Upload folder using huggingface_hub

README.md

@@ -1,3 +1,202 @@
----
-license: apache-2.0
----
+---
+library_name: peft
+base_model: Viet-Mistral/Vistral-7B-Chat
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.9.0

adapter_config.json

@@ -0,0 +1,33 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "Viet-Mistral/Vistral-7B-Chat",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "gate_proj",
+    "o_proj",
+    "down_proj",
+    "up_proj",
+    "q_proj",
+    "k_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 sha256:9cc759aabc4ad48b6147508f9dc71f69f3a47d23e06769d7ce9134cf494cdc8f
+size 671150064

added_tokens.json

@@ -0,0 +1,6 @@
+{
+  "<</SYS>>": 38366,
+  "<<SYS>>": 38365,
+  "[/INST]": 38368,
+  "[INST]": 38367
+}

config.json

@@ -0,0 +1,41 @@
+{
+  "_name_or_path": "Viet-Mistral/Vistral-7B-Chat",
+  "architectures": [
+    "MistralForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "freeze_mm_mlp_adapter": false,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "image_aspect_ratio": "pad",
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "max_position_embeddings": 32768,
+  "mm_hidden_size": 1024,
+  "mm_patch_merge_type": "flat",
+  "mm_projector_lr": 2e-05,
+  "mm_projector_type": "mlp2x_gelu",
+  "mm_use_im_patch_token": false,
+  "mm_use_im_start_end": false,
+  "mm_vision_select_feature": "patch",
+  "mm_vision_select_layer": -2,
+  "mm_vision_tower": "openai/clip-vit-large-patch14-336",
+  "model_type": "llava_mistral",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "rms_norm_eps": 1e-05,
+  "rope_theta": 10000.0,
+  "sliding_window": 4096,
+  "tie_word_embeddings": false,
+  "tokenizer_model_max_length": 2048,
+  "tokenizer_padding_side": "right",
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.37.2",
+  "tune_mm_mlp_adapter": false,
+  "use_cache": true,
+  "use_mm_proj": true,
+  "vocab_size": 38369
+}

latest

@@ -0,0 +1 @@
+global_step728

non_lora_trainables.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5dd53ae020db19f5500f390b8c67e9abf73516cce4f9b7a1a953a2f6a05b9e82
+size 41961648

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69988968d8d5a12db71ef63a90cd57a93d72db7d8e71339dca2ba33b818726e1
+size 15984

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:819889969baf1f3e358df118f7ad14a747265a088892f67796e19cb3f7dcca7b
+size 15984

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c7497ebcd580bdeffdafbe7b595d7a48a070045672d6d51b50a350e9550172f
+size 15984

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f59dada543b7e4e3b0360181c41ca1f7efa525be268e26ff9589995e1c92a097
+size 15984

rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e113c3f62a3c772780f19cffb4b3ddcff1b11c4fb1ec454620a5dc85daae0e8a
+size 15984

rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69393d0b8187bbc613fbb46a570986eaa4465b9e38d93157e80b89ea42fc6575
+size 15984

rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d66e118e69864543fbe9a5664a456902dbdddcc359771cd4ed80d254c8294b12
+size 15984

rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:faf8236fd4ce6a5aedd45ef71ee41022ec4b1be6f9f2582fa4c459f4592b4b8b
+size 15984

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92c701ceec84f359a0e0880888aead6bdb12fb5bee91ba73be6ba08e36bf3858
+size 1064

special_tokens_map.json

@@ -0,0 +1,29 @@
+{
+  "additional_special_tokens": [
+    "<unk>",
+    "<s>",
+    "</s>"
+  ],
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<unk>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e792a804bbfc19a96b61b87109b8f2b0b7c92830025f285b402ba27c0c309c6f
+size 596883

tokenizer_config.json

@@ -0,0 +1,81 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "38365": {
+      "content": "<<SYS>>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "38366": {
+      "content": "<</SYS>>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "38367": {
+      "content": "[INST]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "38368": {
+      "content": "[/INST]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<unk>",
+    "<s>",
+    "</s>"
+  ],
+  "bos_token": "<s>",
+  "chat_template": "{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% set system_message = false %}{% endif %}{% for message in loop_messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if loop.index0 == 0 and system_message != false %}{% set content = '<<SYS>>\\n' + system_message + '\\n<</SYS>>\\n\\n' + message['content'] %}{% else %}{% set content = message['content'] %}{% endif %}{% if message['role'] == 'user' %}{{ bos_token + '[INST] ' + content.strip() + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ ' '  + content.strip() + ' ' + eos_token }}{% endif %}{% endfor %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 2048,
+  "pad_token": "<unk>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false,
+  "use_fast": true
+}

trainer_state.json

@@ -0,0 +1,1473 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9989708404802744,
+  "eval_steps": 500,
+  "global_step": 728,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.090909090909091e-07,
+      "loss": 1.4022,
+      "step": 3
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 1.8181818181818183e-06,
+      "loss": 1.4239,
+      "step": 6
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.7272727272727272e-06,
+      "loss": 1.3843,
+      "step": 9
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 3.6363636363636366e-06,
+      "loss": 1.3722,
+      "step": 12
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 4.5454545454545455e-06,
+      "loss": 1.3411,
+      "step": 15
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 5.4545454545454545e-06,
+      "loss": 1.3187,
+      "step": 18
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 6.363636363636364e-06,
+      "loss": 1.284,
+      "step": 21
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 7.272727272727273e-06,
+      "loss": 1.2492,
+      "step": 24
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 8.181818181818183e-06,
+      "loss": 1.2658,
+      "step": 27
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 9.090909090909091e-06,
+      "loss": 1.2173,
+      "step": 30
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1e-05,
+      "loss": 1.2302,
+      "step": 33
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1.0909090909090909e-05,
+      "loss": 1.2301,
+      "step": 36
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1.181818181818182e-05,
+      "loss": 1.1855,
+      "step": 39
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 1.2727272727272728e-05,
+      "loss": 1.2094,
+      "step": 42
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 1.3636363636363637e-05,
+      "loss": 1.1788,
+      "step": 45
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1.4545454545454546e-05,
+      "loss": 1.1804,
+      "step": 48
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1.5454545454545454e-05,
+      "loss": 1.166,
+      "step": 51
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1.6363636363636366e-05,
+      "loss": 1.1256,
+      "step": 54
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 1.7272727272727274e-05,
+      "loss": 1.1289,
+      "step": 57
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 1.8181818181818182e-05,
+      "loss": 1.1392,
+      "step": 60
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 1.9090909090909094e-05,
+      "loss": 1.131,
+      "step": 63
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 2e-05,
+      "loss": 1.1288,
+      "step": 66
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 1.9999900994429424e-05,
+      "loss": 1.1198,
+      "step": 69
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1.999960397967811e-05,
+      "loss": 1.1281,
+      "step": 72
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1.9999108961627284e-05,
+      "loss": 1.134,
+      "step": 75
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 1.9998415950078858e-05,
+      "loss": 1.1148,
+      "step": 78
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 1.9997524958755226e-05,
+      "loss": 1.1162,
+      "step": 81
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.9996436005299013e-05,
+      "loss": 1.12,
+      "step": 84
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.999514911127271e-05,
+      "loss": 1.12,
+      "step": 87
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.9993664302158255e-05,
+      "loss": 1.0938,
+      "step": 90
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1.9991981607356517e-05,
+      "loss": 1.0838,
+      "step": 93
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1.9990101060186732e-05,
+      "loss": 1.1078,
+      "step": 96
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1.998802269788583e-05,
+      "loss": 1.1037,
+      "step": 99
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1.9985746561607696e-05,
+      "loss": 1.0804,
+      "step": 102
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1.998327269642237e-05,
+      "loss": 1.0977,
+      "step": 105
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 1.998060115131513e-05,
+      "loss": 1.1036,
+      "step": 108
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 1.9977731979185556e-05,
+      "loss": 1.1109,
+      "step": 111
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.9974665236846443e-05,
+      "loss": 1.0937,
+      "step": 114
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.9971400985022712e-05,
+      "loss": 1.0834,
+      "step": 117
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.9967939288350184e-05,
+      "loss": 1.1002,
+      "step": 120
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1.9964280215374312e-05,
+      "loss": 1.0847,
+      "step": 123
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1.9960423838548814e-05,
+      "loss": 1.0845,
+      "step": 126
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 1.995637023423425e-05,
+      "loss": 1.0984,
+      "step": 129
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 1.9952119482696504e-05,
+      "loss": 1.0836,
+      "step": 132
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1.9947671668105185e-05,
+      "loss": 1.082,
+      "step": 135
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1.9943026878531985e-05,
+      "loss": 1.0707,
+      "step": 138
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1.9938185205948906e-05,
+      "loss": 1.0545,
+      "step": 141
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 1.993314674622646e-05,
+      "loss": 1.0618,
+      "step": 144
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 1.992791159913177e-05,
+      "loss": 1.0514,
+      "step": 147
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1.992247986832658e-05,
+      "loss": 1.0733,
+      "step": 150
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1.99168516613652e-05,
+      "loss": 1.0712,
+      "step": 153
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1.991102708969241e-05,
+      "loss": 1.0788,
+      "step": 156
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 1.9905006268641212e-05,
+      "loss": 1.0744,
+      "step": 159
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 1.9898789317430577e-05,
+      "loss": 1.0621,
+      "step": 162
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.9892376359163058e-05,
+      "loss": 1.0598,
+      "step": 165
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.9885767520822377e-05,
+      "loss": 1.095,
+      "step": 168
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.9878962933270896e-05,
+      "loss": 1.0666,
+      "step": 171
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1.987196273124703e-05,
+      "loss": 1.0657,
+      "step": 174
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1.986476705336258e-05,
+      "loss": 1.0691,
+      "step": 177
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 1.9857376042099982e-05,
+      "loss": 1.0663,
+      "step": 180
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 1.9849789843809496e-05,
+      "loss": 1.0476,
+      "step": 183
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.9842008608706295e-05,
+      "loss": 1.0509,
+      "step": 186
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.983403249086751e-05,
+      "loss": 1.0622,
+      "step": 189
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.9825861648229154e-05,
+      "loss": 1.0708,
+      "step": 192
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 1.981749624258302e-05,
+      "loss": 1.0672,
+      "step": 195
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 1.9808936439573455e-05,
+      "loss": 1.0627,
+      "step": 198
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1.9800182408694096e-05,
+      "loss": 1.0726,
+      "step": 201
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1.9791234323284515e-05,
+      "loss": 1.0558,
+      "step": 204
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1.9782092360526763e-05,
+      "loss": 1.0677,
+      "step": 207
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1.977275670144189e-05,
+      "loss": 1.0422,
+      "step": 210
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1.9763227530886348e-05,
+      "loss": 1.0364,
+      "step": 213
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1.9753505037548334e-05,
+      "loss": 1.0475,
+      "step": 216
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1.974358941394404e-05,
+      "loss": 1.0508,
+      "step": 219
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1.973348085641387e-05,
+      "loss": 1.0595,
+      "step": 222
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1.972317956511852e-05,
+      "loss": 1.0528,
+      "step": 225
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1.971268574403503e-05,
+      "loss": 1.0562,
+      "step": 228
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 1.970199960095276e-05,
+      "loss": 1.0329,
+      "step": 231
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 1.9691121347469235e-05,
+      "loss": 1.045,
+      "step": 234
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1.9680051198986004e-05,
+      "loss": 1.0561,
+      "step": 237
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1.9668789374704337e-05,
+      "loss": 1.0449,
+      "step": 240
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1.9657336097620904e-05,
+      "loss": 1.0359,
+      "step": 243
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1.964569159452335e-05,
+      "loss": 1.0359,
+      "step": 246
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1.963385609598581e-05,
+      "loss": 1.0271,
+      "step": 249
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1.9621829836364335e-05,
+      "loss": 1.0563,
+      "step": 252
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1.9609613053792276e-05,
+      "loss": 1.0416,
+      "step": 255
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1.9597205990175528e-05,
+      "loss": 1.0578,
+      "step": 258
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1.958460889118778e-05,
+      "loss": 1.0461,
+      "step": 261
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1.9571822006265623e-05,
+      "loss": 1.0262,
+      "step": 264
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 1.9558845588603625e-05,
+      "loss": 1.0254,
+      "step": 267
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 1.9545679895149315e-05,
+      "loss": 1.0642,
+      "step": 270
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 1.9532325186598093e-05,
+      "loss": 1.0456,
+      "step": 273
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.951878172738806e-05,
+      "loss": 1.0358,
+      "step": 276
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.9505049785694803e-05,
+      "loss": 1.0409,
+      "step": 279
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 1.9491129633426068e-05,
+      "loss": 1.0382,
+      "step": 282
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 1.9477021546216376e-05,
+      "loss": 1.0415,
+      "step": 285
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1.9462725803421566e-05,
+      "loss": 1.0308,
+      "step": 288
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1.9448242688113286e-05,
+      "loss": 1.0376,
+      "step": 291
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1.9433572487073343e-05,
+      "loss": 1.0259,
+      "step": 294
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1.9418715490788066e-05,
+      "loss": 1.0496,
+      "step": 297
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1.9403671993442534e-05,
+      "loss": 1.0519,
+      "step": 300
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1.9388442292914754e-05,
+      "loss": 1.0418,
+      "step": 303
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1.937302669076976e-05,
+      "loss": 1.0372,
+      "step": 306
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1.9357425492253662e-05,
+      "loss": 1.0347,
+      "step": 309
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1.934163900628756e-05,
+      "loss": 1.0253,
+      "step": 312
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1.9325667545461466e-05,
+      "loss": 1.0401,
+      "step": 315
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 1.9309511426028105e-05,
+      "loss": 1.0282,
+      "step": 318
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 1.9293170967896632e-05,
+      "loss": 1.0306,
+      "step": 321
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 1.9276646494626333e-05,
+      "loss": 1.0313,
+      "step": 324
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1.9259938333420183e-05,
+      "loss": 1.0433,
+      "step": 327
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1.9243046815118387e-05,
+      "loss": 1.0232,
+      "step": 330
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 1.922597227419183e-05,
+      "loss": 1.0222,
+      "step": 333
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 1.9208715048735446e-05,
+      "loss": 1.0186,
+      "step": 336
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1.9191275480461525e-05,
+      "loss": 1.033,
+      "step": 339
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1.9173653914692947e-05,
+      "loss": 1.0342,
+      "step": 342
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1.9155850700356345e-05,
+      "loss": 1.035,
+      "step": 345
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 1.91378661899752e-05,
+      "loss": 1.0206,
+      "step": 348
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 1.9119700739662857e-05,
+      "loss": 1.0435,
+      "step": 351
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1.910135470911547e-05,
+      "loss": 1.0181,
+      "step": 354
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1.908282846160488e-05,
+      "loss": 1.0267,
+      "step": 357
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1.9064122363971426e-05,
+      "loss": 1.0365,
+      "step": 360
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.904523678661669e-05,
+      "loss": 1.0381,
+      "step": 363
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.9026172103496138e-05,
+      "loss": 1.0048,
+      "step": 366
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.900692869211174e-05,
+      "loss": 1.0392,
+      "step": 369
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.898750693350447e-05,
+      "loss": 1.0278,
+      "step": 372
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1.8967907212246803e-05,
+      "loss": 1.013,
+      "step": 375
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.8948129916435048e-05,
+      "loss": 1.0385,
+      "step": 378
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1.8928175437681698e-05,
+      "loss": 1.0168,
+      "step": 381
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8908044171107658e-05,
+      "loss": 1.0123,
+      "step": 384
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1.8887736515334443e-05,
+      "loss": 1.015,
+      "step": 387
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8867252872476255e-05,
+      "loss": 1.0265,
+      "step": 390
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.884659364813205e-05,
+      "loss": 0.9997,
+      "step": 393
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1.8825759251377484e-05,
+      "loss": 1.0109,
+      "step": 396
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8804750094756827e-05,
+      "loss": 1.0199,
+      "step": 399
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1.8783566594274783e-05,
+      "loss": 0.9998,
+      "step": 402
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.8762209169388262e-05,
+      "loss": 1.0088,
+      "step": 405
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.8740678242998077e-05,
+      "loss": 1.0022,
+      "step": 408
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1.8718974241440552e-05,
+      "loss": 1.0216,
+      "step": 411
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.8697097594479103e-05,
+      "loss": 1.0248,
+      "step": 414
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1.867504873529571e-05,
+      "loss": 0.9974,
+      "step": 417
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.865282810048235e-05,
+      "loss": 1.0138,
+      "step": 420
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.8630436130032353e-05,
+      "loss": 1.0004,
+      "step": 423
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1.860787326733168e-05,
+      "loss": 1.0081,
+      "step": 426
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.8585139959150144e-05,
+      "loss": 1.0238,
+      "step": 429
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1.856223665563258e-05,
+      "loss": 1.0328,
+      "step": 432
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.8539163810289914e-05,
+      "loss": 1.0071,
+      "step": 435
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1.8515921879990187e-05,
+      "loss": 1.0134,
+      "step": 438
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.8492511324949516e-05,
+      "loss": 1.0181,
+      "step": 441
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.8468932608722975e-05,
+      "loss": 1.0363,
+      "step": 444
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1.8445186198195406e-05,
+      "loss": 1.0011,
+      "step": 447
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.8421272563572202e-05,
+      "loss": 0.9993,
+      "step": 450
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1.8397192178369965e-05,
+      "loss": 1.0201,
+      "step": 453
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.837294551940716e-05,
+      "loss": 0.987,
+      "step": 456
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.834853306679464e-05,
+      "loss": 1.0106,
+      "step": 459
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1.8323955303926165e-05,
+      "loss": 1.0034,
+      "step": 462
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.8299212717468825e-05,
+      "loss": 1.0095,
+      "step": 465
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.8274305797353397e-05,
+      "loss": 0.9921,
+      "step": 468
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.824923503676465e-05,
+      "loss": 0.9859,
+      "step": 471
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.822400093213157e-05,
+      "loss": 1.017,
+      "step": 474
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1.8198603983117546e-05,
+      "loss": 1.0118,
+      "step": 477
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.8173044692610466e-05,
+      "loss": 0.9912,
+      "step": 480
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1.8147323566712755e-05,
+      "loss": 1.0162,
+      "step": 483
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.8121441114731366e-05,
+      "loss": 1.0089,
+      "step": 486
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1.809539784916768e-05,
+      "loss": 0.9752,
+      "step": 489
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.806919428570737e-05,
+      "loss": 1.007,
+      "step": 492
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.804283094321019e-05,
+      "loss": 1.0145,
+      "step": 495
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1.8016308343699686e-05,
+      "loss": 1.0008,
+      "step": 498
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.798962701235289e-05,
+      "loss": 1.0067,
+      "step": 501
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1.796278747748988e-05,
+      "loss": 1.0017,
+      "step": 504
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.7935790270563345e-05,
+      "loss": 1.0086,
+      "step": 507
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.790863592614807e-05,
+      "loss": 0.9884,
+      "step": 510
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.788132498193032e-05,
+      "loss": 1.0028,
+      "step": 513
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.7853857978697223e-05,
+      "loss": 1.0055,
+      "step": 516
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1.7826235460326043e-05,
+      "loss": 1.005,
+      "step": 519
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.7798457973773418e-05,
+      "loss": 1.002,
+      "step": 522
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.7770526069064525e-05,
+      "loss": 0.9838,
+      "step": 525
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1.7742440299282203e-05,
+      "loss": 1.001,
+      "step": 528
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.7714201220555982e-05,
+      "loss": 0.9984,
+      "step": 531
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1.7685809392051084e-05,
+      "loss": 1.0035,
+      "step": 534
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.765726537595734e-05,
+      "loss": 1.0076,
+      "step": 537
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1.7628569737478076e-05,
+      "loss": 0.9936,
+      "step": 540
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.7599723044818898e-05,
+      "loss": 1.0053,
+      "step": 543
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.7570725869176468e-05,
+      "loss": 0.9968,
+      "step": 546
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1.7541578784727163e-05,
+      "loss": 1.0059,
+      "step": 549
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 1.751228236861573e-05,
+      "loss": 1.0059,
+      "step": 552
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 1.7482837200943845e-05,
+      "loss": 1.0081,
+      "step": 555
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1.7453243864758638e-05,
+      "loss": 1.0215,
+      "step": 558
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1.7423502946041133e-05,
+      "loss": 0.9935,
+      "step": 561
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1.739361503369466e-05,
+      "loss": 0.9945,
+      "step": 564
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1.7363580719533173e-05,
+      "loss": 0.9926,
+      "step": 567
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1.733340059826956e-05,
+      "loss": 0.9946,
+      "step": 570
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 1.7303075267503845e-05,
+      "loss": 1.0079,
+      "step": 573
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 1.7272605327711364e-05,
+      "loss": 1.0212,
+      "step": 576
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 1.7241991382230872e-05,
+      "loss": 0.993,
+      "step": 579
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1.72112340372526e-05,
+      "loss": 0.9843,
+      "step": 582
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1.718033390180624e-05,
+      "loss": 0.9837,
+      "step": 585
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 1.71492915877489e-05,
+      "loss": 0.959,
+      "step": 588
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 1.7118107709752986e-05,
+      "loss": 0.9895,
+      "step": 591
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1.7086782885294026e-05,
+      "loss": 0.99,
+      "step": 594
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1.7055317734638444e-05,
+      "loss": 1.006,
+      "step": 597
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1.702371288083127e-05,
+      "loss": 1.0009,
+      "step": 600
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 1.6991968949683835e-05,
+      "loss": 0.9758,
+      "step": 603
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 1.6960086569761332e-05,
+      "loss": 0.9801,
+      "step": 606
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1.6928066372370407e-05,
+      "loss": 0.9833,
+      "step": 609
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1.689590899154664e-05,
+      "loss": 0.9846,
+      "step": 612
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1.6863615064042003e-05,
+      "loss": 0.9752,
+      "step": 615
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1.6831185229312237e-05,
+      "loss": 0.9869,
+      "step": 618
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1.67986201295042e-05,
+      "loss": 0.9869,
+      "step": 621
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1.676592040944315e-05,
+      "loss": 0.9878,
+      "step": 624
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1.6733086716619976e-05,
+      "loss": 0.9938,
+      "step": 627
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1.6700119701178378e-05,
+      "loss": 1.0045,
+      "step": 630
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 1.666702001590199e-05,
+      "loss": 1.0088,
+      "step": 633
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 1.6633788316201455e-05,
+      "loss": 0.998,
+      "step": 636
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.6600425260101453e-05,
+      "loss": 1.0017,
+      "step": 639
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.6566931508227663e-05,
+      "loss": 0.9995,
+      "step": 642
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.6533307723793688e-05,
+      "loss": 1.0012,
+      "step": 645
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.649955457258792e-05,
+      "loss": 0.9807,
+      "step": 648
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1.6465672722960365e-05,
+      "loss": 0.9664,
+      "step": 651
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.6431662845809388e-05,
+      "loss": 0.9707,
+      "step": 654
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.6397525614568446e-05,
+      "loss": 0.983,
+      "step": 657
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.6363261705192757e-05,
+      "loss": 1.0061,
+      "step": 660
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.6328871796145894e-05,
+      "loss": 0.9899,
+      "step": 663
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1.629435656838637e-05,
+      "loss": 0.9795,
+      "step": 666
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 1.6259716705354154e-05,
+      "loss": 1.0002,
+      "step": 669
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 1.6224952892957122e-05,
+      "loss": 0.9837,
+      "step": 672
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 1.6190065819557496e-05,
+      "loss": 0.9872,
+      "step": 675
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 1.615505617595819e-05,
+      "loss": 0.9797,
+      "step": 678
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 1.6119924655389158e-05,
+      "loss": 0.9926,
+      "step": 681
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 1.6084671953493645e-05,
+      "loss": 0.9884,
+      "step": 684
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 1.6049298768314425e-05,
+      "loss": 0.9918,
+      "step": 687
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 1.6013805800279977e-05,
+      "loss": 0.9829,
+      "step": 690
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 1.5978193752190607e-05,
+      "loss": 0.9854,
+      "step": 693
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.5942463329204546e-05,
+      "loss": 0.9751,
+      "step": 696
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.5906615238823974e-05,
+      "loss": 0.9945,
+      "step": 699
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1.5870650190881023e-05,
+      "loss": 0.9957,
+      "step": 702
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.583456889752371e-05,
+      "loss": 1.0047,
+      "step": 705
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1.579837207320184e-05,
+      "loss": 0.9921,
+      "step": 708
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.5762060434652863e-05,
+      "loss": 0.9839,
+      "step": 711
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.572563470088768e-05,
+      "loss": 0.9922,
+      "step": 714
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1.56890955931764e-05,
+      "loss": 0.9752,
+      "step": 717
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1.565244383503407e-05,
+      "loss": 0.9778,
+      "step": 720
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1.5615680152206324e-05,
+      "loss": 0.9795,
+      "step": 723
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1.557880527265505e-05,
+      "loss": 0.9774,
+      "step": 726
+    }
+  ],
+  "logging_steps": 3,
+  "max_steps": 2184,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500.0,
+  "total_flos": 4.694048596218085e+18,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d50359a332e42ea966d472131e23398b23d29c578499d527e2012fe207eae898
+size 6264

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)