Training in progress, step 850, checkpoint

checkpoint-850/README.md

@@ -0,0 +1,202 @@
+---
+base_model: microsoft/Phi-3-mini-4k-instruct
+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 0.14.0

checkpoint-850/adapter_config.json

@@ -0,0 +1,34 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "microsoft/Phi-3-mini-4k-instruct",
+  "bias": "none",
+  "eva_config": null,
+  "exclude_modules": null,
+  "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": 16,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "qkv_proj",
+    "gate_up_proj",
+    "o_proj",
+    "down_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-850/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:065ea2f87f143869b48d6c38a544bae4a7c1d851c8604f05cf76f8939acfe83f
+size 25200088

checkpoint-850/added_tokens.json

@@ -0,0 +1,13 @@
+{
+  "<|assistant|>": 32001,
+  "<|endoftext|>": 32000,
+  "<|end|>": 32007,
+  "<|placeholder1|>": 32002,
+  "<|placeholder2|>": 32003,
+  "<|placeholder3|>": 32004,
+  "<|placeholder4|>": 32005,
+  "<|placeholder5|>": 32008,
+  "<|placeholder6|>": 32009,
+  "<|system|>": 32006,
+  "<|user|>": 32010
+}

checkpoint-850/global_step850/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a48065427637cbb91bbb38e62b2d28b078a8be4ccf34b933dac521dea3ee463f
+size 18881328

checkpoint-850/global_step850/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffea10bb04f757da0020ebaefafd7dbe88e9436c19d9756542bf36f33bad970a
+size 18881328

checkpoint-850/global_step850/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d4a15e9ab92b37b329f417dcbc3d78eec4c9fcb15043fec16962fa934985c0c
+size 18881328

checkpoint-850/global_step850/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8be1311482434adfcd1d4f714d05f3fd82a0934636a5b420cd8d703939052057
+size 18881392

checkpoint-850/global_step850/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79a99ef3e7366c45d54611bd8300c6981818146f939d6dfddeab59dbf07f8369
+size 18881392

checkpoint-850/global_step850/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b255d22f5ea63ee3a648cf0096a1766ad09d0724afd4e2240ecffe4a7af4e07
+size 18881392

checkpoint-850/global_step850/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf879ee70e63316cd2dfb63301a8adcf28f0499cbfaf2f8323b5c21b922608a0
+size 18881392

checkpoint-850/global_step850/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e01e8ebb4c45bd36bd2922edac1d7d9d7665cb04a6e31b6f66ddfdf603eae4d
+size 18881392

checkpoint-850/global_step850/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:836fd0ef3ac8dc04701acb2f5e0c2bc781573e512a286bbaa49ccc4a121c826a
+size 25379244

checkpoint-850/latest

@@ -0,0 +1 @@
+global_step850

checkpoint-850/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f5c4738c31c5c9a38e1f586256d59a0e8e7d02641b9b9af2afdbe078440aeb4
+size 15984

checkpoint-850/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d374b3390eb52ec7f6161c06272d4f26cb715692bdf2ad5374287b6de420ca3
+size 15984

checkpoint-850/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24111edc5a6a2994166cd410155ee3c630816d0fe21c13808ebd2a2ae45bc9d8
+size 15984

checkpoint-850/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:157b21eda1c7f898e519251deed08049767ffba123797289de56343a92ba7380
+size 15984

checkpoint-850/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ccb615552e5845759bc13aa2ae50c0525fbf941fa76ee2e2c20cb9838fe1995
+size 15984

checkpoint-850/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9fcf720fc22147ce563d6f2c2f6f3d916a7e8b7af174b480d072b5c822e992aa
+size 15984

checkpoint-850/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d055d3b033dc8e6fc2a19aa95162960544ab94a903988874315efe4ed5aa8e13
+size 15984

checkpoint-850/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e03c685f2e019350bfdd41f006495a18690aacbccd7ffc1f40de827f433eb87
+size 15984

checkpoint-850/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e018ec51f51e576ee0ab0d945ed5b24487041ac34863e79688b43f75d30d7673
+size 1064

checkpoint-850/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|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
+  }
+}

checkpoint-850/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

checkpoint-850/tokenizer_config.json

@@ -0,0 +1,133 @@
+{
+  "add_bos_token": false,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "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": true,
+      "single_word": false,
+      "special": false
+    },
+    "32000": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32001": {
+      "content": "<|assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32002": {
+      "content": "<|placeholder1|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32003": {
+      "content": "<|placeholder2|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32004": {
+      "content": "<|placeholder3|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32005": {
+      "content": "<|placeholder4|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32006": {
+      "content": "<|system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32007": {
+      "content": "<|end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32008": {
+      "content": "<|placeholder5|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32009": {
+      "content": "<|placeholder6|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32010": {
+      "content": "<|user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "chat_template": "{% set system_message = 'You are a helpful AI assistant.' %}{% if messages[0]['role'] == 'system' %}{% set system_message = messages[0]['content'] %}{% endif %}{% if system_message is defined %}{{ '<s>' + '<|system|>\n' + system_message + '<|end|>\n' }}{% endif %}{% for message in messages %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ '<|user|>\n' + content + '<|end|>\n<|assistant|>\n' }}{% elif message['role'] == 'assistant' %}{{ content + '<|end|>' + '\n' }}{% endif %}{% endfor %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|end|>",
+  "extra_special_tokens": {},
+  "legacy": false,
+  "model_max_length": 4096,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "split_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

checkpoint-850/trainer_state.json

@@ -0,0 +1,1580 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.780980820029861,
+  "eval_steps": 50,
+  "global_step": 850,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.00918800964741013,
+      "grad_norm": 0.036612071096897125,
+      "learning_rate": 4.999451708687114e-06,
+      "logits/chosen": 15.01579761505127,
+      "logits/rejected": 15.359031677246094,
+      "logps/chosen": -0.2681262791156769,
+      "logps/rejected": -0.31947994232177734,
+      "loss": 0.9551,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.40218934416770935,
+      "rewards/margins": 0.07703053951263428,
+      "rewards/rejected": -0.479219913482666,
+      "step": 10
+    },
+    {
+      "epoch": 0.01837601929482026,
+      "grad_norm": 0.05575725808739662,
+      "learning_rate": 4.997807075247147e-06,
+      "logits/chosen": 14.570712089538574,
+      "logits/rejected": 15.321355819702148,
+      "logps/chosen": -0.2867889404296875,
+      "logps/rejected": -0.3514837622642517,
+      "loss": 0.923,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.43018341064453125,
+      "rewards/margins": 0.09704220294952393,
+      "rewards/rejected": -0.5272256135940552,
+      "step": 20
+    },
+    {
+      "epoch": 0.02756402894223039,
+      "grad_norm": 0.0492466576397419,
+      "learning_rate": 4.9950668210706795e-06,
+      "logits/chosen": 14.748420715332031,
+      "logits/rejected": 14.969354629516602,
+      "logps/chosen": -0.28405922651290894,
+      "logps/rejected": -0.32855403423309326,
+      "loss": 0.9357,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.426088809967041,
+      "rewards/margins": 0.06674225628376007,
+      "rewards/rejected": -0.4928310811519623,
+      "step": 30
+    },
+    {
+      "epoch": 0.03675203858964052,
+      "grad_norm": 0.05719422921538353,
+      "learning_rate": 4.9912321481237616e-06,
+      "logits/chosen": 14.28278923034668,
+      "logits/rejected": 14.76964282989502,
+      "logps/chosen": -0.27940627932548523,
+      "logps/rejected": -0.3408831059932709,
+      "loss": 0.9215,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.41910940408706665,
+      "rewards/margins": 0.09221524000167847,
+      "rewards/rejected": -0.5113246440887451,
+      "step": 40
+    },
+    {
+      "epoch": 0.04594004823705065,
+      "grad_norm": 0.06247895210981369,
+      "learning_rate": 4.986304738420684e-06,
+      "logits/chosen": 14.943578720092773,
+      "logits/rejected": 14.936178207397461,
+      "logps/chosen": -0.2819541394710541,
+      "logps/rejected": -0.3245392441749573,
+      "loss": 0.9464,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.4229312539100647,
+      "rewards/margins": 0.06387762725353241,
+      "rewards/rejected": -0.4868088662624359,
+      "step": 50
+    },
+    {
+      "epoch": 0.04594004823705065,
+      "eval_logits/chosen": 14.7594575881958,
+      "eval_logits/rejected": 15.193694114685059,
+      "eval_logps/chosen": -0.2807807922363281,
+      "eval_logps/rejected": -0.36209535598754883,
+      "eval_loss": 0.9397181868553162,
+      "eval_rewards/accuracies": 0.5681818127632141,
+      "eval_rewards/chosen": -0.4211711883544922,
+      "eval_rewards/margins": 0.12197184562683105,
+      "eval_rewards/rejected": -0.5431429743766785,
+      "eval_runtime": 24.9762,
+      "eval_samples_per_second": 28.187,
+      "eval_steps_per_second": 3.523,
+      "step": 50
+    },
+    {
+      "epoch": 0.05512805788446078,
+      "grad_norm": 0.11519577354192734,
+      "learning_rate": 4.980286753286196e-06,
+      "logits/chosen": 14.996228218078613,
+      "logits/rejected": 15.37781810760498,
+      "logps/chosen": -0.2809831202030182,
+      "logps/rejected": -0.35486167669296265,
+      "loss": 0.9318,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.4214746952056885,
+      "rewards/margins": 0.1108178049325943,
+      "rewards/rejected": -0.5322924852371216,
+      "step": 60
+    },
+    {
+      "epoch": 0.06431606753187091,
+      "grad_norm": 0.06691388040781021,
+      "learning_rate": 4.973180832407471e-06,
+      "logits/chosen": 14.612454414367676,
+      "logits/rejected": 15.678136825561523,
+      "logps/chosen": -0.2569667100906372,
+      "logps/rejected": -0.40047627687454224,
+      "loss": 0.9158,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.3854501247406006,
+      "rewards/margins": 0.21526429057121277,
+      "rewards/rejected": -0.600714385509491,
+      "step": 70
+    },
+    {
+      "epoch": 0.07350407717928104,
+      "grad_norm": 0.05976058170199394,
+      "learning_rate": 4.964990092676263e-06,
+      "logits/chosen": 14.873895645141602,
+      "logits/rejected": 15.50474739074707,
+      "logps/chosen": -0.28742527961730957,
+      "logps/rejected": -0.37555089592933655,
+      "loss": 0.9372,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.43113788962364197,
+      "rewards/margins": 0.13218846917152405,
+      "rewards/rejected": -0.5633264183998108,
+      "step": 80
+    },
+    {
+      "epoch": 0.08269208682669117,
+      "grad_norm": 0.0602131113409996,
+      "learning_rate": 4.9557181268217225e-06,
+      "logits/chosen": 14.356691360473633,
+      "logits/rejected": 14.895658493041992,
+      "logps/chosen": -0.2613506317138672,
+      "logps/rejected": -0.3317110538482666,
+      "loss": 0.9324,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.3920259475708008,
+      "rewards/margins": 0.10554064810276031,
+      "rewards/rejected": -0.4975665509700775,
+      "step": 90
+    },
+    {
+      "epoch": 0.0918800964741013,
+      "grad_norm": 0.07126503437757492,
+      "learning_rate": 4.9453690018345144e-06,
+      "logits/chosen": 14.862826347351074,
+      "logits/rejected": 15.257089614868164,
+      "logps/chosen": -0.2707213759422302,
+      "logps/rejected": -0.3511395752429962,
+      "loss": 0.9353,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.4060820937156677,
+      "rewards/margins": 0.1206272691488266,
+      "rewards/rejected": -0.5267094373703003,
+      "step": 100
+    },
+    {
+      "epoch": 0.0918800964741013,
+      "eval_logits/chosen": 14.664334297180176,
+      "eval_logits/rejected": 15.113536834716797,
+      "eval_logps/chosen": -0.2750833034515381,
+      "eval_logps/rejected": -0.36540210247039795,
+      "eval_loss": 0.9324077367782593,
+      "eval_rewards/accuracies": 0.5795454382896423,
+      "eval_rewards/chosen": -0.41262495517730713,
+      "eval_rewards/margins": 0.1354781985282898,
+      "eval_rewards/rejected": -0.5481031537055969,
+      "eval_runtime": 24.4286,
+      "eval_samples_per_second": 28.819,
+      "eval_steps_per_second": 3.602,
+      "step": 100
+    },
+    {
+      "epoch": 0.10106810612151143,
+      "grad_norm": 0.07136944681406021,
+      "learning_rate": 4.933947257182901e-06,
+      "logits/chosen": 14.942098617553711,
+      "logits/rejected": 15.138586044311523,
+      "logps/chosen": -0.2860812246799469,
+      "logps/rejected": -0.36259371042251587,
+      "loss": 0.934,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.42912182211875916,
+      "rewards/margins": 0.11476878076791763,
+      "rewards/rejected": -0.5438905954360962,
+      "step": 110
+    },
+    {
+      "epoch": 0.11025611576892155,
+      "grad_norm": 0.07038908451795578,
+      "learning_rate": 4.921457902821578e-06,
+      "logits/chosen": 14.488851547241211,
+      "logits/rejected": 14.702054023742676,
+      "logps/chosen": -0.2662215232849121,
+      "logps/rejected": -0.3013685941696167,
+      "loss": 0.9202,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.39933228492736816,
+      "rewards/margins": 0.05272058770060539,
+      "rewards/rejected": -0.45205289125442505,
+      "step": 120
+    },
+    {
+      "epoch": 0.11944412541633169,
+      "grad_norm": 0.06875801086425781,
+      "learning_rate": 4.907906416994146e-06,
+      "logits/chosen": 14.075657844543457,
+      "logits/rejected": 14.696513175964355,
+      "logps/chosen": -0.250360369682312,
+      "logps/rejected": -0.3504650592803955,
+      "loss": 0.9266,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.375540554523468,
+      "rewards/margins": 0.15015706419944763,
+      "rewards/rejected": -0.5256975889205933,
+      "step": 130
+    },
+    {
+      "epoch": 0.12863213506374183,
+      "grad_norm": 0.0984601378440857,
+      "learning_rate": 4.893298743830168e-06,
+      "logits/chosen": 13.738212585449219,
+      "logits/rejected": 14.311574935913086,
+      "logps/chosen": -0.26711025834083557,
+      "logps/rejected": -0.3587702810764313,
+      "loss": 0.9185,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.40066537261009216,
+      "rewards/margins": 0.13749003410339355,
+      "rewards/rejected": -0.5381554365158081,
+      "step": 140
+    },
+    {
+      "epoch": 0.13782014471115195,
+      "grad_norm": 0.10201425850391388,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": 13.7462797164917,
+      "logits/rejected": 14.230626106262207,
+      "logps/chosen": -0.25559619069099426,
+      "logps/rejected": -0.3708702623844147,
+      "loss": 0.9106,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.3833943009376526,
+      "rewards/margins": 0.17291104793548584,
+      "rewards/rejected": -0.5563054084777832,
+      "step": 150
+    },
+    {
+      "epoch": 0.13782014471115195,
+      "eval_logits/chosen": 13.458538055419922,
+      "eval_logits/rejected": 13.998083114624023,
+      "eval_logps/chosen": -0.2759075462818146,
+      "eval_logps/rejected": -0.3873325288295746,
+      "eval_loss": 0.9164085388183594,
+      "eval_rewards/accuracies": 0.5795454382896423,
+      "eval_rewards/chosen": -0.41386130452156067,
+      "eval_rewards/margins": 0.1671374887228012,
+      "eval_rewards/rejected": -0.5809988379478455,
+      "eval_runtime": 24.4393,
+      "eval_samples_per_second": 28.806,
+      "eval_steps_per_second": 3.601,
+      "step": 150
+    },
+    {
+      "epoch": 0.14700815435856207,
+      "grad_norm": 0.11537656933069229,
+      "learning_rate": 4.860940925593703e-06,
+      "logits/chosen": 12.686149597167969,
+      "logits/rejected": 13.478736877441406,
+      "logps/chosen": -0.23941929638385773,
+      "logps/rejected": -0.3713286519050598,
+      "loss": 0.9094,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.3591288924217224,
+      "rewards/margins": 0.1978640854358673,
+      "rewards/rejected": -0.5569929480552673,
+      "step": 160
+    },
+    {
+      "epoch": 0.1561961640059722,
+      "grad_norm": 0.1196313351392746,
+      "learning_rate": 4.84320497372973e-06,
+      "logits/chosen": 13.221656799316406,
+      "logits/rejected": 13.317082405090332,
+      "logps/chosen": -0.3033878207206726,
+      "logps/rejected": -0.3784424960613251,
+      "loss": 0.9057,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.4550817608833313,
+      "rewards/margins": 0.11258199065923691,
+      "rewards/rejected": -0.5676637887954712,
+      "step": 170
+    },
+    {
+      "epoch": 0.16538417365338234,
+      "grad_norm": 0.18745549023151398,
+      "learning_rate": 4.824441214720629e-06,
+      "logits/chosen": 11.797627449035645,
+      "logits/rejected": 12.031414985656738,
+      "logps/chosen": -0.2746419608592987,
+      "logps/rejected": -0.3629845976829529,
+      "loss": 0.8954,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.41196292638778687,
+      "rewards/margins": 0.13251398503780365,
+      "rewards/rejected": -0.5444768667221069,
+      "step": 180
+    },
+    {
+      "epoch": 0.17457218330079247,
+      "grad_norm": 0.1806156188249588,
+      "learning_rate": 4.804657878971252e-06,
+      "logits/chosen": 10.275301933288574,
+      "logits/rejected": 10.937273025512695,
+      "logps/chosen": -0.2880379557609558,
+      "logps/rejected": -0.4154580533504486,
+      "loss": 0.8875,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.43205690383911133,
+      "rewards/margins": 0.19113019108772278,
+      "rewards/rejected": -0.6231871247291565,
+      "step": 190
+    },
+    {
+      "epoch": 0.1837601929482026,
+      "grad_norm": 0.1839464157819748,
+      "learning_rate": 4.783863644106502e-06,
+      "logits/chosen": 10.020039558410645,
+      "logits/rejected": 10.66059398651123,
+      "logps/chosen": -0.3136019706726074,
+      "logps/rejected": -0.4385503828525543,
+      "loss": 0.8647,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.47040295600891113,
+      "rewards/margins": 0.18742261826992035,
+      "rewards/rejected": -0.6578255891799927,
+      "step": 200
+    },
+    {
+      "epoch": 0.1837601929482026,
+      "eval_logits/chosen": 9.442557334899902,
+      "eval_logits/rejected": 10.053345680236816,
+      "eval_logps/chosen": -0.3080674409866333,
+      "eval_logps/rejected": -0.4899139702320099,
+      "eval_loss": 0.8702690005302429,
+      "eval_rewards/accuracies": 0.6931818127632141,
+      "eval_rewards/chosen": -0.46210116147994995,
+      "eval_rewards/margins": 0.27276986837387085,
+      "eval_rewards/rejected": -0.7348710894584656,
+      "eval_runtime": 24.4185,
+      "eval_samples_per_second": 28.831,
+      "eval_steps_per_second": 3.604,
+      "step": 200
+    },
+    {
+      "epoch": 0.19294820259561274,
+      "grad_norm": 0.269613116979599,
+      "learning_rate": 4.762067631165049e-06,
+      "logits/chosen": 7.941342353820801,
+      "logits/rejected": 8.542920112609863,
+      "logps/chosen": -0.3083941638469696,
+      "logps/rejected": -0.5024437308311462,
+      "loss": 0.8471,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.4625912606716156,
+      "rewards/margins": 0.29107433557510376,
+      "rewards/rejected": -0.7536656856536865,
+      "step": 210
+    },
+    {
+      "epoch": 0.20213621224302286,
+      "grad_norm": 0.2640094459056854,
+      "learning_rate": 4.7392794005985324e-06,
+      "logits/chosen": 7.587499141693115,
+      "logits/rejected": 7.592519283294678,
+      "logps/chosen": -0.3381899893283844,
+      "logps/rejected": -0.48494213819503784,
+      "loss": 0.8427,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.5072849988937378,
+      "rewards/margins": 0.22012826800346375,
+      "rewards/rejected": -0.7274132966995239,
+      "step": 220
+    },
+    {
+      "epoch": 0.21132422189043298,
+      "grad_norm": 0.29708293080329895,
+      "learning_rate": 4.715508948078037e-06,
+      "logits/chosen": 6.250656604766846,
+      "logits/rejected": 6.7652716636657715,
+      "logps/chosen": -0.3644888997077942,
+      "logps/rejected": -0.5470594167709351,
+      "loss": 0.8201,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.5467333793640137,
+      "rewards/margins": 0.2738557755947113,
+      "rewards/rejected": -0.8205891847610474,
+      "step": 230
+    },
+    {
+      "epoch": 0.2205122315378431,
+      "grad_norm": 0.35299497842788696,
+      "learning_rate": 4.690766700109659e-06,
+      "logits/chosen": 4.6331706047058105,
+      "logits/rejected": 4.710076332092285,
+      "logps/chosen": -0.3634452223777771,
+      "logps/rejected": -0.7193974256515503,
+      "loss": 0.7877,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.545167863368988,
+      "rewards/margins": 0.5339283347129822,
+      "rewards/rejected": -1.0790963172912598,
+      "step": 240
+    },
+    {
+      "epoch": 0.22970024118525326,
+      "grad_norm": 0.4265730082988739,
+      "learning_rate": 4.665063509461098e-06,
+      "logits/chosen": 4.992984771728516,
+      "logits/rejected": 4.606354713439941,
+      "logps/chosen": -0.413116455078125,
+      "logps/rejected": -0.7104976177215576,
+      "loss": 0.7902,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.6196746826171875,
+      "rewards/margins": 0.4460717737674713,
+      "rewards/rejected": -1.0657463073730469,
+      "step": 250
+    },
+    {
+      "epoch": 0.22970024118525326,
+      "eval_logits/chosen": 4.127804279327393,
+      "eval_logits/rejected": 3.742251396179199,
+      "eval_logps/chosen": -0.420327365398407,
+      "eval_logps/rejected": -0.7902651429176331,
+      "eval_loss": 0.7682384252548218,
+      "eval_rewards/accuracies": 0.7159090638160706,
+      "eval_rewards/chosen": -0.6304910182952881,
+      "eval_rewards/margins": 0.5549066662788391,
+      "eval_rewards/rejected": -1.185397744178772,
+      "eval_runtime": 24.4318,
+      "eval_samples_per_second": 28.815,
+      "eval_steps_per_second": 3.602,
+      "step": 250
+    },
+    {
+      "epoch": 0.23888825083266338,
+      "grad_norm": 0.7236106395721436,
+      "learning_rate": 4.638410650401267e-06,
+      "logits/chosen": 2.454423427581787,
+      "logits/rejected": 1.816563367843628,
+      "logps/chosen": -0.4492695927619934,
+      "logps/rejected": -0.8738088607788086,
+      "loss": 0.6911,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.6739044189453125,
+      "rewards/margins": 0.6368088126182556,
+      "rewards/rejected": -1.3107131719589233,
+      "step": 260
+    },
+    {
+      "epoch": 0.2480762604800735,
+      "grad_norm": 0.5856125950813293,
+      "learning_rate": 4.610819813755038e-06,
+      "logits/chosen": 3.2105612754821777,
+      "logits/rejected": 2.5531132221221924,
+      "logps/chosen": -0.537078320980072,
+      "logps/rejected": -1.2025481462478638,
+      "loss": 0.6774,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.80561763048172,
+      "rewards/margins": 0.9982045888900757,
+      "rewards/rejected": -1.8038222789764404,
+      "step": 270
+    },
+    {
+      "epoch": 0.25726427012748365,
+      "grad_norm": 0.7396731972694397,
+      "learning_rate": 4.582303101775249e-06,
+      "logits/chosen": 2.0327231884002686,
+      "logits/rejected": 1.4601097106933594,
+      "logps/chosen": -0.47658151388168335,
+      "logps/rejected": -1.3808696269989014,
+      "loss": 0.628,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.7148722410202026,
+      "rewards/margins": 1.3564319610595703,
+      "rewards/rejected": -2.0713045597076416,
+      "step": 280
+    },
+    {
+      "epoch": 0.2664522797748938,
+      "grad_norm": 2.412203550338745,
+      "learning_rate": 4.55287302283426e-06,
+      "logits/chosen": 1.2262591123580933,
+      "logits/rejected": 0.22599482536315918,
+      "logps/chosen": -0.5671601891517639,
+      "logps/rejected": -1.6760343313217163,
+      "loss": 0.5988,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.8507402539253235,
+      "rewards/margins": 1.6633113622665405,
+      "rewards/rejected": -2.5140514373779297,
+      "step": 290
+    },
+    {
+      "epoch": 0.2756402894223039,
+      "grad_norm": 1.0477895736694336,
+      "learning_rate": 4.522542485937369e-06,
+      "logits/chosen": 1.9952911138534546,
+      "logits/rejected": 1.1298446655273438,
+      "logps/chosen": -0.6152974367141724,
+      "logps/rejected": -2.128481388092041,
+      "loss": 0.5927,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.9229460954666138,
+      "rewards/margins": 2.2697763442993164,
+      "rewards/rejected": -3.192722797393799,
+      "step": 300
+    },
+    {
+      "epoch": 0.2756402894223039,
+      "eval_logits/chosen": 1.6342910528182983,
+      "eval_logits/rejected": 0.633538007736206,
+      "eval_logps/chosen": -0.606099545955658,
+      "eval_logps/rejected": -1.882785439491272,
+      "eval_loss": 0.5978505611419678,
+      "eval_rewards/accuracies": 0.7159090638160706,
+      "eval_rewards/chosen": -0.909149169921875,
+      "eval_rewards/margins": 1.9150291681289673,
+      "eval_rewards/rejected": -2.824178457260132,
+      "eval_runtime": 24.4299,
+      "eval_samples_per_second": 28.817,
+      "eval_steps_per_second": 3.602,
+      "step": 300
+    },
+    {
+      "epoch": 0.284828299069714,
+      "grad_norm": 3.0767388343811035,
+      "learning_rate": 4.491324795060491e-06,
+      "logits/chosen": 1.0374902486801147,
+      "logits/rejected": 0.6220051646232605,
+      "logps/chosen": -0.6778531074523926,
+      "logps/rejected": -1.8290255069732666,
+      "loss": 0.5792,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -1.0167796611785889,
+      "rewards/margins": 1.726758599281311,
+      "rewards/rejected": -2.7435383796691895,
+      "step": 310
+    },
+    {
+      "epoch": 0.29401630871712414,
+      "grad_norm": 0.6015120148658752,
+      "learning_rate": 4.4592336433146e-06,
+      "logits/chosen": 1.0050956010818481,
+      "logits/rejected": -0.016118621453642845,
+      "logps/chosen": -0.6865260004997253,
+      "logps/rejected": -2.113417148590088,
+      "loss": 0.5384,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -1.0297890901565552,
+      "rewards/margins": 2.140336513519287,
+      "rewards/rejected": -3.1701254844665527,
+      "step": 320
+    },
+    {
+      "epoch": 0.30320431836453426,
+      "grad_norm": 0.7415631413459778,
+      "learning_rate": 4.426283106939474e-06,
+      "logits/chosen": 0.5536144971847534,
+      "logits/rejected": -0.1644023358821869,
+      "logps/chosen": -0.8181726336479187,
+      "logps/rejected": -2.581185817718506,
+      "loss": 0.5671,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.2272589206695557,
+      "rewards/margins": 2.644519805908203,
+      "rewards/rejected": -3.871778964996338,
+      "step": 330
+    },
+    {
+      "epoch": 0.3123923280119444,
+      "grad_norm": 0.8956871628761292,
+      "learning_rate": 4.3924876391293915e-06,
+      "logits/chosen": 1.6062015295028687,
+      "logits/rejected": 0.9243733286857605,
+      "logps/chosen": -0.8991573452949524,
+      "logps/rejected": -2.935060977935791,
+      "loss": 0.5124,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -1.348736047744751,
+      "rewards/margins": 3.0538551807403564,
+      "rewards/rejected": -4.402591228485107,
+      "step": 340
+    },
+    {
+      "epoch": 0.32158033765935456,
+      "grad_norm": 1.1822264194488525,
+      "learning_rate": 4.357862063693486e-06,
+      "logits/chosen": 1.8171085119247437,
+      "logits/rejected": 1.228049397468567,
+      "logps/chosen": -0.8822734951972961,
+      "logps/rejected": -2.4744174480438232,
+      "loss": 0.516,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -1.3234103918075562,
+      "rewards/margins": 2.388216018676758,
+      "rewards/rejected": -3.7116265296936035,
+      "step": 350
+    },
+    {
+      "epoch": 0.32158033765935456,
+      "eval_logits/chosen": 1.1368330717086792,
+      "eval_logits/rejected": 0.2795785665512085,
+      "eval_logps/chosen": -0.9485350251197815,
+      "eval_logps/rejected": -2.6484899520874023,
+      "eval_loss": 0.5133901238441467,
+      "eval_rewards/accuracies": 0.7727272510528564,
+      "eval_rewards/chosen": -1.4228025674819946,
+      "eval_rewards/margins": 2.5499324798583984,
+      "eval_rewards/rejected": -3.9727354049682617,
+      "eval_runtime": 24.4277,
+      "eval_samples_per_second": 28.82,
+      "eval_steps_per_second": 3.602,
+      "step": 350
+    },
+    {
+      "epoch": 0.3307683473067647,
+      "grad_norm": 2.5278775691986084,
+      "learning_rate": 4.322421568553529e-06,
+      "logits/chosen": 1.1921958923339844,
+      "logits/rejected": 0.7565670013427734,
+      "logps/chosen": -1.4180412292480469,
+      "logps/rejected": -3.0890870094299316,
+      "loss": 0.4811,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -2.1270618438720703,
+      "rewards/margins": 2.5065689086914062,
+      "rewards/rejected": -4.633630752563477,
+      "step": 360
+    },
+    {
+      "epoch": 0.3399563569541748,
+      "grad_norm": 1.7325788736343384,
+      "learning_rate": 4.286181699082008e-06,
+      "logits/chosen": 0.997096836566925,
+      "logits/rejected": 0.4399908483028412,
+      "logps/chosen": -1.9010308980941772,
+      "logps/rejected": -3.6025185585021973,
+      "loss": 0.4326,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -2.851546287536621,
+      "rewards/margins": 2.552231550216675,
+      "rewards/rejected": -5.403778076171875,
+      "step": 370
+    },
+    {
+      "epoch": 0.34914436660158493,
+      "grad_norm": 4.608370304107666,
+      "learning_rate": 4.249158351283414e-06,
+      "logits/chosen": 1.0715999603271484,
+      "logits/rejected": 0.6113725900650024,
+      "logps/chosen": -2.4032845497131348,
+      "logps/rejected": -3.9940528869628906,
+      "loss": 0.4027,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -3.6049270629882812,
+      "rewards/margins": 2.3861520290374756,
+      "rewards/rejected": -5.991078853607178,
+      "step": 380
+    },
+    {
+      "epoch": 0.35833237624899505,
+      "grad_norm": 4.600816249847412,
+      "learning_rate": 4.211367764821722e-06,
+      "logits/chosen": 1.4443682432174683,
+      "logits/rejected": 0.8617011904716492,
+      "logps/chosen": -2.349093198776245,
+      "logps/rejected": -3.9943203926086426,
+      "loss": 0.4118,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -3.5236401557922363,
+      "rewards/margins": 2.467839479446411,
+      "rewards/rejected": -5.991480350494385,
+      "step": 390
+    },
+    {
+      "epoch": 0.3675203858964052,
+      "grad_norm": 2.1458094120025635,
+      "learning_rate": 4.172826515897146e-06,
+      "logits/chosen": 1.3029582500457764,
+      "logits/rejected": 0.7705980539321899,
+      "logps/chosen": -2.192617416381836,
+      "logps/rejected": -3.8413078784942627,
+      "loss": 0.3557,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -3.288925886154175,
+      "rewards/margins": 2.4730350971221924,
+      "rewards/rejected": -5.761960506439209,
+      "step": 400
+    },
+    {
+      "epoch": 0.3675203858964052,
+      "eval_logits/chosen": 0.8742353320121765,
+      "eval_logits/rejected": 0.14320053160190582,
+      "eval_logps/chosen": -2.0894505977630615,
+      "eval_logps/rejected": -4.20783805847168,
+      "eval_loss": 0.4082850515842438,
+      "eval_rewards/accuracies": 0.8863636255264282,
+      "eval_rewards/chosen": -3.1341757774353027,
+      "eval_rewards/margins": 3.1775810718536377,
+      "eval_rewards/rejected": -6.311756610870361,
+      "eval_runtime": 24.4316,
+      "eval_samples_per_second": 28.815,
+      "eval_steps_per_second": 3.602,
+      "step": 400
+    },
+    {
+      "epoch": 0.3767083955438153,
+      "grad_norm": 2.5496387481689453,
+      "learning_rate": 4.133551509975264e-06,
+      "logits/chosen": 1.9024279117584229,
+      "logits/rejected": 1.4777928590774536,
+      "logps/chosen": -2.7541470527648926,
+      "logps/rejected": -4.542642116546631,
+      "loss": 0.4184,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.13122034072876,
+      "rewards/margins": 2.6827428340911865,
+      "rewards/rejected": -6.813962459564209,
+      "step": 410
+    },
+    {
+      "epoch": 0.3858964051912255,
+      "grad_norm": 2.6187174320220947,
+      "learning_rate": 4.093559974371725e-06,
+      "logits/chosen": 1.9345887899398804,
+      "logits/rejected": 1.4047685861587524,
+      "logps/chosen": -2.389430522918701,
+      "logps/rejected": -4.439882755279541,
+      "loss": 0.4096,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -3.5841457843780518,
+      "rewards/margins": 3.0756776332855225,
+      "rewards/rejected": -6.659823417663574,
+      "step": 420
+    },
+    {
+      "epoch": 0.3950844148386356,
+      "grad_norm": 3.762899398803711,
+      "learning_rate": 4.052869450695776e-06,
+      "logits/chosen": 1.2521915435791016,
+      "logits/rejected": 0.7238092422485352,
+      "logps/chosen": -2.5571534633636475,
+      "logps/rejected": -4.675185203552246,
+      "loss": 0.4083,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -3.835730791091919,
+      "rewards/margins": 3.1770474910736084,
+      "rewards/rejected": -7.012777805328369,
+      "step": 430
+    },
+    {
+      "epoch": 0.4042724244860457,
+      "grad_norm": 3.2343404293060303,
+      "learning_rate": 4.011497787155938e-06,
+      "logits/chosen": 1.9626567363739014,
+      "logits/rejected": 1.3136894702911377,
+      "logps/chosen": -2.892967939376831,
+      "logps/rejected": -5.003688335418701,
+      "loss": 0.3783,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -4.339452266693115,
+      "rewards/margins": 3.1660804748535156,
+      "rewards/rejected": -7.505532264709473,
+      "step": 440
+    },
+    {
+      "epoch": 0.41346043413345585,
+      "grad_norm": 3.2979824542999268,
+      "learning_rate": 3.969463130731183e-06,
+      "logits/chosen": 1.159234881401062,
+      "logits/rejected": 0.5396692752838135,
+      "logps/chosen": -2.7545204162597656,
+      "logps/rejected": -5.4179182052612305,
+      "loss": 0.347,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.131781101226807,
+      "rewards/margins": 3.9950966835021973,
+      "rewards/rejected": -8.126876831054688,
+      "step": 450
+    },
+    {
+      "epoch": 0.41346043413345585,
+      "eval_logits/chosen": 0.9304068088531494,
+      "eval_logits/rejected": 0.30015668272972107,
+      "eval_logps/chosen": -2.485308885574341,
+      "eval_logps/rejected": -4.880238056182861,
+      "eval_loss": 0.36501914262771606,
+      "eval_rewards/accuracies": 0.9204545617103577,
+      "eval_rewards/chosen": -3.7279627323150635,
+      "eval_rewards/margins": 3.592393636703491,
+      "eval_rewards/rejected": -7.320356845855713,
+      "eval_runtime": 24.4234,
+      "eval_samples_per_second": 28.825,
+      "eval_steps_per_second": 3.603,
+      "step": 450
+    },
+    {
+      "epoch": 0.42264844378086597,
+      "grad_norm": 3.155860185623169,
+      "learning_rate": 3.92678391921108e-06,
+      "logits/chosen": 1.633522391319275,
+      "logits/rejected": 1.0575059652328491,
+      "logps/chosen": -2.7214303016662598,
+      "logps/rejected": -4.766176223754883,
+      "loss": 0.3689,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.082144737243652,
+      "rewards/margins": 3.0671191215515137,
+      "rewards/rejected": -7.149264335632324,
+      "step": 460
+    },
+    {
+      "epoch": 0.4318364534282761,
+      "grad_norm": 2.9949357509613037,
+      "learning_rate": 3.88347887310836e-06,
+      "logits/chosen": 1.3139212131500244,
+      "logits/rejected": 0.5838541388511658,
+      "logps/chosen": -2.5324313640594482,
+      "logps/rejected": -4.993292808532715,
+      "loss": 0.3587,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -3.7986464500427246,
+      "rewards/margins": 3.6912925243377686,
+      "rewards/rejected": -7.489938259124756,
+      "step": 470
+    },
+    {
+      "epoch": 0.4410244630756862,
+      "grad_norm": 2.4548208713531494,
+      "learning_rate": 3.839566987447492e-06,
+      "logits/chosen": 1.2342166900634766,
+      "logits/rejected": 0.8118699193000793,
+      "logps/chosen": -2.876863479614258,
+      "logps/rejected": -5.457588195800781,
+      "loss": 0.2922,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -4.315295219421387,
+      "rewards/margins": 3.871086597442627,
+      "rewards/rejected": -8.186381340026855,
+      "step": 480
+    },
+    {
+      "epoch": 0.45021247272309634,
+      "grad_norm": 2.0953762531280518,
+      "learning_rate": 3.795067523432826e-06,
+      "logits/chosen": 1.8121936321258545,
+      "logits/rejected": 1.454637050628662,
+      "logps/chosen": -3.2022106647491455,
+      "logps/rejected": -5.47930908203125,
+      "loss": 0.3086,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.803316116333008,
+      "rewards/margins": 3.4156479835510254,
+      "rewards/rejected": -8.218963623046875,
+      "step": 490
+    },
+    {
+      "epoch": 0.4594004823705065,
+      "grad_norm": 2.4918301105499268,
+      "learning_rate": 3.7500000000000005e-06,
+      "logits/chosen": 1.5386875867843628,
+      "logits/rejected": 1.2544763088226318,
+      "logps/chosen": -3.2174277305603027,
+      "logps/rejected": -6.012864112854004,
+      "loss": 0.3128,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.826140880584717,
+      "rewards/margins": 4.193154335021973,
+      "rewards/rejected": -9.019296646118164,
+      "step": 500
+    },
+    {
+      "epoch": 0.4594004823705065,
+      "eval_logits/chosen": 0.8257808685302734,
+      "eval_logits/rejected": 0.27475622296333313,
+      "eval_logps/chosen": -2.755974054336548,
+      "eval_logps/rejected": -5.469714641571045,
+      "eval_loss": 0.32884541153907776,
+      "eval_rewards/accuracies": 0.9204545617103577,
+      "eval_rewards/chosen": -4.133961200714111,
+      "eval_rewards/margins": 4.070610523223877,
+      "eval_rewards/rejected": -8.204572677612305,
+      "eval_runtime": 24.4191,
+      "eval_samples_per_second": 28.83,
+      "eval_steps_per_second": 3.604,
+      "step": 500
+    },
+    {
+      "epoch": 0.46858849201791664,
+      "grad_norm": 4.017474174499512,
+      "learning_rate": 3.7043841852542884e-06,
+      "logits/chosen": 1.8042447566986084,
+      "logits/rejected": 1.4390740394592285,
+      "logps/chosen": -3.4311797618865967,
+      "logps/rejected": -6.114380359649658,
+      "loss": 0.335,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.1467695236206055,
+      "rewards/margins": 4.0248003005981445,
+      "rewards/rejected": -9.171568870544434,
+      "step": 510
+    },
+    {
+      "epoch": 0.47777650166532676,
+      "grad_norm": 5.516397953033447,
+      "learning_rate": 3.658240087799655e-06,
+      "logits/chosen": 1.1141811609268188,
+      "logits/rejected": 0.7766789197921753,
+      "logps/chosen": -3.1831612586975098,
+      "logps/rejected": -5.615653991699219,
+      "loss": 0.3306,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.774742603302002,
+      "rewards/margins": 3.6487393379211426,
+      "rewards/rejected": -8.423480987548828,
+      "step": 520
+    },
+    {
+      "epoch": 0.4869645113127369,
+      "grad_norm": 4.1005635261535645,
+      "learning_rate": 3.611587947962319e-06,
+      "logits/chosen": 1.5796890258789062,
+      "logits/rejected": 1.1561863422393799,
+      "logps/chosen": -3.286179304122925,
+      "logps/rejected": -5.805339813232422,
+      "loss": 0.3151,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.929268836975098,
+      "rewards/margins": 3.778740644454956,
+      "rewards/rejected": -8.708009719848633,
+      "step": 530
+    },
+    {
+      "epoch": 0.496152520960147,
+      "grad_norm": 2.8616511821746826,
+      "learning_rate": 3.564448228912682e-06,
+      "logits/chosen": 0.6956934928894043,
+      "logits/rejected": 0.1268310248851776,
+      "logps/chosen": -2.8933937549591064,
+      "logps/rejected": -5.794272422790527,
+      "loss": 0.2995,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.340090274810791,
+      "rewards/margins": 4.351318836212158,
+      "rewards/rejected": -8.691408157348633,
+      "step": 540
+    },
+    {
+      "epoch": 0.5053405306075571,
+      "grad_norm": 2.237276315689087,
+      "learning_rate": 3.516841607689501e-06,
+      "logits/chosen": 1.7460235357284546,
+      "logits/rejected": 0.9990445375442505,
+      "logps/chosen": -2.860546827316284,
+      "logps/rejected": -5.517810821533203,
+      "loss": 0.3203,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.290820121765137,
+      "rewards/margins": 3.985896348953247,
+      "rewards/rejected": -8.276716232299805,
+      "step": 550
+    },
+    {
+      "epoch": 0.5053405306075571,
+      "eval_logits/chosen": 0.8703196048736572,
+      "eval_logits/rejected": 0.3152187466621399,
+      "eval_logps/chosen": -2.709866523742676,
+      "eval_logps/rejected": -5.687611103057861,
+      "eval_loss": 0.304200142621994,
+      "eval_rewards/accuracies": 0.9204545617103577,
+      "eval_rewards/chosen": -4.064799785614014,
+      "eval_rewards/margins": 4.466617107391357,
+      "eval_rewards/rejected": -8.531416893005371,
+      "eval_runtime": 24.6083,
+      "eval_samples_per_second": 28.608,
+      "eval_steps_per_second": 3.576,
+      "step": 550
+    },
+    {
+      "epoch": 0.5145285402549673,
+      "grad_norm": 3.7983174324035645,
+      "learning_rate": 3.4687889661302577e-06,
+      "logits/chosen": 1.084142804145813,
+      "logits/rejected": 0.9131366610527039,
+      "logps/chosen": -3.4958484172821045,
+      "logps/rejected": -7.143038749694824,
+      "loss": 0.2728,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.243772506713867,
+      "rewards/margins": 5.470786094665527,
+      "rewards/rejected": -10.714558601379395,
+      "step": 560
+    },
+    {
+      "epoch": 0.5237165499023774,
+      "grad_norm": 2.398188829421997,
+      "learning_rate": 3.4203113817116955e-06,
+      "logits/chosen": 1.2777886390686035,
+      "logits/rejected": 0.6827106475830078,
+      "logps/chosen": -3.4014134407043457,
+      "logps/rejected": -6.260494232177734,
+      "loss": 0.2833,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.102120399475098,
+      "rewards/margins": 4.288620948791504,
+      "rewards/rejected": -9.390741348266602,
+      "step": 570
+    },
+    {
+      "epoch": 0.5329045595497875,
+      "grad_norm": 4.97003173828125,
+      "learning_rate": 3.3714301183045382e-06,
+      "logits/chosen": 1.6978384256362915,
+      "logits/rejected": 1.1092720031738281,
+      "logps/chosen": -3.179011106491089,
+      "logps/rejected": -6.802459716796875,
+      "loss": 0.2942,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.768516540527344,
+      "rewards/margins": 5.435172080993652,
+      "rewards/rejected": -10.203688621520996,
+      "step": 580
+    },
+    {
+      "epoch": 0.5420925691971976,
+      "grad_norm": 4.482264995574951,
+      "learning_rate": 3.3221666168464584e-06,
+      "logits/chosen": 1.0245933532714844,
+      "logits/rejected": 0.7797524929046631,
+      "logps/chosen": -3.2393958568573,
+      "logps/rejected": -5.916412353515625,
+      "loss": 0.3347,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.859094142913818,
+      "rewards/margins": 4.015524864196777,
+      "rewards/rejected": -8.874618530273438,
+      "step": 590
+    },
+    {
+      "epoch": 0.5512805788446078,
+      "grad_norm": 3.913116931915283,
+      "learning_rate": 3.272542485937369e-06,
+      "logits/chosen": 1.3320618867874146,
+      "logits/rejected": 0.9762558937072754,
+      "logps/chosen": -3.2751450538635254,
+      "logps/rejected": -6.7808518409729,
+      "loss": 0.2776,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.912716865539551,
+      "rewards/margins": 5.258560657501221,
+      "rewards/rejected": -10.17127799987793,
+      "step": 600
+    },
+    {
+      "epoch": 0.5512805788446078,
+      "eval_logits/chosen": 0.9448758959770203,
+      "eval_logits/rejected": 0.42817041277885437,
+      "eval_logps/chosen": -2.835143804550171,
+      "eval_logps/rejected": -6.087582588195801,
+      "eval_loss": 0.28658536076545715,
+      "eval_rewards/accuracies": 0.9204545617103577,
+      "eval_rewards/chosen": -4.252715587615967,
+      "eval_rewards/margins": 4.878659248352051,
+      "eval_rewards/rejected": -9.13137435913086,
+      "eval_runtime": 24.4325,
+      "eval_samples_per_second": 28.814,
+      "eval_steps_per_second": 3.602,
+      "step": 600
+    },
+    {
+      "epoch": 0.560468588492018,
+      "grad_norm": 3.8335089683532715,
+      "learning_rate": 3.222579492361179e-06,
+      "logits/chosen": 1.295569658279419,
+      "logits/rejected": 1.0852326154708862,
+      "logps/chosen": -3.4870052337646484,
+      "logps/rejected": -6.572986602783203,
+      "loss": 0.3243,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.230508327484131,
+      "rewards/margins": 4.628971576690674,
+      "rewards/rejected": -9.859478950500488,
+      "step": 610
+    },
+    {
+      "epoch": 0.569656598139428,
+      "grad_norm": 3.095102071762085,
+      "learning_rate": 3.1722995515381644e-06,
+      "logits/chosen": 1.8339803218841553,
+      "logits/rejected": 1.467551589012146,
+      "logps/chosen": -3.6670470237731934,
+      "logps/rejected": -6.880410194396973,
+      "loss": 0.2615,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.500569820404053,
+      "rewards/margins": 4.820044040679932,
+      "rewards/rejected": -10.3206148147583,
+      "step": 620
+    },
+    {
+      "epoch": 0.5788446077868382,
+      "grad_norm": 4.381973743438721,
+      "learning_rate": 3.121724717912138e-06,
+      "logits/chosen": 2.390763521194458,
+      "logits/rejected": 2.155505418777466,
+      "logps/chosen": -3.3569788932800293,
+      "logps/rejected": -6.319291114807129,
+      "loss": 0.2649,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.035468101501465,
+      "rewards/margins": 4.4434685707092285,
+      "rewards/rejected": -9.478937149047852,
+      "step": 630
+    },
+    {
+      "epoch": 0.5880326174342483,
+      "grad_norm": 3.9204964637756348,
+      "learning_rate": 3.0708771752766397e-06,
+      "logits/chosen": 1.4343383312225342,
+      "logits/rejected": 1.1991338729858398,
+      "logps/chosen": -3.654259443283081,
+      "logps/rejected": -6.873109340667725,
+      "loss": 0.2761,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.481389045715332,
+      "rewards/margins": 4.828273773193359,
+      "rewards/rejected": -10.309663772583008,
+      "step": 640
+    },
+    {
+      "epoch": 0.5972206270816585,
+      "grad_norm": 3.179067373275757,
+      "learning_rate": 3.019779227044398e-06,
+      "logits/chosen": 2.0368576049804688,
+      "logits/rejected": 1.8037185668945312,
+      "logps/chosen": -3.791682720184326,
+      "logps/rejected": -6.912911415100098,
+      "loss": 0.2739,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.68752384185791,
+      "rewards/margins": 4.68184232711792,
+      "rewards/rejected": -10.369367599487305,
+      "step": 650
+    },
+    {
+      "epoch": 0.5972206270816585,
+      "eval_logits/chosen": 1.0889618396759033,
+      "eval_logits/rejected": 0.5887767672538757,
+      "eval_logps/chosen": -3.0191192626953125,
+      "eval_logps/rejected": -6.388964653015137,
+      "eval_loss": 0.27443525195121765,
+      "eval_rewards/accuracies": 0.9204545617103577,
+      "eval_rewards/chosen": -4.528679370880127,
+      "eval_rewards/margins": 5.054768085479736,
+      "eval_rewards/rejected": -9.583446502685547,
+      "eval_runtime": 24.4235,
+      "eval_samples_per_second": 28.825,
+      "eval_steps_per_second": 3.603,
+      "step": 650
+    },
+    {
+      "epoch": 0.6064086367290685,
+      "grad_norm": 2.804940700531006,
+      "learning_rate": 2.9684532864643123e-06,
+      "logits/chosen": 1.8347485065460205,
+      "logits/rejected": 1.3636573553085327,
+      "logps/chosen": -3.398799419403076,
+      "logps/rejected": -6.337627410888672,
+      "loss": 0.2678,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.098199367523193,
+      "rewards/margins": 4.408241271972656,
+      "rewards/rejected": -9.506441116333008,
+      "step": 660
+    },
+    {
+      "epoch": 0.6155966463764787,
+      "grad_norm": 3.7368969917297363,
+      "learning_rate": 2.9169218667902562e-06,
+      "logits/chosen": 1.6734691858291626,
+      "logits/rejected": 1.0945308208465576,
+      "logps/chosen": -3.354790210723877,
+      "logps/rejected": -6.3691864013671875,
+      "loss": 0.294,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.0321855545043945,
+      "rewards/margins": 4.521594524383545,
+      "rewards/rejected": -9.553780555725098,
+      "step": 670
+    },
+    {
+      "epoch": 0.6247846560238888,
+      "grad_norm": 4.483130931854248,
+      "learning_rate": 2.8652075714060296e-06,
+      "logits/chosen": 1.640228509902954,
+      "logits/rejected": 1.1519067287445068,
+      "logps/chosen": -3.4700570106506348,
+      "logps/rejected": -6.6276445388793945,
+      "loss": 0.23,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.205085277557373,
+      "rewards/margins": 4.736380577087402,
+      "rewards/rejected": -9.941465377807617,
+      "step": 680
+    },
+    {
+      "epoch": 0.633972665671299,
+      "grad_norm": 8.508203506469727,
+      "learning_rate": 2.813333083910761e-06,
+      "logits/chosen": 1.0501906871795654,
+      "logits/rejected": 0.5691097974777222,
+      "logps/chosen": -3.4982333183288574,
+      "logps/rejected": -7.144225120544434,
+      "loss": 0.2808,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.247350215911865,
+      "rewards/margins": 5.468987464904785,
+      "rewards/rejected": -10.716337203979492,
+      "step": 690
+    },
+    {
+      "epoch": 0.6431606753187091,
+      "grad_norm": 2.271857738494873,
+      "learning_rate": 2.761321158169134e-06,
+      "logits/chosen": 1.2119544744491577,
+      "logits/rejected": 0.7247776389122009,
+      "logps/chosen": -4.096956253051758,
+      "logps/rejected": -7.4536261558532715,
+      "loss": 0.2808,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -6.145434379577637,
+      "rewards/margins": 5.035003662109375,
+      "rewards/rejected": -11.180438995361328,
+      "step": 700
+    },
+    {
+      "epoch": 0.6431606753187091,
+      "eval_logits/chosen": 1.085008978843689,
+      "eval_logits/rejected": 0.5991834998130798,
+      "eval_logps/chosen": -3.179896354675293,
+      "eval_logps/rejected": -6.738409519195557,
+      "eval_loss": 0.26070258021354675,
+      "eval_rewards/accuracies": 0.9090909361839294,
+      "eval_rewards/chosen": -4.7698445320129395,
+      "eval_rewards/margins": 5.337769031524658,
+      "eval_rewards/rejected": -10.107613563537598,
+      "eval_runtime": 24.465,
+      "eval_samples_per_second": 28.776,
+      "eval_steps_per_second": 3.597,
+      "step": 700
+    },
+    {
+      "epoch": 0.6523486849661192,
+      "grad_norm": 4.058814525604248,
+      "learning_rate": 2.70919460833079e-06,
+      "logits/chosen": 1.2182409763336182,
+      "logits/rejected": 1.0481547117233276,
+      "logps/chosen": -3.532151699066162,
+      "logps/rejected": -7.4123854637146,
+      "loss": 0.2565,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.2982282638549805,
+      "rewards/margins": 5.8203511238098145,
+      "rewards/rejected": -11.118578910827637,
+      "step": 710
+    },
+    {
+      "epoch": 0.6615366946135294,
+      "grad_norm": 3.857574939727783,
+      "learning_rate": 2.6569762988232838e-06,
+      "logits/chosen": 0.8797380328178406,
+      "logits/rejected": 0.6432709693908691,
+      "logps/chosen": -3.3780128955841064,
+      "logps/rejected": -6.922214508056641,
+      "loss": 0.2321,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.067019462585449,
+      "rewards/margins": 5.3163018226623535,
+      "rewards/rejected": -10.383320808410645,
+      "step": 720
+    },
+    {
+      "epoch": 0.6707247042609394,
+      "grad_norm": 3.7125725746154785,
+      "learning_rate": 2.604689134322999e-06,
+      "logits/chosen": 1.447474479675293,
+      "logits/rejected": 1.1114810705184937,
+      "logps/chosen": -3.682945728302002,
+      "logps/rejected": -7.297152519226074,
+      "loss": 0.2812,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.524418830871582,
+      "rewards/margins": 5.421310901641846,
+      "rewards/rejected": -10.94572925567627,
+      "step": 730
+    },
+    {
+      "epoch": 0.6799127139083496,
+      "grad_norm": 2.026604652404785,
+      "learning_rate": 2.5523560497083927e-06,
+      "logits/chosen": 1.9674856662750244,
+      "logits/rejected": 1.593400239944458,
+      "logps/chosen": -3.3987834453582764,
+      "logps/rejected": -6.970278263092041,
+      "loss": 0.2603,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.098175048828125,
+      "rewards/margins": 5.357242107391357,
+      "rewards/rejected": -10.45541763305664,
+      "step": 740
+    },
+    {
+      "epoch": 0.6891007235557597,
+      "grad_norm": 2.357570171356201,
+      "learning_rate": 2.5e-06,
+      "logits/chosen": 1.9729163646697998,
+      "logits/rejected": 1.8791675567626953,
+      "logps/chosen": -3.3879425525665283,
+      "logps/rejected": -6.712514400482178,
+      "loss": 0.2348,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.081913948059082,
+      "rewards/margins": 4.986858367919922,
+      "rewards/rejected": -10.068772315979004,
+      "step": 750
+    },
+    {
+      "epoch": 0.6891007235557597,
+      "eval_logits/chosen": 1.1882920265197754,
+      "eval_logits/rejected": 0.7092404961585999,
+      "eval_logps/chosen": -2.855703592300415,
+      "eval_logps/rejected": -6.557665824890137,
+      "eval_loss": 0.25226154923439026,
+      "eval_rewards/accuracies": 0.9090909361839294,
+      "eval_rewards/chosen": -4.283555507659912,
+      "eval_rewards/margins": 5.552942276000977,
+      "eval_rewards/rejected": -9.836498260498047,
+      "eval_runtime": 24.4236,
+      "eval_samples_per_second": 28.825,
+      "eval_steps_per_second": 3.603,
+      "step": 750
+    },
+    {
+      "epoch": 0.6982887332031699,
+      "grad_norm": 3.1369588375091553,
+      "learning_rate": 2.447643950291608e-06,
+      "logits/chosen": 1.908418893814087,
+      "logits/rejected": 1.637202262878418,
+      "logps/chosen": -3.941102981567383,
+      "logps/rejected": -7.562623023986816,
+      "loss": 0.2412,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.911653995513916,
+      "rewards/margins": 5.432280540466309,
+      "rewards/rejected": -11.343935012817383,
+      "step": 760
+    },
+    {
+      "epoch": 0.70747674285058,
+      "grad_norm": 4.2312703132629395,
+      "learning_rate": 2.3953108656770018e-06,
+      "logits/chosen": 1.3125172853469849,
+      "logits/rejected": 0.933386504650116,
+      "logps/chosen": -3.5281143188476562,
+      "logps/rejected": -7.116488456726074,
+      "loss": 0.2528,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.292171001434326,
+      "rewards/margins": 5.382561683654785,
+      "rewards/rejected": -10.674734115600586,
+      "step": 770
+    },
+    {
+      "epoch": 0.7166647524979901,
+      "grad_norm": 5.6951799392700195,
+      "learning_rate": 2.3430237011767166e-06,
+      "logits/chosen": 2.295801877975464,
+      "logits/rejected": 1.8810745477676392,
+      "logps/chosen": -3.3887531757354736,
+      "logps/rejected": -6.8857245445251465,
+      "loss": 0.2672,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.083128929138184,
+      "rewards/margins": 5.245457649230957,
+      "rewards/rejected": -10.32858657836914,
+      "step": 780
+    },
+    {
+      "epoch": 0.7258527621454003,
+      "grad_norm": 3.734528064727783,
+      "learning_rate": 2.290805391669212e-06,
+      "logits/chosen": 1.6154979467391968,
+      "logits/rejected": 1.1886816024780273,
+      "logps/chosen": -3.379617214202881,
+      "logps/rejected": -7.1727166175842285,
+      "loss": 0.2315,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.069426536560059,
+      "rewards/margins": 5.689648628234863,
+      "rewards/rejected": -10.759074211120605,
+      "step": 790
+    },
+    {
+      "epoch": 0.7350407717928104,
+      "grad_norm": 3.2450063228607178,
+      "learning_rate": 2.238678841830867e-06,
+      "logits/chosen": 1.9270827770233154,
+      "logits/rejected": 1.5745903253555298,
+      "logps/chosen": -3.9242210388183594,
+      "logps/rejected": -7.254105567932129,
+      "loss": 0.2301,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.886331081390381,
+      "rewards/margins": 4.9948272705078125,
+      "rewards/rejected": -10.881157875061035,
+      "step": 800
+    },
+    {
+      "epoch": 0.7350407717928104,
+      "eval_logits/chosen": 1.228877067565918,
+      "eval_logits/rejected": 0.7780414819717407,
+      "eval_logps/chosen": -3.096149444580078,
+      "eval_logps/rejected": -6.9255452156066895,
+      "eval_loss": 0.23762211203575134,
+      "eval_rewards/accuracies": 0.9090909361839294,
+      "eval_rewards/chosen": -4.644224166870117,
+      "eval_rewards/margins": 5.744093894958496,
+      "eval_rewards/rejected": -10.388318061828613,
+      "eval_runtime": 24.4202,
+      "eval_samples_per_second": 28.829,
+      "eval_steps_per_second": 3.604,
+      "step": 800
+    },
+    {
+      "epoch": 0.7442287814402205,
+      "grad_norm": 3.6619718074798584,
+      "learning_rate": 2.186666916089239e-06,
+      "logits/chosen": 2.1697134971618652,
+      "logits/rejected": 1.8535985946655273,
+      "logps/chosen": -3.306549549102783,
+      "logps/rejected": -6.570471286773682,
+      "loss": 0.2239,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.959824562072754,
+      "rewards/margins": 4.895882606506348,
+      "rewards/rejected": -9.855707168579102,
+      "step": 810
+    },
+    {
+      "epoch": 0.7534167910876306,
+      "grad_norm": 3.9014956951141357,
+      "learning_rate": 2.134792428593971e-06,
+      "logits/chosen": 1.8228479623794556,
+      "logits/rejected": 1.5440882444381714,
+      "logps/chosen": -3.545902967453003,
+      "logps/rejected": -7.393707275390625,
+      "loss": 0.2429,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.318853855133057,
+      "rewards/margins": 5.7717061042785645,
+      "rewards/rejected": -11.090560913085938,
+      "step": 820
+    },
+    {
+      "epoch": 0.7626048007350408,
+      "grad_norm": 4.266237735748291,
+      "learning_rate": 2.0830781332097446e-06,
+      "logits/chosen": 1.6186097860336304,
+      "logits/rejected": 1.4948759078979492,
+      "logps/chosen": -3.6410841941833496,
+      "logps/rejected": -7.069480895996094,
+      "loss": 0.2213,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.461626052856445,
+      "rewards/margins": 5.142596244812012,
+      "rewards/rejected": -10.604223251342773,
+      "step": 830
+    },
+    {
+      "epoch": 0.771792810382451,
+      "grad_norm": 2.3189163208007812,
+      "learning_rate": 2.031546713535688e-06,
+      "logits/chosen": 2.4443275928497314,
+      "logits/rejected": 2.1987624168395996,
+      "logps/chosen": -4.127468585968018,
+      "logps/rejected": -8.15340805053711,
+      "loss": 0.2525,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.191202640533447,
+      "rewards/margins": 6.038909435272217,
+      "rewards/rejected": -12.230112075805664,
+      "step": 840
+    },
+    {
+      "epoch": 0.780980820029861,
+      "grad_norm": 5.227460861206055,
+      "learning_rate": 1.9802207729556023e-06,
+      "logits/chosen": 2.0057969093322754,
+      "logits/rejected": 1.7481991052627563,
+      "logps/chosen": -3.890474796295166,
+      "logps/rejected": -7.659049034118652,
+      "loss": 0.2253,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.83571195602417,
+      "rewards/margins": 5.65286111831665,
+      "rewards/rejected": -11.48857307434082,
+      "step": 850
+    },
+    {
+      "epoch": 0.780980820029861,
+      "eval_logits/chosen": 1.2305774688720703,
+      "eval_logits/rejected": 0.8189592957496643,
+      "eval_logps/chosen": -3.205787420272827,
+      "eval_logps/rejected": -7.230924606323242,
+      "eval_loss": 0.23406219482421875,
+      "eval_rewards/accuracies": 0.9090909361839294,
+      "eval_rewards/chosen": -4.808681488037109,
+      "eval_rewards/margins": 6.0377044677734375,
+      "eval_rewards/rejected": -10.846386909484863,
+      "eval_runtime": 24.5257,
+      "eval_samples_per_second": 28.705,
+      "eval_steps_per_second": 3.588,
+      "step": 850
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 1500,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "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": 2.0658307641628426e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-850/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35e758dd0e4a754b8adb96ec29d9d5e6d185c2cfb0f0a27add29180a481d3863
+size 7224

checkpoint-850/zero_to_fp32.py

@@ -0,0 +1,674 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example:
+#   python zero_to_fp32.py . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import json
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        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 tqdm(param_shapes.items(), desc='Gathering Sharded Weights'):
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # 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_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        state_dict_split = split_torch_state_dict_into_shards(state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard = {tensor: state_dict[tensor].contiguous() for tensor in tensors}
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard, output_path)
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument("output_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+                                               args.output_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)