.522906332068865,
+      "grad_norm": 0.007760294512478719,
+      "learning_rate": 1.0463347715432488e-05,
+      "loss": 0.7196,
+      "step": 164
+    },
+    {
+      "epoch": 1.532243945141523,
+      "grad_norm": 0.007716707922551307,
+      "learning_rate": 1.0360432546094341e-05,
+      "loss": 0.7197,
+      "step": 165
+    },
+    {
+      "epoch": 1.5415815582141814,
+      "grad_norm": 0.0072311410748593635,
+      "learning_rate": 1.0257479136549889e-05,
+      "loss": 0.7096,
+      "step": 166
+    },
+    {
+      "epoch": 1.5509191712868398,
+      "grad_norm": 0.006676953574505626,
+      "learning_rate": 1.0154498409674051e-05,
+      "loss": 0.6982,
+      "step": 167
+    },
+    {
+      "epoch": 1.560256784359498,
+      "grad_norm": 0.006468587987533156,
+      "learning_rate": 1.0051501291240008e-05,
+      "loss": 0.6774,
+      "step": 168
+    },
+    {
+      "epoch": 1.5695943974321565,
+      "grad_norm": 0.006253950592479604,
+      "learning_rate": 9.948498708759993e-06,
+      "loss": 0.688,
+      "step": 169
+    },
+    {
+      "epoch": 1.5789320105048148,
+      "grad_norm": 0.0065410655967948214,
+      "learning_rate": 9.845501590325949e-06,
+      "loss": 0.7207,
+      "step": 170
+    },
+    {
+      "epoch": 1.5882696235774731,
+      "grad_norm": 0.006364666727868252,
+      "learning_rate": 9.742520863450116e-06,
+      "loss": 0.7027,
+      "step": 171
+    },
+    {
+      "epoch": 1.5976072366501313,
+      "grad_norm": 0.006174978009841133,
+      "learning_rate": 9.639567453905662e-06,
+      "loss": 0.7021,
+      "step": 172
+    },
+    {
+      "epoch": 1.6069448497227896,
+      "grad_norm": 0.006819554213970836,
+      "learning_rate": 9.536652284567514e-06,
+      "loss": 0.6707,
+      "step": 173
+    },
+    {
+      "epoch": 1.616282462795448,
+      "grad_norm": 0.006939549814926166,
+      "learning_rate": 9.433786274253496e-06,
+      "loss": 0.688,
+      "step": 174
+    },
+    {
+      "epoch": 1.625620075868106,
+      "grad_norm": 0.006694022655954106,
+      "learning_rate": 9.330980336565887e-06,
+      "loss": 0.6986,
+      "step": 175
+    },
+    {
+      "epoch": 1.6349576889407644,
+      "grad_norm": 0.006672750770558918,
+      "learning_rate": 9.228245378733537e-06,
+      "loss": 0.7094,
+      "step": 176
+    },
+    {
+      "epoch": 1.6442953020134228,
+      "grad_norm": 0.007263430275454483,
+      "learning_rate": 9.125592300454675e-06,
+      "loss": 0.6912,
+      "step": 177
+    },
+    {
+      "epoch": 1.6536329150860811,
+      "grad_norm": 0.007540180086839557,
+      "learning_rate": 9.023031992740488e-06,
+      "loss": 0.6792,
+      "step": 178
+    },
+    {
+      "epoch": 1.6629705281587395,
+      "grad_norm": 0.0064199964583914505,
+      "learning_rate": 8.92057533675963e-06,
+      "loss": 0.688,
+      "step": 179
+    },
+    {
+      "epoch": 1.6723081412313978,
+      "grad_norm": 0.006196992820903783,
+      "learning_rate": 8.818233202683815e-06,
+      "loss": 0.6928,
+      "step": 180
+    },
+    {
+      "epoch": 1.6723081412313978,
+      "eval_loss": 0.6554006934165955,
+      "eval_runtime": 16.766,
+      "eval_samples_per_second": 16.462,
+      "eval_steps_per_second": 2.088,
+      "step": 180
+    },
+    {
+      "epoch": 1.6816457543040562,
+      "grad_norm": 0.006747686143721935,
+      "learning_rate": 8.71601644853449e-06,
+      "loss": 0.7055,
+      "step": 181
+    },
+    {
+      "epoch": 1.6909833673767143,
+      "grad_norm": 0.006142383641399304,
+      "learning_rate": 8.613935919030908e-06,
+      "loss": 0.6897,
+      "step": 182
+    },
+    {
+      "epoch": 1.7003209804493726,
+      "grad_norm": 0.005877324668671395,
+      "learning_rate": 8.512002444439502e-06,
+      "loss": 0.7044,
+      "step": 183
+    },
+    {
+      "epoch": 1.709658593522031,
+      "grad_norm": 0.006855381402076154,
+      "learning_rate": 8.410226839424871e-06,
+      "loss": 0.6819,
+      "step": 184
+    },
+    {
+      "epoch": 1.718996206594689,
+      "grad_norm": 0.005424415467423937,
+      "learning_rate": 8.308619901902406e-06,
+      "loss": 0.7078,
+      "step": 185
+    },
+    {
+      "epoch": 1.7283338196673474,
+      "grad_norm": 0.0061132223152236785,
+      "learning_rate": 8.207192411892645e-06,
+      "loss": 0.6931,
+      "step": 186
+    },
+    {
+      "epoch": 1.7376714327400058,
+      "grad_norm": 0.006419370473537233,
+      "learning_rate": 8.1059551303776e-06,
+      "loss": 0.7123,
+      "step": 187
+    },
+    {
+      "epoch": 1.7470090458126641,
+      "grad_norm": 0.0063388625830988875,
+      "learning_rate": 8.004918798159046e-06,
+      "loss": 0.6867,
+      "step": 188
+    },
+    {
+      "epoch": 1.7563466588853225,
+      "grad_norm": 0.0058968715521128074,
+      "learning_rate": 7.904094134718975e-06,
+      "loss": 0.6864,
+      "step": 189
+    },
+    {
+      "epoch": 1.7656842719579808,
+      "grad_norm": 0.0064607553768266796,
+      "learning_rate": 7.803491837082324e-06,
+      "loss": 0.6901,
+      "step": 190
+    },
+    {
+      "epoch": 1.7750218850306392,
+      "grad_norm": 0.0067065346597906346,
+      "learning_rate": 7.703122578682047e-06,
+      "loss": 0.7002,
+      "step": 191
+    },
+    {
+      "epoch": 1.7843594981032973,
+      "grad_norm": 0.006375540418233348,
+      "learning_rate": 7.602997008226725e-06,
+      "loss": 0.6952,
+      "step": 192
+    },
+    {
+      "epoch": 1.7936971111759556,
+      "grad_norm": 0.005964849793104097,
+      "learning_rate": 7.503125748570801e-06,
+      "loss": 0.7012,
+      "step": 193
+    },
+    {
+      "epoch": 1.803034724248614,
+      "grad_norm": 0.006923688795252922,
+      "learning_rate": 7.403519395587522e-06,
+      "loss": 0.6873,
+      "step": 194
+    },
+    {
+      "epoch": 1.8123723373212721,
+      "grad_norm": 0.006884309979239693,
+      "learning_rate": 7.304188517044774e-06,
+      "loss": 0.7075,
+      "step": 195
+    },
+    {
+      "epoch": 1.8217099503939305,
+      "grad_norm": 0.006370636558687697,
+      "learning_rate": 7.2051436514839064e-06,
+      "loss": 0.6988,
+      "step": 196
+    },
+    {
+      "epoch": 1.8310475634665888,
+      "grad_norm": 0.00628278997180673,
+      "learning_rate": 7.106395307101621e-06,
+      "loss": 0.6934,
+      "step": 197
+    },
+    {
+      "epoch": 1.8403851765392472,
+      "grad_norm": 0.006032860815901838,
+      "learning_rate": 7.007953960635109e-06,
+      "loss": 0.6938,
+      "step": 198
+    },
+    {
+      "epoch": 1.8497227896119055,
+      "grad_norm": 0.005962967202398698,
+      "learning_rate": 6.909830056250527e-06,
+      "loss": 0.6854,
+      "step": 199
+    },
+    {
+      "epoch": 1.8590604026845639,
+      "grad_norm": 0.00554268821549102,
+      "learning_rate": 6.812034004434904e-06,
+      "loss": 0.6851,
+      "step": 200
+    },
+    {
+      "epoch": 1.8683980157572222,
+      "grad_norm": 0.005725484983585548,
+      "learning_rate": 6.714576180891653e-06,
+      "loss": 0.6936,
+      "step": 201
+    },
+    {
+      "epoch": 1.8777356288298803,
+      "grad_norm": 0.006054690781627555,
+      "learning_rate": 6.617466925439746e-06,
+      "loss": 0.6851,
+      "step": 202
+    },
+    {
+      "epoch": 1.8870732419025387,
+      "grad_norm": 0.005708307473106662,
+      "learning_rate": 6.520716540916709e-06,
+      "loss": 0.678,
+      "step": 203
+    },
+    {
+      "epoch": 1.896410854975197,
+      "grad_norm": 0.0059546695263407905,
+      "learning_rate": 6.424335292085553e-06,
+      "loss": 0.6854,
+      "step": 204
+    },
+    {
+      "epoch": 1.9057484680478551,
+      "grad_norm": 0.006488681044848436,
+      "learning_rate": 6.32833340454571e-06,
+      "loss": 0.6796,
+      "step": 205
+    },
+    {
+      "epoch": 1.9150860811205135,
+      "grad_norm": 0.006493914236731745,
+      "learning_rate": 6.232721063648148e-06,
+      "loss": 0.6745,
+      "step": 206
+    },
+    {
+      "epoch": 1.9244236941931718,
+      "grad_norm": 0.006261095082612158,
+      "learning_rate": 6.137508413414784e-06,
+      "loss": 0.6925,
+      "step": 207
+    },
+    {
+      "epoch": 1.9337613072658302,
+      "grad_norm": 0.00619237862435736,
+      "learning_rate": 6.042705555462192e-06,
+      "loss": 0.693,
+      "step": 208
+    },
+    {
+      "epoch": 1.9430989203384885,
+      "grad_norm": 0.005985860419714863,
+      "learning_rate": 5.948322547929939e-06,
+      "loss": 0.6882,
+      "step": 209
+    },
+    {
+      "epoch": 1.9524365334111469,
+      "grad_norm": 0.006018365152445982,
+      "learning_rate": 5.8543694044133984e-06,
+      "loss": 0.7059,
+      "step": 210
+    },
+    {
+      "epoch": 1.9617741464838052,
+      "grad_norm": 0.006178576965302145,
+      "learning_rate": 5.760856092901394e-06,
+      "loss": 0.6872,
+      "step": 211
+    },
+    {
+      "epoch": 1.9711117595564633,
+      "grad_norm": 0.006209928845152174,
+      "learning_rate": 5.667792534718639e-06,
+      "loss": 0.6955,
+      "step": 212
+    },
+    {
+      "epoch": 1.9804493726291217,
+      "grad_norm": 0.006362809796961763,
+      "learning_rate": 5.575188603473112e-06,
+      "loss": 0.6781,
+      "step": 213
+    },
+    {
+      "epoch": 1.98978698570178,
+      "grad_norm": 0.0073402479875403605,
+      "learning_rate": 5.483054124008528e-06,
+      "loss": 0.669,
+      "step": 214
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 321,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 107,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.043316961713062e+18,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-214/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c22096e6a9c9dea84f12ad6313d09723c960a987f46ad9af04dc6c1de407c91
+size 10872

checkpoint-214/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)

checkpoint-321/added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

checkpoint-321/config.json

@@ -0,0 +1,31 @@
+{
+  "_name_or_path": "Qwen/Qwen2.5-Math-1.5B-Instruct",
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 8960,
+  "max_position_embeddings": 32768,
+  "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": {
+    "factor": 8.0,
+    "type": "linear"
+  },
+  "rope_theta": 10000.0,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.48.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151665
+}

checkpoint-321/generation_config.json

@@ -0,0 +1,10 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "transformers_version": "4.48.3"
+}