Upload 34 files

README.md

@@ -1,3 +1,43 @@
----
-license: cc-by-4.0
----
+---
+base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
+library_name: peft
+---
+
+# Model Card for Reviewer-14B
+
+## Model Details
+
+### Model Description
+
+Reviewer-14B is a fine-tuned on [**DeepSeek-R1-Distill-Qwen-14B**](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B), optimized for selecting the best patch among multiple patches generated by our DARS agent while solving software engineering problems.
+
+### Model Sources
+
+- **Repository:** [DARS-14B Repository](https://github.com/darsagent/DARS-Agent)
+- **Paper:** ["DARS: Dynamic Action Re-Sampling to Enhance Coding Agent Performance by Adaptive Tree Traversal"](https://drive.google.com/file/d/1DMAZ-fkirC8uKl8819cOq9J3BQ4E7GXR/view?usp=drive_link)
+
+## How to Get Started with the Model
+We use vLLM to deploy and infer the model. Please follow this tutorial [here]((https://docs.vllm.ai/en/latest/features/lora.html)) to use our LoRA weights with vLLM.
+
+## Training Details
+
+### Dataset
+
+We use our [code review dataset](https://huggingface.co/datasets/AGENTDARS/generated-critiques) where each instance contains several git patches with critiques for each each patch. The model learns to generate critiques for multiple patches and select the best patch.
+
+### Training Procedure
+
+| Hyperparameter       | Value                                      |
+|----------------------|--------------------------------------------|
+| Training regime      | BF16 mixed precision                       |
+| Optimizer            | AdamW with cosine learning rate scheduler  |
+| LoRA Configuration   | rank=8, alpha=32, dropout=0.1              |
+| Batch Size           | 48                           |
+| Learning Rate        | 5e-6                                       |
+| Sequence Length      | 14K tokens                                 |
+| Fine-tuning Epochs   | 1                                          |
+| Compute Environment  | DeepSpeed for memory-efficient distributed training |
+| Compute Infrastructure  | 8x H100 |
+
+## Results
+Using this model as a reviewer with DARS trajectories generated using Claude 3.5 Sonnet V2 achieves 41.7% on SWE-Bench Lite.

adapter_config.json

@@ -0,0 +1,29 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "/home/dars/Qwen2.5-Coder/models/deepseek-14b/",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_dropout": 0.1,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "v_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:157c15152bce8e797dfa01f6e466d29429fa6c83df9d2278c90f23d4b64ebae9
+size 12608496

global_step2000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f8a8aad76113e84b54cf7d3c303a9b19125ac980c9f3bc26ff48c123355f41f
+size 9440912

global_step2000/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a92407e6c77b137b11809a3d6e3332ce01c6b44dfaf906f931ed6c37da568fb
+size 9440912

global_step2000/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3b8fad66457b17e20b005a16d3afbbe7284aa4191b7e41d4141687f8cebb058
+size 9440912

global_step2000/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8c9401c5e7a03c7880f43bb7a862a7b82063f8bfa147b366a529d8c44eb396e
+size 9440912

global_step2000/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e116b3abfbf167542dfb4cf1d05cf724f6e395d41ec236218a47882a54b3a1c1
+size 9440912

global_step2000/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5027c1fa90f40b1c4b9649b6e2a7a4f9e64ac55c5845b383e4fcbf78c499c4c8
+size 9440912

global_step2000/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2b4a4f5b2c0add41f7b9471da74ca7340ae888ae11ac5ab60be2bcc1b2857b9
+size 9440912

global_step2000/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f761562e37a8fb2fc45449f1efdb2898b0ed05d1a200f27e9d6d9ec426137e5
+size 9440912

global_step2000/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3be128cab192df1962334a33f6aaab5f0b2e506c62a5a20dd9ce97c8e7b5cb4
+size 228398

global_step2000/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:463d30b2fb4f41dafe42c4f4bb16db903a7fb192b056f3615bc84331c7d912c9
+size 228398

global_step2000/zero_pp_rank_2_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1969eb6fe91040b1904eeecfcf4cf5918ca5bde4906b9a453d30f3958a357b3
+size 228398

global_step2000/zero_pp_rank_3_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2315308eca50010ad291aa15be29979a4ff7517c31afaabc9965f892e6087863
+size 228398

global_step2000/zero_pp_rank_4_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:329e7389a9356d81428ca5cca5914e87c2c252fa101b8dac3b0fc86bf61dbf80
+size 228398

global_step2000/zero_pp_rank_5_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d108ae6ebfd137dd064635fe64a9abf2031a45ee99359cd357c5f5359a8d5305
+size 228398

global_step2000/zero_pp_rank_6_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6f10280fac8ea18a1b9716f90c4f21ebd1c17f3ec585294a20dee201700ff98
+size 228398

global_step2000/zero_pp_rank_7_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a5782a01e1ef7abdf7d083035263a60a8a479ba17ec06304cbc0426ba3f45fc
+size 228398

latest

@@ -0,0 +1 @@
+global_step2000

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e832b43e588d740d3f07dc0a44d65ab5ce8a3e06e28c23014052b0ba2d54ad21
+size 15920

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d09ac95eeabbb6f44c57a6d15a7dab491542295abbb790c351f496fb5ad3319
+size 15920

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53cb30810e690ff974d7d060095a4a9fb8587590bff41323eb6ee094f71c9647
+size 15920

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06a8c290881b49fb26b7e3876ab4e6409d7506a8d64bd0d3b4e2479e976e5fae
+size 15920

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e3bd90ae12b6576ffc76b434c14036de3a16c05721e44bcb23fc3b67101e66a
+size 15920

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b59c3b71c7beadeea672b59b1f0f6c37896012b23b5763cd194fabf9fc01b41
+size 15920

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76b9e1212c0a9b1c0778a79289f15e73c750fd9510950ae55fdd47b4db37c7a4
+size 15920

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00814da94919dc36b80b7dc43dd9e7f49c59fd651eadb84f772bdf9598171438
+size 15920

special_tokens_map.json

@@ -0,0 +1,27 @@
+{
+  "additional_special_tokens": [
+    {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    }
+  ],
+  "bos_token": {
+    "content": "<｜begin▁of▁sentence｜>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": "<|im_end|>",
+  "pad_token": "<|endoftext|>"
+}

tokenizer_config.json

@@ -0,0 +1,223 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<｜end▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<｜User｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151645": {
+      "content": "<｜Assistant｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151646": {
+      "content": "<｜begin▁of▁sentence｜>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|EOT|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151648": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151649": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151665": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151666": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151667": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_end|>",
+    "<|im_start|>"
+  ],
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\\n' + '```json' + '\\n' + tool['function']['arguments'] + '\\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\\n' + '```json' + '\\n' + tool['function']['arguments'] + '\\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜>'}}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "legacy": true,
+  "model_max_length": 14000,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "truncation": true,
+  "unk_token": null,
+  "use_default_system_prompt": false
+}

trainer_state.json

@@ -0,0 +1,653 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.5482456140350878,
+  "eval_steps": 500,
+  "global_step": 2000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.005482456140350877,
+      "learning_rate": 3.2525749891599526e-06,
+      "loss": 1.627,
+      "step": 20
+    },
+    {
+      "epoch": 0.010964912280701754,
+      "learning_rate": 4.0051499783199055e-06,
+      "loss": 1.6081,
+      "step": 40
+    },
+    {
+      "epoch": 0.01644736842105263,
+      "learning_rate": 4.445378125959108e-06,
+      "loss": 1.587,
+      "step": 60
+    },
+    {
+      "epoch": 0.021929824561403508,
+      "learning_rate": 4.757724967479858e-06,
+      "loss": 1.5592,
+      "step": 80
+    },
+    {
+      "epoch": 0.027412280701754384,
+      "learning_rate": 4.9999999999999996e-06,
+      "loss": 1.5009,
+      "step": 100
+    },
+    {
+      "epoch": 0.03289473684210526,
+      "learning_rate": 4.9756410256410255e-06,
+      "loss": 1.4438,
+      "step": 120
+    },
+    {
+      "epoch": 0.03837719298245614,
+      "learning_rate": 4.95e-06,
+      "loss": 1.3777,
+      "step": 140
+    },
+    {
+      "epoch": 0.043859649122807015,
+      "learning_rate": 4.9243589743589745e-06,
+      "loss": 1.3215,
+      "step": 160
+    },
+    {
+      "epoch": 0.049342105263157895,
+      "learning_rate": 4.898717948717949e-06,
+      "loss": 1.267,
+      "step": 180
+    },
+    {
+      "epoch": 0.05482456140350877,
+      "learning_rate": 4.8730769230769235e-06,
+      "loss": 1.2095,
+      "step": 200
+    },
+    {
+      "epoch": 0.06030701754385965,
+      "learning_rate": 4.847435897435897e-06,
+      "loss": 1.1684,
+      "step": 220
+    },
+    {
+      "epoch": 0.06578947368421052,
+      "learning_rate": 4.821794871794872e-06,
+      "loss": 1.1054,
+      "step": 240
+    },
+    {
+      "epoch": 0.0712719298245614,
+      "learning_rate": 4.796153846153846e-06,
+      "loss": 1.0593,
+      "step": 260
+    },
+    {
+      "epoch": 0.07675438596491228,
+      "learning_rate": 4.770512820512821e-06,
+      "loss": 1.0069,
+      "step": 280
+    },
+    {
+      "epoch": 0.08223684210526316,
+      "learning_rate": 4.744871794871795e-06,
+      "loss": 0.9498,
+      "step": 300
+    },
+    {
+      "epoch": 0.08771929824561403,
+      "learning_rate": 4.71923076923077e-06,
+      "loss": 0.8949,
+      "step": 320
+    },
+    {
+      "epoch": 0.09320175438596491,
+      "learning_rate": 4.693589743589744e-06,
+      "loss": 0.8369,
+      "step": 340
+    },
+    {
+      "epoch": 0.09868421052631579,
+      "learning_rate": 4.667948717948718e-06,
+      "loss": 0.8047,
+      "step": 360
+    },
+    {
+      "epoch": 0.10416666666666667,
+      "learning_rate": 4.642307692307692e-06,
+      "loss": 0.781,
+      "step": 380
+    },
+    {
+      "epoch": 0.10964912280701754,
+      "learning_rate": 4.616666666666667e-06,
+      "loss": 0.7535,
+      "step": 400
+    },
+    {
+      "epoch": 0.11513157894736842,
+      "learning_rate": 4.591025641025641e-06,
+      "loss": 0.7332,
+      "step": 420
+    },
+    {
+      "epoch": 0.1206140350877193,
+      "learning_rate": 4.565384615384616e-06,
+      "loss": 0.7171,
+      "step": 440
+    },
+    {
+      "epoch": 0.12609649122807018,
+      "learning_rate": 4.53974358974359e-06,
+      "loss": 0.7074,
+      "step": 460
+    },
+    {
+      "epoch": 0.13157894736842105,
+      "learning_rate": 4.514102564102564e-06,
+      "loss": 0.6905,
+      "step": 480
+    },
+    {
+      "epoch": 0.13706140350877194,
+      "learning_rate": 4.4884615384615384e-06,
+      "loss": 0.6812,
+      "step": 500
+    },
+    {
+      "epoch": 0.13706140350877194,
+      "eval_loss": 0.6900227665901184,
+      "eval_runtime": 7.1872,
+      "eval_samples_per_second": 13.635,
+      "eval_steps_per_second": 1.809,
+      "step": 500
+    },
+    {
+      "epoch": 0.1425438596491228,
+      "learning_rate": 4.462820512820513e-06,
+      "loss": 0.6753,
+      "step": 520
+    },
+    {
+      "epoch": 0.14802631578947367,
+      "learning_rate": 4.437179487179487e-06,
+      "loss": 0.6637,
+      "step": 540
+    },
+    {
+      "epoch": 0.15350877192982457,
+      "learning_rate": 4.411538461538462e-06,
+      "loss": 0.6615,
+      "step": 560
+    },
+    {
+      "epoch": 0.15899122807017543,
+      "learning_rate": 4.385897435897436e-06,
+      "loss": 0.6547,
+      "step": 580
+    },
+    {
+      "epoch": 0.16447368421052633,
+      "learning_rate": 4.36025641025641e-06,
+      "loss": 0.6487,
+      "step": 600
+    },
+    {
+      "epoch": 0.1699561403508772,
+      "learning_rate": 4.3346153846153846e-06,
+      "loss": 0.6441,
+      "step": 620
+    },
+    {
+      "epoch": 0.17543859649122806,
+      "learning_rate": 4.308974358974359e-06,
+      "loss": 0.6358,
+      "step": 640
+    },
+    {
+      "epoch": 0.18092105263157895,
+      "learning_rate": 4.2833333333333335e-06,
+      "loss": 0.636,
+      "step": 660
+    },
+    {
+      "epoch": 0.18640350877192982,
+      "learning_rate": 4.257692307692308e-06,
+      "loss": 0.6326,
+      "step": 680
+    },
+    {
+      "epoch": 0.19188596491228072,
+      "learning_rate": 4.2320512820512825e-06,
+      "loss": 0.6246,
+      "step": 700
+    },
+    {
+      "epoch": 0.19736842105263158,
+      "learning_rate": 4.206410256410256e-06,
+      "loss": 0.6237,
+      "step": 720
+    },
+    {
+      "epoch": 0.20285087719298245,
+      "learning_rate": 4.180769230769231e-06,
+      "loss": 0.6203,
+      "step": 740
+    },
+    {
+      "epoch": 0.20833333333333334,
+      "learning_rate": 4.155128205128205e-06,
+      "loss": 0.6164,
+      "step": 760
+    },
+    {
+      "epoch": 0.2138157894736842,
+      "learning_rate": 4.12948717948718e-06,
+      "loss": 0.6118,
+      "step": 780
+    },
+    {
+      "epoch": 0.21929824561403508,
+      "learning_rate": 4.103846153846154e-06,
+      "loss": 0.6101,
+      "step": 800
+    },
+    {
+      "epoch": 0.22478070175438597,
+      "learning_rate": 4.078205128205129e-06,
+      "loss": 0.6092,
+      "step": 820
+    },
+    {
+      "epoch": 0.23026315789473684,
+      "learning_rate": 4.052564102564102e-06,
+      "loss": 0.6039,
+      "step": 840
+    },
+    {
+      "epoch": 0.23574561403508773,
+      "learning_rate": 4.026923076923077e-06,
+      "loss": 0.5989,
+      "step": 860
+    },
+    {
+      "epoch": 0.2412280701754386,
+      "learning_rate": 4.001282051282051e-06,
+      "loss": 0.5979,
+      "step": 880
+    },
+    {
+      "epoch": 0.24671052631578946,
+      "learning_rate": 3.975641025641026e-06,
+      "loss": 0.5965,
+      "step": 900
+    },
+    {
+      "epoch": 0.25219298245614036,
+      "learning_rate": 3.95e-06,
+      "loss": 0.5969,
+      "step": 920
+    },
+    {
+      "epoch": 0.2576754385964912,
+      "learning_rate": 3.924358974358975e-06,
+      "loss": 0.5902,
+      "step": 940
+    },
+    {
+      "epoch": 0.2631578947368421,
+      "learning_rate": 3.8987179487179484e-06,
+      "loss": 0.5909,
+      "step": 960
+    },
+    {
+      "epoch": 0.26864035087719296,
+      "learning_rate": 3.873076923076923e-06,
+      "loss": 0.5914,
+      "step": 980
+    },
+    {
+      "epoch": 0.2741228070175439,
+      "learning_rate": 3.8474358974358974e-06,
+      "loss": 0.589,
+      "step": 1000
+    },
+    {
+      "epoch": 0.2741228070175439,
+      "eval_loss": 0.5962496399879456,
+      "eval_runtime": 6.5411,
+      "eval_samples_per_second": 14.982,
+      "eval_steps_per_second": 1.987,
+      "step": 1000
+    },
+    {
+      "epoch": 0.27960526315789475,
+      "learning_rate": 3.821794871794872e-06,
+      "loss": 0.5844,
+      "step": 1020
+    },
+    {
+      "epoch": 0.2850877192982456,
+      "learning_rate": 3.7961538461538464e-06,
+      "loss": 0.5819,
+      "step": 1040
+    },
+    {
+      "epoch": 0.2905701754385965,
+      "learning_rate": 3.770512820512821e-06,
+      "loss": 0.5828,
+      "step": 1060
+    },
+    {
+      "epoch": 0.29605263157894735,
+      "learning_rate": 3.744871794871795e-06,
+      "loss": 0.5816,
+      "step": 1080
+    },
+    {
+      "epoch": 0.30153508771929827,
+      "learning_rate": 3.7192307692307695e-06,
+      "loss": 0.5773,
+      "step": 1100
+    },
+    {
+      "epoch": 0.30701754385964913,
+      "learning_rate": 3.693589743589744e-06,
+      "loss": 0.5782,
+      "step": 1120
+    },
+    {
+      "epoch": 0.3125,
+      "learning_rate": 3.6679487179487185e-06,
+      "loss": 0.5774,
+      "step": 1140
+    },
+    {
+      "epoch": 0.31798245614035087,
+      "learning_rate": 3.642307692307693e-06,
+      "loss": 0.5733,
+      "step": 1160
+    },
+    {
+      "epoch": 0.32346491228070173,
+      "learning_rate": 3.616666666666667e-06,
+      "loss": 0.5722,
+      "step": 1180
+    },
+    {
+      "epoch": 0.32894736842105265,
+      "learning_rate": 3.591025641025641e-06,
+      "loss": 0.569,
+      "step": 1200
+    },
+    {
+      "epoch": 0.3344298245614035,
+      "learning_rate": 3.5653846153846156e-06,
+      "loss": 0.5699,
+      "step": 1220
+    },
+    {
+      "epoch": 0.3399122807017544,
+      "learning_rate": 3.53974358974359e-06,
+      "loss": 0.5691,
+      "step": 1240
+    },
+    {
+      "epoch": 0.34539473684210525,
+      "learning_rate": 3.5141025641025646e-06,
+      "loss": 0.5706,
+      "step": 1260
+    },
+    {
+      "epoch": 0.3508771929824561,
+      "learning_rate": 3.488461538461539e-06,
+      "loss": 0.565,
+      "step": 1280
+    },
+    {
+      "epoch": 0.35635964912280704,
+      "learning_rate": 3.462820512820513e-06,
+      "loss": 0.5652,
+      "step": 1300
+    },
+    {
+      "epoch": 0.3618421052631579,
+      "learning_rate": 3.4371794871794873e-06,
+      "loss": 0.5631,
+      "step": 1320
+    },
+    {
+      "epoch": 0.3673245614035088,
+      "learning_rate": 3.4115384615384617e-06,
+      "loss": 0.5652,
+      "step": 1340
+    },
+    {
+      "epoch": 0.37280701754385964,
+      "learning_rate": 3.3858974358974362e-06,
+      "loss": 0.5627,
+      "step": 1360
+    },
+    {
+      "epoch": 0.3782894736842105,
+      "learning_rate": 3.3602564102564107e-06,
+      "loss": 0.5662,
+      "step": 1380
+    },
+    {
+      "epoch": 0.38377192982456143,
+      "learning_rate": 3.3346153846153852e-06,
+      "loss": 0.5606,
+      "step": 1400
+    },
+    {
+      "epoch": 0.3892543859649123,
+      "learning_rate": 3.3089743589743593e-06,
+      "loss": 0.5616,
+      "step": 1420
+    },
+    {
+      "epoch": 0.39473684210526316,
+      "learning_rate": 3.2833333333333334e-06,
+      "loss": 0.5624,
+      "step": 1440
+    },
+    {
+      "epoch": 0.40021929824561403,
+      "learning_rate": 3.257692307692308e-06,
+      "loss": 0.5578,
+      "step": 1460
+    },
+    {
+      "epoch": 0.4057017543859649,
+      "learning_rate": 3.2320512820512824e-06,
+      "loss": 0.5577,
+      "step": 1480
+    },
+    {
+      "epoch": 0.41118421052631576,
+      "learning_rate": 3.206410256410257e-06,
+      "loss": 0.5586,
+      "step": 1500
+    },
+    {
+      "epoch": 0.41118421052631576,
+      "eval_loss": 0.5718954801559448,
+      "eval_runtime": 6.5348,
+      "eval_samples_per_second": 14.997,
+      "eval_steps_per_second": 1.989,
+      "step": 1500
+    },
+    {
+      "epoch": 0.4166666666666667,
+      "learning_rate": 3.1807692307692314e-06,
+      "loss": 0.5603,
+      "step": 1520
+    },
+    {
+      "epoch": 0.42214912280701755,
+      "learning_rate": 3.1551282051282054e-06,
+      "loss": 0.5567,
+      "step": 1540
+    },
+    {
+      "epoch": 0.4276315789473684,
+      "learning_rate": 3.1294871794871795e-06,
+      "loss": 0.5565,
+      "step": 1560
+    },
+    {
+      "epoch": 0.4331140350877193,
+      "learning_rate": 3.103846153846154e-06,
+      "loss": 0.552,
+      "step": 1580
+    },
+    {
+      "epoch": 0.43859649122807015,
+      "learning_rate": 3.0782051282051285e-06,
+      "loss": 0.5553,
+      "step": 1600
+    },
+    {
+      "epoch": 0.4440789473684211,
+      "learning_rate": 3.052564102564103e-06,
+      "loss": 0.5528,
+      "step": 1620
+    },
+    {
+      "epoch": 0.44956140350877194,
+      "learning_rate": 3.0269230769230775e-06,
+      "loss": 0.5446,
+      "step": 1640
+    },
+    {
+      "epoch": 0.4550438596491228,
+      "learning_rate": 3.0012820512820516e-06,
+      "loss": 0.5525,
+      "step": 1660
+    },
+    {
+      "epoch": 0.4605263157894737,
+      "learning_rate": 2.9756410256410256e-06,
+      "loss": 0.5504,
+      "step": 1680
+    },
+    {
+      "epoch": 0.46600877192982454,
+      "learning_rate": 2.95e-06,
+      "loss": 0.5528,
+      "step": 1700
+    },
+    {
+      "epoch": 0.47149122807017546,
+      "learning_rate": 2.9243589743589746e-06,
+      "loss": 0.5502,
+      "step": 1720
+    },
+    {
+      "epoch": 0.4769736842105263,
+      "learning_rate": 2.898717948717949e-06,
+      "loss": 0.5512,
+      "step": 1740
+    },
+    {
+      "epoch": 0.4824561403508772,
+      "learning_rate": 2.8730769230769236e-06,
+      "loss": 0.5476,
+      "step": 1760
+    },
+    {
+      "epoch": 0.48793859649122806,
+      "learning_rate": 2.8474358974358977e-06,
+      "loss": 0.5489,
+      "step": 1780
+    },
+    {
+      "epoch": 0.4934210526315789,
+      "learning_rate": 2.8217948717948718e-06,
+      "loss": 0.5522,
+      "step": 1800
+    },
+    {
+      "epoch": 0.49890350877192985,
+      "learning_rate": 2.7961538461538463e-06,
+      "loss": 0.5463,
+      "step": 1820
+    },
+    {
+      "epoch": 0.5043859649122807,
+      "learning_rate": 2.7705128205128208e-06,
+      "loss": 0.5498,
+      "step": 1840
+    },
+    {
+      "epoch": 0.5098684210526315,
+      "learning_rate": 2.7448717948717953e-06,
+      "loss": 0.5437,
+      "step": 1860
+    },
+    {
+      "epoch": 0.5153508771929824,
+      "learning_rate": 2.7192307692307697e-06,
+      "loss": 0.5477,
+      "step": 1880
+    },
+    {
+      "epoch": 0.5208333333333334,
+      "learning_rate": 2.693589743589744e-06,
+      "loss": 0.5464,
+      "step": 1900
+    },
+    {
+      "epoch": 0.5263157894736842,
+      "learning_rate": 2.667948717948718e-06,
+      "loss": 0.5421,
+      "step": 1920
+    },
+    {
+      "epoch": 0.5317982456140351,
+      "learning_rate": 2.6423076923076924e-06,
+      "loss": 0.5451,
+      "step": 1940
+    },
+    {
+      "epoch": 0.5372807017543859,
+      "learning_rate": 2.616666666666667e-06,
+      "loss": 0.545,
+      "step": 1960
+    },
+    {
+      "epoch": 0.5427631578947368,
+      "learning_rate": 2.5910256410256414e-06,
+      "loss": 0.5416,
+      "step": 1980
+    },
+    {
+      "epoch": 0.5482456140350878,
+      "learning_rate": 2.565384615384616e-06,
+      "loss": 0.5378,
+      "step": 2000
+    },
+    {
+      "epoch": 0.5482456140350878,
+      "eval_loss": 0.5599762797355652,
+      "eval_runtime": 6.5788,
+      "eval_samples_per_second": 14.896,
+      "eval_steps_per_second": 1.976,
+      "step": 2000
+    }
+  ],
+  "logging_steps": 20,
+  "max_steps": 4000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 500,
+  "total_flos": 4034749748215808.0,
+  "train_batch_size": 6,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1666fb8227e6013863ea69147019d309d9fc5bb193cba9a4f7ddb486b04d057
+size 7224

zero_to_fp32.py

@@ -0,0 +1,604 @@
+#!/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 . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+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)
+
+        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 files:
+        state_dict = torch.load(f, map_location=device)
+        # 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}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, 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")
+
+
+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 = fp32_flat_groups[0].numel() * 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
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed 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 get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=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
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, 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_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-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_file,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)