Upload folder using huggingface_hub

README.md

@@ -0,0 +1,202 @@
+---
+base_model: /root/workspace/digirl/outputs/minicpmv_pt1000
+library_name: peft
+---
+
+# 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 0.12.0

adapter_config.json

@@ -0,0 +1,32 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": {
+    "base_model_class": "MiniCPMV",
+    "parent_library": "transformers_modules.openbmb.MiniCPM-V-2_6.4e4be000cd81feda8b96d14b53f1791b4010b038.modeling_minicpmv"
+  },
+  "base_model_name_or_path": "/root/workspace/digirl/outputs/minicpmv_pt1000",
+  "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": 64,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": [
+    "embed_tokens",
+    "resampler"
+  ],
+  "peft_type": "LORA",
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": "llm\\..*layers\\.\\d+\\.self_attn\\.(q_proj|k_proj|v_proj|o_proj)",
+  "task_type": null,
+  "use_dora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e113c2b54a97fff8dd57be79c22119b5a1e44265b5396009e0bc8ef4dd173ca
+size 1305154312

added_tokens.json

@@ -0,0 +1,25 @@
+{
+  "</box>": 151651,
+  "</image>": 151647,
+  "</image_id>": 151659,
+  "</point>": 151655,
+  "</quad>": 151653,
+  "</ref>": 151649,
+  "</slice>": 151657,
+  "<box>": 151650,
+  "<image>": 151646,
+  "<image_id>": 151658,
+  "<point>": 151654,
+  "<quad>": 151652,
+  "<ref>": 151648,
+  "<slice>": 151656,
+  "<|endoftext|>": 151643,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|reserved_special_token_0|>": 151660,
+  "<|reserved_special_token_1|>": 151661,
+  "<|reserved_special_token_2|>": 151662,
+  "<|reserved_special_token_3|>": 151663,
+  "<|reserved_special_token_4|>": 151664,
+  "<|reserved_special_token_5|>": 151665
+}

latest

@@ -0,0 +1 @@
+global_step250

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed6a2d05cc7a06addf1550e979e589da3472cd6b1edac74354584fabab31282d
+size 15024

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de348fc647233fdc2fcb9e4379bedc7385ae9066915b91dfad1264742c86998e
+size 15024

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4959a127b60a7d53e124b5e14292bc21cfb045313c23cb2516dbfdd0fdedca96
+size 15024

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f27b59b3e263669823622cbe27cd950550735cfb6e20d4b32e64c97c3fdd785a
+size 15024

special_tokens_map.json

@@ -0,0 +1,52 @@
+{
+  "additional_special_tokens": [
+    "<image>",
+    "</image>",
+    "<ref>",
+    "</ref>",
+    "<box>",
+    "</box>",
+    "<quad>",
+    "</quad>",
+    "<point>",
+    "</point>",
+    "<slice>",
+    "</slice>",
+    "<image_id>",
+    "</image_id>",
+    "<|reserved_special_token_0|>",
+    "<|reserved_special_token_1|>",
+    "<|reserved_special_token_2|>",
+    "<|reserved_special_token_3|>",
+    "<|reserved_special_token_4|>",
+    "<|reserved_special_token_5|>"
+  ],
+  "bos_token": {
+    "content": "<|im_start|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenization_minicpmv_fast.py

@@ -0,0 +1,66 @@
+from transformers.models.qwen2 import Qwen2TokenizerFast
+
+
+class MiniCPMVTokenizerFast(Qwen2TokenizerFast):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self.im_start = "<image>"
+        self.im_end = "</image>"
+        self.ref_start = "<ref>"
+        self.ref_end = "</ref>"
+        self.box_start = "<box>"
+        self.box_end = "</box>"
+        self.quad_start = "<quad>"
+        self.quad_end = "</quad>"
+        self.slice_start = "<slice>"
+        self.slice_end = "</slice>"
+        self.im_id_start = "<image_id>"
+        self.im_id_end = "</image_id>"
+
+    @property
+    def eos_id(self):
+        return self.eos_token_id
+
+    @property
+    def bos_id(self):
+        return self.bos_token_id
+
+    @property
+    def unk_id(self):
+        return self.unk_token_id
+
+    @property
+    def im_start_id(self):
+        return self.convert_tokens_to_ids(self.im_start)
+
+    @property
+    def im_end_id(self):
+        return self.convert_tokens_to_ids(self.im_end)
+
+    @property
+    def slice_start_id(self):
+        return self.convert_tokens_to_ids(self.slice_start)
+    
+    @property
+    def slice_end_id(self):
+        return self.convert_tokens_to_ids(self.slice_end)
+
+    @property
+    def im_id_start_id(self):
+        return self.convert_tokens_to_ids(self.im_id_start)
+    
+    @property
+    def im_id_end_id(self):
+        return self.convert_tokens_to_ids(self.im_id_end)
+    
+    @property
+    def newline_id(self):
+        return self.convert_tokens_to_ids('\n')
+
+    @staticmethod
+    def escape(text: str) -> str:
+        return text
+
+    @staticmethod
+    def unescape(text: str) -> str:
+        return text

tokenizer_config.json

@@ -0,0 +1,235 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "128244": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<image>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "</image>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<ref>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "</ref>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<box>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "</box>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<quad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "</quad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<point>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "</point>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<slice>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "</slice>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151658": {
+      "content": "<image_id>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151659": {
+      "content": "</image_id>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151660": {
+      "content": "<|reserved_special_token_0|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151661": {
+      "content": "<|reserved_special_token_1|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151662": {
+      "content": "<|reserved_special_token_2|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151663": {
+      "content": "<|reserved_special_token_3|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151664": {
+      "content": "<|reserved_special_token_4|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151665": {
+      "content": "<|reserved_special_token_5|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<image>",
+    "</image>",
+    "<ref>",
+    "</ref>",
+    "<box>",
+    "</box>",
+    "<quad>",
+    "</quad>",
+    "<point>",
+    "</point>",
+    "<slice>",
+    "</slice>",
+    "<image_id>",
+    "</image_id>",
+    "<|reserved_special_token_0|>",
+    "<|reserved_special_token_1|>",
+    "<|reserved_special_token_2|>",
+    "<|reserved_special_token_3|>",
+    "<|reserved_special_token_4|>",
+    "<|reserved_special_token_5|>"
+  ],
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_qwen2.Qwen2Tokenizer",
+      "tokenization_minicpmv_fast.MiniCPMVTokenizerFast"
+    ]
+  },
+  "bos_token": "<|im_start|>",
+  "chat_template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "MiniCPMVTokenizer",
+  "unk_token": "<unk>"
+}

trainer_state.json

@@ -0,0 +1,1783 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 125.0,
+  "eval_steps": 1000,
+  "global_step": 250,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.5,
+      "grad_norm": 3.334595203399658,
+      "learning_rate": 0.0,
+      "loss": 1.3402,
+      "step": 1
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 3.309346914291382,
+      "learning_rate": 1.5051499783199055e-07,
+      "loss": 1.284,
+      "step": 2
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 3.4695301055908203,
+      "learning_rate": 2.385606273598312e-07,
+      "loss": 1.3362,
+      "step": 3
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 2.9988322257995605,
+      "learning_rate": 3.010299956639811e-07,
+      "loss": 1.2535,
+      "step": 4
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 3.0436394214630127,
+      "learning_rate": 3.494850021680093e-07,
+      "loss": 1.2326,
+      "step": 5
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 3.329270601272583,
+      "learning_rate": 3.8907562519182173e-07,
+      "loss": 1.3378,
+      "step": 6
+    },
+    {
+      "epoch": 3.5,
+      "grad_norm": 3.0574333667755127,
+      "learning_rate": 4.2254902000712834e-07,
+      "loss": 1.4162,
+      "step": 7
+    },
+    {
+      "epoch": 4.0,
+      "grad_norm": 3.348349094390869,
+      "learning_rate": 4.5154499349597166e-07,
+      "loss": 1.3598,
+      "step": 8
+    },
+    {
+      "epoch": 4.5,
+      "grad_norm": 2.9368350505828857,
+      "learning_rate": 4.771212547196623e-07,
+      "loss": 1.1551,
+      "step": 9
+    },
+    {
+      "epoch": 5.0,
+      "grad_norm": 3.3986520767211914,
+      "learning_rate": 4.999999999999999e-07,
+      "loss": 1.4285,
+      "step": 10
+    },
+    {
+      "epoch": 5.5,
+      "grad_norm": 3.004727363586426,
+      "learning_rate": 5.206963425791124e-07,
+      "loss": 1.2629,
+      "step": 11
+    },
+    {
+      "epoch": 6.0,
+      "grad_norm": 3.090939998626709,
+      "learning_rate": 5.395906230238123e-07,
+      "loss": 1.3378,
+      "step": 12
+    },
+    {
+      "epoch": 6.5,
+      "grad_norm": 2.906198740005493,
+      "learning_rate": 5.569716761534182e-07,
+      "loss": 1.3384,
+      "step": 13
+    },
+    {
+      "epoch": 7.0,
+      "grad_norm": 2.9036598205566406,
+      "learning_rate": 5.730640178391189e-07,
+      "loss": 1.2693,
+      "step": 14
+    },
+    {
+      "epoch": 7.5,
+      "grad_norm": 2.924349069595337,
+      "learning_rate": 5.880456295278405e-07,
+      "loss": 1.2711,
+      "step": 15
+    },
+    {
+      "epoch": 8.0,
+      "grad_norm": 3.110586166381836,
+      "learning_rate": 6.020599913279622e-07,
+      "loss": 1.2708,
+      "step": 16
+    },
+    {
+      "epoch": 8.5,
+      "grad_norm": 3.2055583000183105,
+      "learning_rate": 6.15224460689137e-07,
+      "loss": 1.4757,
+      "step": 17
+    },
+    {
+      "epoch": 9.0,
+      "grad_norm": 2.972228527069092,
+      "learning_rate": 6.276362525516529e-07,
+      "loss": 1.3115,
+      "step": 18
+    },
+    {
+      "epoch": 9.5,
+      "grad_norm": 2.447122573852539,
+      "learning_rate": 6.393768004764143e-07,
+      "loss": 1.241,
+      "step": 19
+    },
+    {
+      "epoch": 10.0,
+      "grad_norm": 2.3147361278533936,
+      "learning_rate": 6.505149978319905e-07,
+      "loss": 1.2135,
+      "step": 20
+    },
+    {
+      "epoch": 10.5,
+      "grad_norm": 2.9482200145721436,
+      "learning_rate": 6.611096473669595e-07,
+      "loss": 1.4155,
+      "step": 21
+    },
+    {
+      "epoch": 11.0,
+      "grad_norm": 2.6261355876922607,
+      "learning_rate": 6.712113404111031e-07,
+      "loss": 1.2434,
+      "step": 22
+    },
+    {
+      "epoch": 11.5,
+      "grad_norm": 2.448061943054199,
+      "learning_rate": 6.808639180087963e-07,
+      "loss": 1.2424,
+      "step": 23
+    },
+    {
+      "epoch": 12.0,
+      "grad_norm": 2.22385311126709,
+      "learning_rate": 6.901056208558029e-07,
+      "loss": 1.1388,
+      "step": 24
+    },
+    {
+      "epoch": 12.5,
+      "grad_norm": 2.7259082794189453,
+      "learning_rate": 6.989700043360186e-07,
+      "loss": 1.4238,
+      "step": 25
+    },
+    {
+      "epoch": 13.0,
+      "grad_norm": 2.03949236869812,
+      "learning_rate": 7.074866739854088e-07,
+      "loss": 1.0655,
+      "step": 26
+    },
+    {
+      "epoch": 13.5,
+      "grad_norm": 2.341221809387207,
+      "learning_rate": 7.156818820794935e-07,
+      "loss": 1.2652,
+      "step": 27
+    },
+    {
+      "epoch": 14.0,
+      "grad_norm": 2.2585113048553467,
+      "learning_rate": 7.235790156711094e-07,
+      "loss": 1.1248,
+      "step": 28
+    },
+    {
+      "epoch": 14.5,
+      "grad_norm": 2.2248682975769043,
+      "learning_rate": 7.311989989494779e-07,
+      "loss": 1.1755,
+      "step": 29
+    },
+    {
+      "epoch": 15.0,
+      "grad_norm": 2.115250825881958,
+      "learning_rate": 7.38560627359831e-07,
+      "loss": 1.1932,
+      "step": 30
+    },
+    {
+      "epoch": 15.5,
+      "grad_norm": 2.030606269836426,
+      "learning_rate": 7.456808469171361e-07,
+      "loss": 1.2845,
+      "step": 31
+    },
+    {
+      "epoch": 16.0,
+      "grad_norm": 2.0579230785369873,
+      "learning_rate": 7.525749891599529e-07,
+      "loss": 1.1964,
+      "step": 32
+    },
+    {
+      "epoch": 16.5,
+      "grad_norm": 2.074784517288208,
+      "learning_rate": 7.592569699389436e-07,
+      "loss": 1.1445,
+      "step": 33
+    },
+    {
+      "epoch": 17.0,
+      "grad_norm": 2.064431667327881,
+      "learning_rate": 7.657394585211274e-07,
+      "loss": 1.2398,
+      "step": 34
+    },
+    {
+      "epoch": 17.5,
+      "grad_norm": 1.8380582332611084,
+      "learning_rate": 7.720340221751376e-07,
+      "loss": 1.1373,
+      "step": 35
+    },
+    {
+      "epoch": 18.0,
+      "grad_norm": 2.3023219108581543,
+      "learning_rate": 7.781512503836435e-07,
+      "loss": 1.0978,
+      "step": 36
+    },
+    {
+      "epoch": 18.5,
+      "grad_norm": 2.0860466957092285,
+      "learning_rate": 7.841008620334974e-07,
+      "loss": 1.1017,
+      "step": 37
+    },
+    {
+      "epoch": 19.0,
+      "grad_norm": 1.9423589706420898,
+      "learning_rate": 7.89891798308405e-07,
+      "loss": 1.1538,
+      "step": 38
+    },
+    {
+      "epoch": 19.5,
+      "grad_norm": 2.018376350402832,
+      "learning_rate": 7.955323035132494e-07,
+      "loss": 1.1345,
+      "step": 39
+    },
+    {
+      "epoch": 20.0,
+      "grad_norm": 1.9834052324295044,
+      "learning_rate": 8.01029995663981e-07,
+      "loss": 1.1596,
+      "step": 40
+    },
+    {
+      "epoch": 20.5,
+      "grad_norm": 1.8433727025985718,
+      "learning_rate": 8.063919283598676e-07,
+      "loss": 0.9934,
+      "step": 41
+    },
+    {
+      "epoch": 21.0,
+      "grad_norm": 2.0120983123779297,
+      "learning_rate": 8.116246451989502e-07,
+      "loss": 1.1849,
+      "step": 42
+    },
+    {
+      "epoch": 21.5,
+      "grad_norm": 1.7347400188446045,
+      "learning_rate": 8.16734227789793e-07,
+      "loss": 0.9969,
+      "step": 43
+    },
+    {
+      "epoch": 22.0,
+      "grad_norm": 2.1413559913635254,
+      "learning_rate": 8.217263382430935e-07,
+      "loss": 1.2613,
+      "step": 44
+    },
+    {
+      "epoch": 22.5,
+      "grad_norm": 1.9589574337005615,
+      "learning_rate": 8.266062568876716e-07,
+      "loss": 1.134,
+      "step": 45
+    },
+    {
+      "epoch": 23.0,
+      "grad_norm": 1.7819244861602783,
+      "learning_rate": 8.313789158407869e-07,
+      "loss": 1.0627,
+      "step": 46
+    },
+    {
+      "epoch": 23.5,
+      "grad_norm": 2.12137508392334,
+      "learning_rate": 8.360489289678585e-07,
+      "loss": 1.1799,
+      "step": 47
+    },
+    {
+      "epoch": 24.0,
+      "grad_norm": 1.7399541139602661,
+      "learning_rate": 8.406206186877934e-07,
+      "loss": 0.9974,
+      "step": 48
+    },
+    {
+      "epoch": 24.5,
+      "grad_norm": 2.0634093284606934,
+      "learning_rate": 8.450980400142567e-07,
+      "loss": 0.9758,
+      "step": 49
+    },
+    {
+      "epoch": 25.0,
+      "grad_norm": 1.8642668724060059,
+      "learning_rate": 8.494850021680092e-07,
+      "loss": 1.1103,
+      "step": 50
+    },
+    {
+      "epoch": 25.5,
+      "grad_norm": 1.8973793983459473,
+      "learning_rate": 8.53785088048968e-07,
+      "loss": 1.1013,
+      "step": 51
+    },
+    {
+      "epoch": 26.0,
+      "grad_norm": 1.8340080976486206,
+      "learning_rate": 8.580016718173995e-07,
+      "loss": 0.9959,
+      "step": 52
+    },
+    {
+      "epoch": 26.5,
+      "grad_norm": 1.8867207765579224,
+      "learning_rate": 8.621379348003944e-07,
+      "loss": 1.1526,
+      "step": 53
+    },
+    {
+      "epoch": 27.0,
+      "grad_norm": 1.7188408374786377,
+      "learning_rate": 8.661968799114842e-07,
+      "loss": 0.9184,
+      "step": 54
+    },
+    {
+      "epoch": 27.5,
+      "grad_norm": 1.8246105909347534,
+      "learning_rate": 8.701813447471218e-07,
+      "loss": 1.0568,
+      "step": 55
+    },
+    {
+      "epoch": 28.0,
+      "grad_norm": 1.9179924726486206,
+      "learning_rate": 8.740940135031001e-07,
+      "loss": 1.1564,
+      "step": 56
+    },
+    {
+      "epoch": 28.5,
+      "grad_norm": 1.7202187776565552,
+      "learning_rate": 8.779374278362456e-07,
+      "loss": 0.9801,
+      "step": 57
+    },
+    {
+      "epoch": 29.0,
+      "grad_norm": 1.891327977180481,
+      "learning_rate": 8.817139967814684e-07,
+      "loss": 1.1344,
+      "step": 58
+    },
+    {
+      "epoch": 29.5,
+      "grad_norm": 1.6394891738891602,
+      "learning_rate": 8.854260058210719e-07,
+      "loss": 0.9748,
+      "step": 59
+    },
+    {
+      "epoch": 30.0,
+      "grad_norm": 1.8808232545852661,
+      "learning_rate": 8.890756251918216e-07,
+      "loss": 0.9423,
+      "step": 60
+    },
+    {
+      "epoch": 30.5,
+      "grad_norm": 1.8192319869995117,
+      "learning_rate": 8.926649175053833e-07,
+      "loss": 1.003,
+      "step": 61
+    },
+    {
+      "epoch": 31.0,
+      "grad_norm": 1.7621928453445435,
+      "learning_rate": 8.961958447491268e-07,
+      "loss": 1.0498,
+      "step": 62
+    },
+    {
+      "epoch": 31.5,
+      "grad_norm": 1.8957831859588623,
+      "learning_rate": 8.996702747267907e-07,
+      "loss": 1.1259,
+      "step": 63
+    },
+    {
+      "epoch": 32.0,
+      "grad_norm": 1.6946312189102173,
+      "learning_rate": 9.030899869919433e-07,
+      "loss": 0.9092,
+      "step": 64
+    },
+    {
+      "epoch": 32.5,
+      "grad_norm": 1.6719154119491577,
+      "learning_rate": 9.064566783214276e-07,
+      "loss": 0.9273,
+      "step": 65
+    },
+    {
+      "epoch": 33.0,
+      "grad_norm": 1.6526826620101929,
+      "learning_rate": 9.097719677709341e-07,
+      "loss": 0.9469,
+      "step": 66
+    },
+    {
+      "epoch": 33.5,
+      "grad_norm": 1.7876310348510742,
+      "learning_rate": 9.13037401350413e-07,
+      "loss": 0.8734,
+      "step": 67
+    },
+    {
+      "epoch": 34.0,
+      "grad_norm": 1.517971396446228,
+      "learning_rate": 9.162544563531181e-07,
+      "loss": 0.9527,
+      "step": 68
+    },
+    {
+      "epoch": 34.5,
+      "grad_norm": 1.7323296070098877,
+      "learning_rate": 9.194245453686276e-07,
+      "loss": 0.8279,
+      "step": 69
+    },
+    {
+      "epoch": 35.0,
+      "grad_norm": 1.5928230285644531,
+      "learning_rate": 9.225490200071283e-07,
+      "loss": 0.9627,
+      "step": 70
+    },
+    {
+      "epoch": 35.5,
+      "grad_norm": 1.6561415195465088,
+      "learning_rate": 9.256291743595375e-07,
+      "loss": 0.9138,
+      "step": 71
+    },
+    {
+      "epoch": 36.0,
+      "grad_norm": 1.628674030303955,
+      "learning_rate": 9.28666248215634e-07,
+      "loss": 0.8956,
+      "step": 72
+    },
+    {
+      "epoch": 36.5,
+      "grad_norm": 1.5087110996246338,
+      "learning_rate": 9.316614300602277e-07,
+      "loss": 0.8223,
+      "step": 73
+    },
+    {
+      "epoch": 37.0,
+      "grad_norm": 1.6503626108169556,
+      "learning_rate": 9.346158598654879e-07,
+      "loss": 0.9776,
+      "step": 74
+    },
+    {
+      "epoch": 37.5,
+      "grad_norm": 1.6680738925933838,
+      "learning_rate": 9.375306316958498e-07,
+      "loss": 0.8301,
+      "step": 75
+    },
+    {
+      "epoch": 38.0,
+      "grad_norm": 1.8239458799362183,
+      "learning_rate": 9.404067961403955e-07,
+      "loss": 0.9891,
+      "step": 76
+    },
+    {
+      "epoch": 38.5,
+      "grad_norm": 1.669043779373169,
+      "learning_rate": 9.432453625862408e-07,
+      "loss": 0.9085,
+      "step": 77
+    },
+    {
+      "epoch": 39.0,
+      "grad_norm": 1.8124521970748901,
+      "learning_rate": 9.4604730134524e-07,
+      "loss": 0.87,
+      "step": 78
+    },
+    {
+      "epoch": 39.5,
+      "grad_norm": 1.6593950986862183,
+      "learning_rate": 9.488135456452205e-07,
+      "loss": 0.8142,
+      "step": 79
+    },
+    {
+      "epoch": 40.0,
+      "grad_norm": 1.6837782859802246,
+      "learning_rate": 9.515449934959715e-07,
+      "loss": 0.8246,
+      "step": 80
+    },
+    {
+      "epoch": 40.5,
+      "grad_norm": 1.7322601079940796,
+      "learning_rate": 9.542425094393247e-07,
+      "loss": 0.8752,
+      "step": 81
+    },
+    {
+      "epoch": 41.0,
+      "grad_norm": 1.5245649814605713,
+      "learning_rate": 9.569069261918583e-07,
+      "loss": 0.7944,
+      "step": 82
+    },
+    {
+      "epoch": 41.5,
+      "grad_norm": 1.713905692100525,
+      "learning_rate": 9.59539046188037e-07,
+      "loss": 0.8122,
+      "step": 83
+    },
+    {
+      "epoch": 42.0,
+      "grad_norm": 1.7115576267242432,
+      "learning_rate": 9.621396430309406e-07,
+      "loss": 0.8612,
+      "step": 84
+    },
+    {
+      "epoch": 42.5,
+      "grad_norm": 1.770555019378662,
+      "learning_rate": 9.647094628571462e-07,
+      "loss": 0.9131,
+      "step": 85
+    },
+    {
+      "epoch": 43.0,
+      "grad_norm": 1.8751977682113647,
+      "learning_rate": 9.672492256217836e-07,
+      "loss": 0.8756,
+      "step": 86
+    },
+    {
+      "epoch": 43.5,
+      "grad_norm": 1.8345999717712402,
+      "learning_rate": 9.69759626309309e-07,
+      "loss": 0.9098,
+      "step": 87
+    },
+    {
+      "epoch": 44.0,
+      "grad_norm": 1.6411185264587402,
+      "learning_rate": 9.722413360750842e-07,
+      "loss": 0.7975,
+      "step": 88
+    },
+    {
+      "epoch": 44.5,
+      "grad_norm": 1.744828462600708,
+      "learning_rate": 9.74695003322456e-07,
+      "loss": 0.8444,
+      "step": 89
+    },
+    {
+      "epoch": 45.0,
+      "grad_norm": 1.6588188409805298,
+      "learning_rate": 9.771212547196622e-07,
+      "loss": 0.8376,
+      "step": 90
+    },
+    {
+      "epoch": 45.5,
+      "grad_norm": 1.8046928644180298,
+      "learning_rate": 9.795206961605466e-07,
+      "loss": 0.7546,
+      "step": 91
+    },
+    {
+      "epoch": 46.0,
+      "grad_norm": 1.8828351497650146,
+      "learning_rate": 9.818939136727774e-07,
+      "loss": 0.8505,
+      "step": 92
+    },
+    {
+      "epoch": 46.5,
+      "grad_norm": 1.841956377029419,
+      "learning_rate": 9.842414742769674e-07,
+      "loss": 0.7847,
+      "step": 93
+    },
+    {
+      "epoch": 47.0,
+      "grad_norm": 1.682256817817688,
+      "learning_rate": 9.865639267998492e-07,
+      "loss": 0.7712,
+      "step": 94
+    },
+    {
+      "epoch": 47.5,
+      "grad_norm": 1.8375487327575684,
+      "learning_rate": 9.888618026444236e-07,
+      "loss": 0.7228,
+      "step": 95
+    },
+    {
+      "epoch": 48.0,
+      "grad_norm": 1.7198117971420288,
+      "learning_rate": 9.91135616519784e-07,
+      "loss": 0.7755,
+      "step": 96
+    },
+    {
+      "epoch": 48.5,
+      "grad_norm": 1.760452389717102,
+      "learning_rate": 9.933858671331222e-07,
+      "loss": 0.7045,
+      "step": 97
+    },
+    {
+      "epoch": 49.0,
+      "grad_norm": 1.735704779624939,
+      "learning_rate": 9.956130378462473e-07,
+      "loss": 0.8024,
+      "step": 98
+    },
+    {
+      "epoch": 49.5,
+      "grad_norm": 1.6422948837280273,
+      "learning_rate": 9.978175972987748e-07,
+      "loss": 0.7368,
+      "step": 99
+    },
+    {
+      "epoch": 50.0,
+      "grad_norm": 1.8960306644439697,
+      "learning_rate": 9.999999999999997e-07,
+      "loss": 0.7557,
+      "step": 100
+    },
+    {
+      "epoch": 50.5,
+      "grad_norm": 1.6727304458618164,
+      "learning_rate": 1e-06,
+      "loss": 0.7325,
+      "step": 101
+    },
+    {
+      "epoch": 51.0,
+      "grad_norm": 1.6515084505081177,
+      "learning_rate": 1e-06,
+      "loss": 0.7351,
+      "step": 102
+    },
+    {
+      "epoch": 51.5,
+      "grad_norm": 1.7705847024917603,
+      "learning_rate": 1e-06,
+      "loss": 0.7907,
+      "step": 103
+    },
+    {
+      "epoch": 52.0,
+      "grad_norm": 1.7057950496673584,
+      "learning_rate": 1e-06,
+      "loss": 0.6447,
+      "step": 104
+    },
+    {
+      "epoch": 52.5,
+      "grad_norm": 1.6130571365356445,
+      "learning_rate": 1e-06,
+      "loss": 0.7079,
+      "step": 105
+    },
+    {
+      "epoch": 53.0,
+      "grad_norm": 2.063298463821411,
+      "learning_rate": 1e-06,
+      "loss": 0.693,
+      "step": 106
+    },
+    {
+      "epoch": 53.5,
+      "grad_norm": 2.0730509757995605,
+      "learning_rate": 1e-06,
+      "loss": 0.8002,
+      "step": 107
+    },
+    {
+      "epoch": 54.0,
+      "grad_norm": 1.6381713151931763,
+      "learning_rate": 1e-06,
+      "loss": 0.657,
+      "step": 108
+    },
+    {
+      "epoch": 54.5,
+      "grad_norm": 1.5659828186035156,
+      "learning_rate": 1e-06,
+      "loss": 0.7202,
+      "step": 109
+    },
+    {
+      "epoch": 55.0,
+      "grad_norm": 1.594575047492981,
+      "learning_rate": 1e-06,
+      "loss": 0.6627,
+      "step": 110
+    },
+    {
+      "epoch": 55.5,
+      "grad_norm": 1.497917652130127,
+      "learning_rate": 1e-06,
+      "loss": 0.6693,
+      "step": 111
+    },
+    {
+      "epoch": 56.0,
+      "grad_norm": 2.5011086463928223,
+      "learning_rate": 1e-06,
+      "loss": 0.6946,
+      "step": 112
+    },
+    {
+      "epoch": 56.5,
+      "grad_norm": 1.9602758884429932,
+      "learning_rate": 1e-06,
+      "loss": 0.6926,
+      "step": 113
+    },
+    {
+      "epoch": 57.0,
+      "grad_norm": 1.4793510437011719,
+      "learning_rate": 1e-06,
+      "loss": 0.5861,
+      "step": 114
+    },
+    {
+      "epoch": 57.5,
+      "grad_norm": 1.6028777360916138,
+      "learning_rate": 1e-06,
+      "loss": 0.5788,
+      "step": 115
+    },
+    {
+      "epoch": 58.0,
+      "grad_norm": 1.6478813886642456,
+      "learning_rate": 1e-06,
+      "loss": 0.6395,
+      "step": 116
+    },
+    {
+      "epoch": 58.5,
+      "grad_norm": 1.5423738956451416,
+      "learning_rate": 1e-06,
+      "loss": 0.5959,
+      "step": 117
+    },
+    {
+      "epoch": 59.0,
+      "grad_norm": 1.8497169017791748,
+      "learning_rate": 1e-06,
+      "loss": 0.584,
+      "step": 118
+    },
+    {
+      "epoch": 59.5,
+      "grad_norm": 1.6440547704696655,
+      "learning_rate": 1e-06,
+      "loss": 0.5916,
+      "step": 119
+    },
+    {
+      "epoch": 60.0,
+      "grad_norm": 1.765620231628418,
+      "learning_rate": 1e-06,
+      "loss": 0.6226,
+      "step": 120
+    },
+    {
+      "epoch": 60.5,
+      "grad_norm": 1.543800950050354,
+      "learning_rate": 1e-06,
+      "loss": 0.5966,
+      "step": 121
+    },
+    {
+      "epoch": 61.0,
+      "grad_norm": 1.4944016933441162,
+      "learning_rate": 1e-06,
+      "loss": 0.5305,
+      "step": 122
+    },
+    {
+      "epoch": 61.5,
+      "grad_norm": 1.968621850013733,
+      "learning_rate": 1e-06,
+      "loss": 0.6673,
+      "step": 123
+    },
+    {
+      "epoch": 62.0,
+      "grad_norm": 1.523604393005371,
+      "learning_rate": 1e-06,
+      "loss": 0.5424,
+      "step": 124
+    },
+    {
+      "epoch": 62.5,
+      "grad_norm": 1.6466797590255737,
+      "learning_rate": 1e-06,
+      "loss": 0.6007,
+      "step": 125
+    },
+    {
+      "epoch": 63.0,
+      "grad_norm": 1.7836798429489136,
+      "learning_rate": 1e-06,
+      "loss": 0.6201,
+      "step": 126
+    },
+    {
+      "epoch": 63.5,
+      "grad_norm": 1.6673424243927002,
+      "learning_rate": 1e-06,
+      "loss": 0.5786,
+      "step": 127
+    },
+    {
+      "epoch": 64.0,
+      "grad_norm": 1.6889145374298096,
+      "learning_rate": 1e-06,
+      "loss": 0.5211,
+      "step": 128
+    },
+    {
+      "epoch": 64.5,
+      "grad_norm": 1.4834386110305786,
+      "learning_rate": 1e-06,
+      "loss": 0.4521,
+      "step": 129
+    },
+    {
+      "epoch": 65.0,
+      "grad_norm": 1.743851661682129,
+      "learning_rate": 1e-06,
+      "loss": 0.5363,
+      "step": 130
+    },
+    {
+      "epoch": 65.5,
+      "grad_norm": 1.8134723901748657,
+      "learning_rate": 1e-06,
+      "loss": 0.554,
+      "step": 131
+    },
+    {
+      "epoch": 66.0,
+      "grad_norm": 1.508358120918274,
+      "learning_rate": 1e-06,
+      "loss": 0.5104,
+      "step": 132
+    },
+    {
+      "epoch": 66.5,
+      "grad_norm": 1.6829733848571777,
+      "learning_rate": 1e-06,
+      "loss": 0.4658,
+      "step": 133
+    },
+    {
+      "epoch": 67.0,
+      "grad_norm": 1.526950716972351,
+      "learning_rate": 1e-06,
+      "loss": 0.4892,
+      "step": 134
+    },
+    {
+      "epoch": 67.5,
+      "grad_norm": 1.8935024738311768,
+      "learning_rate": 1e-06,
+      "loss": 0.4979,
+      "step": 135
+    },
+    {
+      "epoch": 68.0,
+      "grad_norm": 1.4638999700546265,
+      "learning_rate": 1e-06,
+      "loss": 0.5144,
+      "step": 136
+    },
+    {
+      "epoch": 68.5,
+      "grad_norm": 1.9910645484924316,
+      "learning_rate": 1e-06,
+      "loss": 0.5521,
+      "step": 137
+    },
+    {
+      "epoch": 69.0,
+      "grad_norm": 1.6257317066192627,
+      "learning_rate": 1e-06,
+      "loss": 0.4937,
+      "step": 138
+    },
+    {
+      "epoch": 69.5,
+      "grad_norm": 1.4498651027679443,
+      "learning_rate": 1e-06,
+      "loss": 0.4749,
+      "step": 139
+    },
+    {
+      "epoch": 70.0,
+      "grad_norm": 1.8104501962661743,
+      "learning_rate": 1e-06,
+      "loss": 0.4564,
+      "step": 140
+    },
+    {
+      "epoch": 70.5,
+      "grad_norm": 2.0244479179382324,
+      "learning_rate": 1e-06,
+      "loss": 0.4228,
+      "step": 141
+    },
+    {
+      "epoch": 71.0,
+      "grad_norm": 1.5190598964691162,
+      "learning_rate": 1e-06,
+      "loss": 0.4735,
+      "step": 142
+    },
+    {
+      "epoch": 71.5,
+      "grad_norm": 1.7180043458938599,
+      "learning_rate": 1e-06,
+      "loss": 0.4776,
+      "step": 143
+    },
+    {
+      "epoch": 72.0,
+      "grad_norm": 1.5680577754974365,
+      "learning_rate": 1e-06,
+      "loss": 0.4343,
+      "step": 144
+    },
+    {
+      "epoch": 72.5,
+      "grad_norm": 1.6798756122589111,
+      "learning_rate": 1e-06,
+      "loss": 0.4074,
+      "step": 145
+    },
+    {
+      "epoch": 73.0,
+      "grad_norm": 1.4644179344177246,
+      "learning_rate": 1e-06,
+      "loss": 0.4982,
+      "step": 146
+    },
+    {
+      "epoch": 73.5,
+      "grad_norm": 1.5461561679840088,
+      "learning_rate": 1e-06,
+      "loss": 0.3539,
+      "step": 147
+    },
+    {
+      "epoch": 74.0,
+      "grad_norm": 1.7116854190826416,
+      "learning_rate": 1e-06,
+      "loss": 0.4267,
+      "step": 148
+    },
+    {
+      "epoch": 74.5,
+      "grad_norm": 1.6357485055923462,
+      "learning_rate": 1e-06,
+      "loss": 0.4563,
+      "step": 149
+    },
+    {
+      "epoch": 75.0,
+      "grad_norm": 1.3843780755996704,
+      "learning_rate": 1e-06,
+      "loss": 0.4072,
+      "step": 150
+    },
+    {
+      "epoch": 75.5,
+      "grad_norm": 1.6510047912597656,
+      "learning_rate": 1e-06,
+      "loss": 0.4619,
+      "step": 151
+    },
+    {
+      "epoch": 76.0,
+      "grad_norm": 1.5008376836776733,
+      "learning_rate": 1e-06,
+      "loss": 0.3768,
+      "step": 152
+    },
+    {
+      "epoch": 76.5,
+      "grad_norm": 1.4433045387268066,
+      "learning_rate": 1e-06,
+      "loss": 0.4502,
+      "step": 153
+    },
+    {
+      "epoch": 77.0,
+      "grad_norm": 1.4826611280441284,
+      "learning_rate": 1e-06,
+      "loss": 0.3686,
+      "step": 154
+    },
+    {
+      "epoch": 77.5,
+      "grad_norm": 1.5890164375305176,
+      "learning_rate": 1e-06,
+      "loss": 0.3642,
+      "step": 155
+    },
+    {
+      "epoch": 78.0,
+      "grad_norm": 1.5281238555908203,
+      "learning_rate": 1e-06,
+      "loss": 0.4034,
+      "step": 156
+    },
+    {
+      "epoch": 78.5,
+      "grad_norm": 1.3185185194015503,
+      "learning_rate": 1e-06,
+      "loss": 0.336,
+      "step": 157
+    },
+    {
+      "epoch": 79.0,
+      "grad_norm": 1.6037932634353638,
+      "learning_rate": 1e-06,
+      "loss": 0.4349,
+      "step": 158
+    },
+    {
+      "epoch": 79.5,
+      "grad_norm": 1.384059190750122,
+      "learning_rate": 1e-06,
+      "loss": 0.3431,
+      "step": 159
+    },
+    {
+      "epoch": 80.0,
+      "grad_norm": 1.339905858039856,
+      "learning_rate": 1e-06,
+      "loss": 0.3745,
+      "step": 160
+    },
+    {
+      "epoch": 80.5,
+      "grad_norm": 1.2671548128128052,
+      "learning_rate": 1e-06,
+      "loss": 0.3476,
+      "step": 161
+    },
+    {
+      "epoch": 81.0,
+      "grad_norm": 1.5032880306243896,
+      "learning_rate": 1e-06,
+      "loss": 0.3297,
+      "step": 162
+    },
+    {
+      "epoch": 81.5,
+      "grad_norm": 1.432960033416748,
+      "learning_rate": 1e-06,
+      "loss": 0.3829,
+      "step": 163
+    },
+    {
+      "epoch": 82.0,
+      "grad_norm": 1.785606026649475,
+      "learning_rate": 1e-06,
+      "loss": 0.4002,
+      "step": 164
+    },
+    {
+      "epoch": 82.5,
+      "grad_norm": 1.599700927734375,
+      "learning_rate": 1e-06,
+      "loss": 0.3272,
+      "step": 165
+    },
+    {
+      "epoch": 83.0,
+      "grad_norm": 1.3606281280517578,
+      "learning_rate": 1e-06,
+      "loss": 0.3462,
+      "step": 166
+    },
+    {
+      "epoch": 83.5,
+      "grad_norm": 1.311733603477478,
+      "learning_rate": 1e-06,
+      "loss": 0.3459,
+      "step": 167
+    },
+    {
+      "epoch": 84.0,
+      "grad_norm": 1.577045202255249,
+      "learning_rate": 1e-06,
+      "loss": 0.278,
+      "step": 168
+    },
+    {
+      "epoch": 84.5,
+      "grad_norm": 1.5641367435455322,
+      "learning_rate": 1e-06,
+      "loss": 0.3636,
+      "step": 169
+    },
+    {
+      "epoch": 85.0,
+      "grad_norm": 1.2674757242202759,
+      "learning_rate": 1e-06,
+      "loss": 0.3074,
+      "step": 170
+    },
+    {
+      "epoch": 85.5,
+      "grad_norm": 1.423398494720459,
+      "learning_rate": 1e-06,
+      "loss": 0.32,
+      "step": 171
+    },
+    {
+      "epoch": 86.0,
+      "grad_norm": 1.149396538734436,
+      "learning_rate": 1e-06,
+      "loss": 0.2577,
+      "step": 172
+    },
+    {
+      "epoch": 86.5,
+      "grad_norm": 1.687155842781067,
+      "learning_rate": 1e-06,
+      "loss": 0.2998,
+      "step": 173
+    },
+    {
+      "epoch": 87.0,
+      "grad_norm": 1.0938485860824585,
+      "learning_rate": 1e-06,
+      "loss": 0.2963,
+      "step": 174
+    },
+    {
+      "epoch": 87.5,
+      "grad_norm": 1.2464781999588013,
+      "learning_rate": 1e-06,
+      "loss": 0.2691,
+      "step": 175
+    },
+    {
+      "epoch": 88.0,
+      "grad_norm": 1.259631633758545,
+      "learning_rate": 1e-06,
+      "loss": 0.2815,
+      "step": 176
+    },
+    {
+      "epoch": 88.5,
+      "grad_norm": 1.2384026050567627,
+      "learning_rate": 1e-06,
+      "loss": 0.1886,
+      "step": 177
+    },
+    {
+      "epoch": 89.0,
+      "grad_norm": 1.209479808807373,
+      "learning_rate": 1e-06,
+      "loss": 0.3283,
+      "step": 178
+    },
+    {
+      "epoch": 89.5,
+      "grad_norm": 1.2056385278701782,
+      "learning_rate": 1e-06,
+      "loss": 0.2795,
+      "step": 179
+    },
+    {
+      "epoch": 90.0,
+      "grad_norm": 1.256142258644104,
+      "learning_rate": 1e-06,
+      "loss": 0.3071,
+      "step": 180
+    },
+    {
+      "epoch": 90.5,
+      "grad_norm": 1.2020200490951538,
+      "learning_rate": 1e-06,
+      "loss": 0.242,
+      "step": 181
+    },
+    {
+      "epoch": 91.0,
+      "grad_norm": 1.275436520576477,
+      "learning_rate": 1e-06,
+      "loss": 0.2505,
+      "step": 182
+    },
+    {
+      "epoch": 91.5,
+      "grad_norm": 1.1096285581588745,
+      "learning_rate": 1e-06,
+      "loss": 0.2833,
+      "step": 183
+    },
+    {
+      "epoch": 92.0,
+      "grad_norm": 1.0823484659194946,
+      "learning_rate": 1e-06,
+      "loss": 0.216,
+      "step": 184
+    },
+    {
+      "epoch": 92.5,
+      "grad_norm": 1.112586498260498,
+      "learning_rate": 1e-06,
+      "loss": 0.2292,
+      "step": 185
+    },
+    {
+      "epoch": 93.0,
+      "grad_norm": 1.004947543144226,
+      "learning_rate": 1e-06,
+      "loss": 0.2399,
+      "step": 186
+    },
+    {
+      "epoch": 93.5,
+      "grad_norm": 1.10011887550354,
+      "learning_rate": 1e-06,
+      "loss": 0.263,
+      "step": 187
+    },
+    {
+      "epoch": 94.0,
+      "grad_norm": 0.9535015821456909,
+      "learning_rate": 1e-06,
+      "loss": 0.218,
+      "step": 188
+    },
+    {
+      "epoch": 94.5,
+      "grad_norm": 1.0121976137161255,
+      "learning_rate": 1e-06,
+      "loss": 0.191,
+      "step": 189
+    },
+    {
+      "epoch": 95.0,
+      "grad_norm": 0.9026556611061096,
+      "learning_rate": 1e-06,
+      "loss": 0.2278,
+      "step": 190
+    },
+    {
+      "epoch": 95.5,
+      "grad_norm": 0.9730249643325806,
+      "learning_rate": 1e-06,
+      "loss": 0.2422,
+      "step": 191
+    },
+    {
+      "epoch": 96.0,
+      "grad_norm": 0.9288642406463623,
+      "learning_rate": 1e-06,
+      "loss": 0.2353,
+      "step": 192
+    },
+    {
+      "epoch": 96.5,
+      "grad_norm": 0.8509739637374878,
+      "learning_rate": 1e-06,
+      "loss": 0.2292,
+      "step": 193
+    },
+    {
+      "epoch": 97.0,
+      "grad_norm": 0.9947998523712158,
+      "learning_rate": 1e-06,
+      "loss": 0.2309,
+      "step": 194
+    },
+    {
+      "epoch": 97.5,
+      "grad_norm": 1.109282374382019,
+      "learning_rate": 1e-06,
+      "loss": 0.2369,
+      "step": 195
+    },
+    {
+      "epoch": 98.0,
+      "grad_norm": 0.8555991053581238,
+      "learning_rate": 1e-06,
+      "loss": 0.2011,
+      "step": 196
+    },
+    {
+      "epoch": 98.5,
+      "grad_norm": 0.9674638509750366,
+      "learning_rate": 1e-06,
+      "loss": 0.2385,
+      "step": 197
+    },
+    {
+      "epoch": 99.0,
+      "grad_norm": 0.781050443649292,
+      "learning_rate": 1e-06,
+      "loss": 0.1881,
+      "step": 198
+    },
+    {
+      "epoch": 99.5,
+      "grad_norm": 0.8599874377250671,
+      "learning_rate": 1e-06,
+      "loss": 0.2031,
+      "step": 199
+    },
+    {
+      "epoch": 100.0,
+      "grad_norm": 0.8711087703704834,
+      "learning_rate": 1e-06,
+      "loss": 0.2214,
+      "step": 200
+    },
+    {
+      "epoch": 100.5,
+      "grad_norm": 0.9213354587554932,
+      "learning_rate": 1e-06,
+      "loss": 0.2313,
+      "step": 201
+    },
+    {
+      "epoch": 101.0,
+      "grad_norm": 0.871462345123291,
+      "learning_rate": 1e-06,
+      "loss": 0.1978,
+      "step": 202
+    },
+    {
+      "epoch": 101.5,
+      "grad_norm": 0.7935155630111694,
+      "learning_rate": 1e-06,
+      "loss": 0.1873,
+      "step": 203
+    },
+    {
+      "epoch": 102.0,
+      "grad_norm": 0.9139618277549744,
+      "learning_rate": 1e-06,
+      "loss": 0.2283,
+      "step": 204
+    },
+    {
+      "epoch": 102.5,
+      "grad_norm": 0.8635255694389343,
+      "learning_rate": 1e-06,
+      "loss": 0.228,
+      "step": 205
+    },
+    {
+      "epoch": 103.0,
+      "grad_norm": 0.9213907122612,
+      "learning_rate": 1e-06,
+      "loss": 0.1837,
+      "step": 206
+    },
+    {
+      "epoch": 103.5,
+      "grad_norm": 0.7787233591079712,
+      "learning_rate": 1e-06,
+      "loss": 0.1652,
+      "step": 207
+    },
+    {
+      "epoch": 104.0,
+      "grad_norm": 0.8260976076126099,
+      "learning_rate": 1e-06,
+      "loss": 0.1986,
+      "step": 208
+    },
+    {
+      "epoch": 104.5,
+      "grad_norm": 0.8949348330497742,
+      "learning_rate": 1e-06,
+      "loss": 0.172,
+      "step": 209
+    },
+    {
+      "epoch": 105.0,
+      "grad_norm": 0.8772971630096436,
+      "learning_rate": 1e-06,
+      "loss": 0.201,
+      "step": 210
+    },
+    {
+      "epoch": 105.5,
+      "grad_norm": 0.7942510843276978,
+      "learning_rate": 1e-06,
+      "loss": 0.1754,
+      "step": 211
+    },
+    {
+      "epoch": 106.0,
+      "grad_norm": 0.8099932670593262,
+      "learning_rate": 1e-06,
+      "loss": 0.1586,
+      "step": 212
+    },
+    {
+      "epoch": 106.5,
+      "grad_norm": 0.880547285079956,
+      "learning_rate": 1e-06,
+      "loss": 0.1516,
+      "step": 213
+    },
+    {
+      "epoch": 107.0,
+      "grad_norm": 0.8132925033569336,
+      "learning_rate": 1e-06,
+      "loss": 0.1657,
+      "step": 214
+    },
+    {
+      "epoch": 107.5,
+      "grad_norm": 0.8455451726913452,
+      "learning_rate": 1e-06,
+      "loss": 0.1994,
+      "step": 215
+    },
+    {
+      "epoch": 108.0,
+      "grad_norm": 0.9202403426170349,
+      "learning_rate": 1e-06,
+      "loss": 0.1486,
+      "step": 216
+    },
+    {
+      "epoch": 108.5,
+      "grad_norm": 0.8958231806755066,
+      "learning_rate": 1e-06,
+      "loss": 0.1949,
+      "step": 217
+    },
+    {
+      "epoch": 109.0,
+      "grad_norm": 0.8252700567245483,
+      "learning_rate": 1e-06,
+      "loss": 0.1645,
+      "step": 218
+    },
+    {
+      "epoch": 109.5,
+      "grad_norm": 0.796977698802948,
+      "learning_rate": 1e-06,
+      "loss": 0.1297,
+      "step": 219
+    },
+    {
+      "epoch": 110.0,
+      "grad_norm": 0.8288230895996094,
+      "learning_rate": 1e-06,
+      "loss": 0.1967,
+      "step": 220
+    },
+    {
+      "epoch": 110.5,
+      "grad_norm": 0.9239948987960815,
+      "learning_rate": 1e-06,
+      "loss": 0.1546,
+      "step": 221
+    },
+    {
+      "epoch": 111.0,
+      "grad_norm": 0.8271680474281311,
+      "learning_rate": 1e-06,
+      "loss": 0.1748,
+      "step": 222
+    },
+    {
+      "epoch": 111.5,
+      "grad_norm": 0.7675459980964661,
+      "learning_rate": 1e-06,
+      "loss": 0.1262,
+      "step": 223
+    },
+    {
+      "epoch": 112.0,
+      "grad_norm": 0.7924964427947998,
+      "learning_rate": 1e-06,
+      "loss": 0.1434,
+      "step": 224
+    },
+    {
+      "epoch": 112.5,
+      "grad_norm": 0.9103841185569763,
+      "learning_rate": 1e-06,
+      "loss": 0.186,
+      "step": 225
+    },
+    {
+      "epoch": 113.0,
+      "grad_norm": 0.8457869291305542,
+      "learning_rate": 1e-06,
+      "loss": 0.1384,
+      "step": 226
+    },
+    {
+      "epoch": 113.5,
+      "grad_norm": 0.8478854298591614,
+      "learning_rate": 1e-06,
+      "loss": 0.1743,
+      "step": 227
+    },
+    {
+      "epoch": 114.0,
+      "grad_norm": 0.8645926713943481,
+      "learning_rate": 1e-06,
+      "loss": 0.1514,
+      "step": 228
+    },
+    {
+      "epoch": 114.5,
+      "grad_norm": 0.9108607769012451,
+      "learning_rate": 1e-06,
+      "loss": 0.1759,
+      "step": 229
+    },
+    {
+      "epoch": 115.0,
+      "grad_norm": 0.8300096392631531,
+      "learning_rate": 1e-06,
+      "loss": 0.1375,
+      "step": 230
+    },
+    {
+      "epoch": 115.5,
+      "grad_norm": 0.9206691384315491,
+      "learning_rate": 1e-06,
+      "loss": 0.1369,
+      "step": 231
+    },
+    {
+      "epoch": 116.0,
+      "grad_norm": 0.7558128833770752,
+      "learning_rate": 1e-06,
+      "loss": 0.1242,
+      "step": 232
+    },
+    {
+      "epoch": 116.5,
+      "grad_norm": 0.8597300052642822,
+      "learning_rate": 1e-06,
+      "loss": 0.128,
+      "step": 233
+    },
+    {
+      "epoch": 117.0,
+      "grad_norm": 0.8134746551513672,
+      "learning_rate": 1e-06,
+      "loss": 0.1547,
+      "step": 234
+    },
+    {
+      "epoch": 117.5,
+      "grad_norm": 0.9657474160194397,
+      "learning_rate": 1e-06,
+      "loss": 0.1534,
+      "step": 235
+    },
+    {
+      "epoch": 118.0,
+      "grad_norm": 0.7481112480163574,
+      "learning_rate": 1e-06,
+      "loss": 0.1282,
+      "step": 236
+    },
+    {
+      "epoch": 118.5,
+      "grad_norm": 0.6953885555267334,
+      "learning_rate": 1e-06,
+      "loss": 0.0988,
+      "step": 237
+    },
+    {
+      "epoch": 119.0,
+      "grad_norm": 0.8225458860397339,
+      "learning_rate": 1e-06,
+      "loss": 0.1135,
+      "step": 238
+    },
+    {
+      "epoch": 119.5,
+      "grad_norm": 0.7915026545524597,
+      "learning_rate": 1e-06,
+      "loss": 0.1043,
+      "step": 239
+    },
+    {
+      "epoch": 120.0,
+      "grad_norm": 0.7963205575942993,
+      "learning_rate": 1e-06,
+      "loss": 0.1474,
+      "step": 240
+    },
+    {
+      "epoch": 120.5,
+      "grad_norm": 0.796559751033783,
+      "learning_rate": 1e-06,
+      "loss": 0.1295,
+      "step": 241
+    },
+    {
+      "epoch": 121.0,
+      "grad_norm": 0.8174726963043213,
+      "learning_rate": 1e-06,
+      "loss": 0.1305,
+      "step": 242
+    },
+    {
+      "epoch": 121.5,
+      "grad_norm": 0.7992000579833984,
+      "learning_rate": 1e-06,
+      "loss": 0.1078,
+      "step": 243
+    },
+    {
+      "epoch": 122.0,
+      "grad_norm": 0.7742059826850891,
+      "learning_rate": 1e-06,
+      "loss": 0.1082,
+      "step": 244
+    },
+    {
+      "epoch": 122.5,
+      "grad_norm": 0.7738575339317322,
+      "learning_rate": 1e-06,
+      "loss": 0.1075,
+      "step": 245
+    },
+    {
+      "epoch": 123.0,
+      "grad_norm": 0.7644574642181396,
+      "learning_rate": 1e-06,
+      "loss": 0.1228,
+      "step": 246
+    },
+    {
+      "epoch": 123.5,
+      "grad_norm": 0.7060183882713318,
+      "learning_rate": 1e-06,
+      "loss": 0.1115,
+      "step": 247
+    },
+    {
+      "epoch": 124.0,
+      "grad_norm": 0.7386205196380615,
+      "learning_rate": 1e-06,
+      "loss": 0.1181,
+      "step": 248
+    },
+    {
+      "epoch": 124.5,
+      "grad_norm": 0.7068957090377808,
+      "learning_rate": 1e-06,
+      "loss": 0.1156,
+      "step": 249
+    },
+    {
+      "epoch": 125.0,
+      "grad_norm": 0.7993431687355042,
+      "learning_rate": 1e-06,
+      "loss": 0.088,
+      "step": 250
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 10000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5000,
+  "save_steps": 50,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.317943074155397e+17,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eddbfe2b39be9df083ba4ee7aa28f6cb6116ae440c069e5ea511599f289c0bdc
+size 7032

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)