Training in progress, step 1350, checkpoint

checkpoint-1350/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-1350/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": [
+    "o_proj",
+    "qkv_proj",
+    "gate_up_proj",
+    "down_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-1350/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8f055b87ef70f365b91f97073ca75c2fbab086f89ab95fe7b4feff2b0947df7
+size 25200088

checkpoint-1350/global_step1349/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:b95065078f7b664ce34d46033305e735b4c08527dd846e49daab0cc6ea1b6146
+size 18881328

checkpoint-1350/global_step1349/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a92123cee1eb75bda341a5730273ad23876cb637112a608c66c1a683e58dfd9
+size 18881328

checkpoint-1350/global_step1349/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69711d4f540af3931ec9f18e3cb245d42a1ec6eb26bcd8170a3e44023a81f185
+size 18881328

checkpoint-1350/global_step1349/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:d97d84b72d891b3e5a06f4a9d6f04d2b52dc94d6584a940f09f70317ae977ac3
+size 18881392

checkpoint-1350/global_step1349/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ef3cd2c0663ac3d568ccbe108e637a1e6317b1029ffc9ea21c44977398c2eef
+size 18881392

checkpoint-1350/global_step1349/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:827ea9ee51035cb6869d76128afec5cc556b7b7e18e58e34d0a87835230dfba5
+size 18881392

checkpoint-1350/global_step1349/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:019160ff7bf38ca4db262c7e898b4961995cda7e67f83a77411e43ae2c0dd506
+size 18881392

checkpoint-1350/global_step1349/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:c4cca50f2d6bc1e4c1e3536b0e41b421aef3f74c4acbd3a01ed7fe9fc61f4434
+size 18881392

checkpoint-1350/global_step1349/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6564e3bf722e5e1ae0a8b7d080d87de28a4d158d5b9ef842438b9a15b6c79473
+size 25379244

checkpoint-1350/latest

@@ -0,0 +1 @@
+global_step1349

checkpoint-1350/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3765ef8bd773a01f33d6189c28cd8adc0806701ad4be0ce173a432a02a00453e
+size 15984

checkpoint-1350/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:125d82e66475994a03198cccbb6d6b3f3f0b90e87e3ad8c2ce9dbc11dc5b5675
+size 15984

checkpoint-1350/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e24861fad76311d7a147f5f52607d98bf3fe96ea063b7dfbf1a9905199dc89ae
+size 15984

checkpoint-1350/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8432297426aec314a0edbfee7e2587547675e3d1226f2da8c86342b76733a35f
+size 15984

checkpoint-1350/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11a36c96a40ebc1c664f769f1365b8d2840185512dee05598f8dd28eb82719f9
+size 15984

checkpoint-1350/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46e34171101846c7396fec0bef25165e1e31cb999d62d6f33db7d94a5baf824b
+size 15984

checkpoint-1350/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:acddb7d75b7cda479865f424b489b16f4c79a394ba6912c4308d7d45b8bef340
+size 15984

checkpoint-1350/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec8aca039dd9da6e1b7a6a6945c64efd596a772ca3e5cf3eb06d07471446b4aa
+size 15984

checkpoint-1350/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b94062a46361d2ca678aa6132530caba96001618ab107f09c919e21fc386bb76
+size 1064

checkpoint-1350/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-1350/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-1350/trainer_state.json

@@ -0,0 +1,2490 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.7136507936507936,
+  "eval_steps": 50,
+  "global_step": 1350,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.012698412698412698,
+      "grad_norm": 0.04658036306500435,
+      "learning_rate": 4.999451708687114e-06,
+      "logits/chosen": 15.106437683105469,
+      "logits/rejected": 15.158523559570312,
+      "logps/chosen": -0.30069679021835327,
+      "logps/rejected": -0.30243945121765137,
+      "loss": 0.9981,
+      "rewards/accuracies": 0.3499999940395355,
+      "rewards/chosen": -0.45104512572288513,
+      "rewards/margins": 0.002613987773656845,
+      "rewards/rejected": -0.45365914702415466,
+      "step": 10
+    },
+    {
+      "epoch": 0.025396825396825397,
+      "grad_norm": 0.05173320695757866,
+      "learning_rate": 4.997807075247147e-06,
+      "logits/chosen": 14.680102348327637,
+      "logits/rejected": 14.592491149902344,
+      "logps/chosen": -0.3225177228450775,
+      "logps/rejected": -0.27230435609817505,
+      "loss": 0.9967,
+      "rewards/accuracies": 0.26249998807907104,
+      "rewards/chosen": -0.4837765693664551,
+      "rewards/margins": -0.07532001286745071,
+      "rewards/rejected": -0.40845656394958496,
+      "step": 20
+    },
+    {
+      "epoch": 0.0380952380952381,
+      "grad_norm": 0.04962443560361862,
+      "learning_rate": 4.9950668210706795e-06,
+      "logits/chosen": 15.399703979492188,
+      "logits/rejected": 15.089459419250488,
+      "logps/chosen": -0.2997470498085022,
+      "logps/rejected": -0.28447264432907104,
+      "loss": 0.9988,
+      "rewards/accuracies": 0.3125,
+      "rewards/chosen": -0.4496205747127533,
+      "rewards/margins": -0.022911589592695236,
+      "rewards/rejected": -0.42670899629592896,
+      "step": 30
+    },
+    {
+      "epoch": 0.050793650793650794,
+      "grad_norm": 0.05171326920390129,
+      "learning_rate": 4.9912321481237616e-06,
+      "logits/chosen": 15.099847793579102,
+      "logits/rejected": 15.159170150756836,
+      "logps/chosen": -0.29869550466537476,
+      "logps/rejected": -0.2806033790111542,
+      "loss": 0.9841,
+      "rewards/accuracies": 0.375,
+      "rewards/chosen": -0.44804325699806213,
+      "rewards/margins": -0.02713816799223423,
+      "rewards/rejected": -0.42090511322021484,
+      "step": 40
+    },
+    {
+      "epoch": 0.06349206349206349,
+      "grad_norm": 0.048664532601833344,
+      "learning_rate": 4.986304738420684e-06,
+      "logits/chosen": 15.207334518432617,
+      "logits/rejected": 15.135488510131836,
+      "logps/chosen": -0.3121686577796936,
+      "logps/rejected": -0.3194735050201416,
+      "loss": 0.9953,
+      "rewards/accuracies": 0.3499999940395355,
+      "rewards/chosen": -0.468252956867218,
+      "rewards/margins": 0.010957291349768639,
+      "rewards/rejected": -0.47921022772789,
+      "step": 50
+    },
+    {
+      "epoch": 0.06349206349206349,
+      "eval_logits/chosen": 15.441752433776855,
+      "eval_logits/rejected": 15.288756370544434,
+      "eval_logps/chosen": -0.33557233214378357,
+      "eval_logps/rejected": -0.31198158860206604,
+      "eval_loss": 0.9914231300354004,
+      "eval_rewards/accuracies": 0.3125,
+      "eval_rewards/chosen": -0.5033585429191589,
+      "eval_rewards/margins": -0.03538615256547928,
+      "eval_rewards/rejected": -0.46797239780426025,
+      "eval_runtime": 19.0844,
+      "eval_samples_per_second": 26.723,
+      "eval_steps_per_second": 3.354,
+      "step": 50
+    },
+    {
+      "epoch": 0.0761904761904762,
+      "grad_norm": 0.0722479596734047,
+      "learning_rate": 4.980286753286196e-06,
+      "logits/chosen": 15.281786918640137,
+      "logits/rejected": 15.110156059265137,
+      "logps/chosen": -0.30892473459243774,
+      "logps/rejected": -0.26251715421676636,
+      "loss": 1.0022,
+      "rewards/accuracies": 0.25,
+      "rewards/chosen": -0.4633871018886566,
+      "rewards/margins": -0.0696113258600235,
+      "rewards/rejected": -0.3937757611274719,
+      "step": 60
+    },
+    {
+      "epoch": 0.08888888888888889,
+      "grad_norm": 0.07008200883865356,
+      "learning_rate": 4.973180832407471e-06,
+      "logits/chosen": 14.836176872253418,
+      "logits/rejected": 14.98499584197998,
+      "logps/chosen": -0.29483428597450256,
+      "logps/rejected": -0.29870957136154175,
+      "loss": 0.9849,
+      "rewards/accuracies": 0.4000000059604645,
+      "rewards/chosen": -0.44225144386291504,
+      "rewards/margins": 0.0058129094541072845,
+      "rewards/rejected": -0.44806432723999023,
+      "step": 70
+    },
+    {
+      "epoch": 0.10158730158730159,
+      "grad_norm": 0.06503555178642273,
+      "learning_rate": 4.964990092676263e-06,
+      "logits/chosen": 15.508198738098145,
+      "logits/rejected": 15.592549324035645,
+      "logps/chosen": -0.3066270351409912,
+      "logps/rejected": -0.2757572531700134,
+      "loss": 0.9886,
+      "rewards/accuracies": 0.3125,
+      "rewards/chosen": -0.45994052290916443,
+      "rewards/margins": -0.046304650604724884,
+      "rewards/rejected": -0.41363590955734253,
+      "step": 80
+    },
+    {
+      "epoch": 0.11428571428571428,
+      "grad_norm": 0.10515156388282776,
+      "learning_rate": 4.9557181268217225e-06,
+      "logits/chosen": 15.301602363586426,
+      "logits/rejected": 15.421157836914062,
+      "logps/chosen": -0.31223705410957336,
+      "logps/rejected": -0.29449179768562317,
+      "loss": 0.981,
+      "rewards/accuracies": 0.2750000059604645,
+      "rewards/chosen": -0.46835556626319885,
+      "rewards/margins": -0.02661792002618313,
+      "rewards/rejected": -0.44173765182495117,
+      "step": 90
+    },
+    {
+      "epoch": 0.12698412698412698,
+      "grad_norm": 0.05785346403717995,
+      "learning_rate": 4.9453690018345144e-06,
+      "logits/chosen": 16.00307846069336,
+      "logits/rejected": 15.64977741241455,
+      "logps/chosen": -0.32283931970596313,
+      "logps/rejected": -0.28576889634132385,
+      "loss": 0.9883,
+      "rewards/accuracies": 0.2750000059604645,
+      "rewards/chosen": -0.4842589795589447,
+      "rewards/margins": -0.055605631321668625,
+      "rewards/rejected": -0.42865338921546936,
+      "step": 100
+    },
+    {
+      "epoch": 0.12698412698412698,
+      "eval_logits/chosen": 15.624425888061523,
+      "eval_logits/rejected": 15.548928260803223,
+      "eval_logps/chosen": -0.33450835943222046,
+      "eval_logps/rejected": -0.31935107707977295,
+      "eval_loss": 0.9836427569389343,
+      "eval_rewards/accuracies": 0.359375,
+      "eval_rewards/chosen": -0.5017625689506531,
+      "eval_rewards/margins": -0.022735953330993652,
+      "eval_rewards/rejected": -0.4790266156196594,
+      "eval_runtime": 18.6768,
+      "eval_samples_per_second": 27.307,
+      "eval_steps_per_second": 3.427,
+      "step": 100
+    },
+    {
+      "epoch": 0.13968253968253969,
+      "grad_norm": 0.08603859692811966,
+      "learning_rate": 4.933947257182901e-06,
+      "logits/chosen": 15.40850830078125,
+      "logits/rejected": 15.258935928344727,
+      "logps/chosen": -0.3209790587425232,
+      "logps/rejected": -0.29926618933677673,
+      "loss": 0.9852,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": -0.4814685881137848,
+      "rewards/margins": -0.03256931155920029,
+      "rewards/rejected": -0.4488992691040039,
+      "step": 110
+    },
+    {
+      "epoch": 0.1523809523809524,
+      "grad_norm": 0.13407552242279053,
+      "learning_rate": 4.921457902821578e-06,
+      "logits/chosen": 15.610095024108887,
+      "logits/rejected": 15.964601516723633,
+      "logps/chosen": -0.2977743446826935,
+      "logps/rejected": -0.3102283179759979,
+      "loss": 0.9839,
+      "rewards/accuracies": 0.38749998807907104,
+      "rewards/chosen": -0.4466615617275238,
+      "rewards/margins": 0.018680967390537262,
+      "rewards/rejected": -0.4653424620628357,
+      "step": 120
+    },
+    {
+      "epoch": 0.16507936507936508,
+      "grad_norm": 0.1397980898618698,
+      "learning_rate": 4.907906416994146e-06,
+      "logits/chosen": 15.487627983093262,
+      "logits/rejected": 15.767982482910156,
+      "logps/chosen": -0.2769243121147156,
+      "logps/rejected": -0.3168947100639343,
+      "loss": 0.9737,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.415386438369751,
+      "rewards/margins": 0.05995568633079529,
+      "rewards/rejected": -0.47534212470054626,
+      "step": 130
+    },
+    {
+      "epoch": 0.17777777777777778,
+      "grad_norm": 0.09849797189235687,
+      "learning_rate": 4.893298743830168e-06,
+      "logits/chosen": 15.779914855957031,
+      "logits/rejected": 15.66816234588623,
+      "logps/chosen": -0.2959491014480591,
+      "logps/rejected": -0.3029848635196686,
+      "loss": 0.9804,
+      "rewards/accuracies": 0.375,
+      "rewards/chosen": -0.4439236521720886,
+      "rewards/margins": 0.010553586296737194,
+      "rewards/rejected": -0.4544772207736969,
+      "step": 140
+    },
+    {
+      "epoch": 0.19047619047619047,
+      "grad_norm": 0.08089074492454529,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": 15.029818534851074,
+      "logits/rejected": 15.431653022766113,
+      "logps/chosen": -0.2956623435020447,
+      "logps/rejected": -0.3162347376346588,
+      "loss": 0.9728,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.443493515253067,
+      "rewards/margins": 0.030858617275953293,
+      "rewards/rejected": -0.4743521809577942,
+      "step": 150
+    },
+    {
+      "epoch": 0.19047619047619047,
+      "eval_logits/chosen": 15.575506210327148,
+      "eval_logits/rejected": 15.54050064086914,
+      "eval_logps/chosen": -0.3363308906555176,
+      "eval_logps/rejected": -0.3436908721923828,
+      "eval_loss": 0.9609583616256714,
+      "eval_rewards/accuracies": 0.4375,
+      "eval_rewards/chosen": -0.5044962763786316,
+      "eval_rewards/margins": 0.01103996392339468,
+      "eval_rewards/rejected": -0.5155363082885742,
+      "eval_runtime": 18.6082,
+      "eval_samples_per_second": 27.407,
+      "eval_steps_per_second": 3.439,
+      "step": 150
+    },
+    {
+      "epoch": 0.20317460317460317,
+      "grad_norm": 0.12168499082326889,
+      "learning_rate": 4.860940925593703e-06,
+      "logits/chosen": 15.47050952911377,
+      "logits/rejected": 15.637664794921875,
+      "logps/chosen": -0.32601848244667053,
+      "logps/rejected": -0.35739919543266296,
+      "loss": 0.9531,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.489027738571167,
+      "rewards/margins": 0.04707105830311775,
+      "rewards/rejected": -0.5360987782478333,
+      "step": 160
+    },
+    {
+      "epoch": 0.21587301587301588,
+      "grad_norm": 0.12861700356006622,
+      "learning_rate": 4.84320497372973e-06,
+      "logits/chosen": 15.594339370727539,
+      "logits/rejected": 15.680140495300293,
+      "logps/chosen": -0.29271024465560913,
+      "logps/rejected": -0.3383347690105438,
+      "loss": 0.9386,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.4390653669834137,
+      "rewards/margins": 0.06843684613704681,
+      "rewards/rejected": -0.5075021982192993,
+      "step": 170
+    },
+    {
+      "epoch": 0.22857142857142856,
+      "grad_norm": 0.10320646315813065,
+      "learning_rate": 4.824441214720629e-06,
+      "logits/chosen": 15.644658088684082,
+      "logits/rejected": 15.526695251464844,
+      "logps/chosen": -0.30950039625167847,
+      "logps/rejected": -0.33660295605659485,
+      "loss": 0.9313,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": -0.46425050497055054,
+      "rewards/margins": 0.04065385088324547,
+      "rewards/rejected": -0.5049043893814087,
+      "step": 180
+    },
+    {
+      "epoch": 0.24126984126984127,
+      "grad_norm": 0.29178574681282043,
+      "learning_rate": 4.804657878971252e-06,
+      "logits/chosen": 15.795066833496094,
+      "logits/rejected": 15.640788078308105,
+      "logps/chosen": -0.34224197268486023,
+      "logps/rejected": -0.34523850679397583,
+      "loss": 0.9406,
+      "rewards/accuracies": 0.375,
+      "rewards/chosen": -0.5133630037307739,
+      "rewards/margins": 0.0044947536662220955,
+      "rewards/rejected": -0.5178577303886414,
+      "step": 190
+    },
+    {
+      "epoch": 0.25396825396825395,
+      "grad_norm": 0.15747429430484772,
+      "learning_rate": 4.783863644106502e-06,
+      "logits/chosen": 15.24070930480957,
+      "logits/rejected": 15.132087707519531,
+      "logps/chosen": -0.3026728630065918,
+      "logps/rejected": -0.3765440583229065,
+      "loss": 0.9031,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.4540092945098877,
+      "rewards/margins": 0.11080671846866608,
+      "rewards/rejected": -0.5648160576820374,
+      "step": 200
+    },
+    {
+      "epoch": 0.25396825396825395,
+      "eval_logits/chosen": 15.073077201843262,
+      "eval_logits/rejected": 15.098322868347168,
+      "eval_logps/chosen": -0.3540771007537842,
+      "eval_logps/rejected": -0.41381165385246277,
+      "eval_loss": 0.9153187274932861,
+      "eval_rewards/accuracies": 0.5,
+      "eval_rewards/chosen": -0.5311156511306763,
+      "eval_rewards/margins": 0.08960187435150146,
+      "eval_rewards/rejected": -0.6207175254821777,
+      "eval_runtime": 18.5936,
+      "eval_samples_per_second": 27.429,
+      "eval_steps_per_second": 3.442,
+      "step": 200
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 0.9226244688034058,
+      "learning_rate": 4.762067631165049e-06,
+      "logits/chosen": 15.300783157348633,
+      "logits/rejected": 15.6528902053833,
+      "logps/chosen": -0.3110392093658447,
+      "logps/rejected": -0.4790540635585785,
+      "loss": 0.8977,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.4665588438510895,
+      "rewards/margins": 0.25202232599258423,
+      "rewards/rejected": -0.7185810804367065,
+      "step": 210
+    },
+    {
+      "epoch": 0.27936507936507937,
+      "grad_norm": 0.187363401055336,
+      "learning_rate": 4.7392794005985324e-06,
+      "logits/chosen": 15.123028755187988,
+      "logits/rejected": 14.940861701965332,
+      "logps/chosen": -0.33024150133132935,
+      "logps/rejected": -0.35755541920661926,
+      "loss": 0.9074,
+      "rewards/accuracies": 0.42500001192092896,
+      "rewards/chosen": -0.49536222219467163,
+      "rewards/margins": 0.04097090661525726,
+      "rewards/rejected": -0.5363331437110901,
+      "step": 220
+    },
+    {
+      "epoch": 0.2920634920634921,
+      "grad_norm": 0.27345994114875793,
+      "learning_rate": 4.715508948078037e-06,
+      "logits/chosen": 14.54762077331543,
+      "logits/rejected": 14.207303047180176,
+      "logps/chosen": -0.2951691150665283,
+      "logps/rejected": -0.41360992193222046,
+      "loss": 0.896,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.4427536427974701,
+      "rewards/margins": 0.1776612401008606,
+      "rewards/rejected": -0.6204149723052979,
+      "step": 230
+    },
+    {
+      "epoch": 0.3047619047619048,
+      "grad_norm": 0.20160575211048126,
+      "learning_rate": 4.690766700109659e-06,
+      "logits/chosen": 14.768750190734863,
+      "logits/rejected": 14.940885543823242,
+      "logps/chosen": -0.3044833838939667,
+      "logps/rejected": -0.4275297224521637,
+      "loss": 0.891,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.4567251205444336,
+      "rewards/margins": 0.18456946313381195,
+      "rewards/rejected": -0.6412945985794067,
+      "step": 240
+    },
+    {
+      "epoch": 0.31746031746031744,
+      "grad_norm": 0.9059060215950012,
+      "learning_rate": 4.665063509461098e-06,
+      "logits/chosen": 14.839933395385742,
+      "logits/rejected": 14.69981861114502,
+      "logps/chosen": -0.3137063980102539,
+      "logps/rejected": -0.49661844968795776,
+      "loss": 0.8621,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.47055959701538086,
+      "rewards/margins": 0.2743679881095886,
+      "rewards/rejected": -0.7449275851249695,
+      "step": 250
+    },
+    {
+      "epoch": 0.31746031746031744,
+      "eval_logits/chosen": 14.19374942779541,
+      "eval_logits/rejected": 14.215425491333008,
+      "eval_logps/chosen": -0.3999108076095581,
+      "eval_logps/rejected": -0.7892026305198669,
+      "eval_loss": 0.7948001623153687,
+      "eval_rewards/accuracies": 0.53125,
+      "eval_rewards/chosen": -0.5998662114143372,
+      "eval_rewards/margins": 0.5839377641677856,
+      "eval_rewards/rejected": -1.1838040351867676,
+      "eval_runtime": 18.6098,
+      "eval_samples_per_second": 27.405,
+      "eval_steps_per_second": 3.439,
+      "step": 250
+    },
+    {
+      "epoch": 0.33015873015873015,
+      "grad_norm": 0.8104033470153809,
+      "learning_rate": 4.638410650401267e-06,
+      "logits/chosen": 14.144885063171387,
+      "logits/rejected": 14.350593566894531,
+      "logps/chosen": -0.3469873368740082,
+      "logps/rejected": -0.8288809657096863,
+      "loss": 0.7939,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.5204810500144958,
+      "rewards/margins": 0.7228401899337769,
+      "rewards/rejected": -1.243321418762207,
+      "step": 260
+    },
+    {
+      "epoch": 0.34285714285714286,
+      "grad_norm": 0.39150306582450867,
+      "learning_rate": 4.610819813755038e-06,
+      "logits/chosen": 14.15583324432373,
+      "logits/rejected": 13.783352851867676,
+      "logps/chosen": -0.34353378415107727,
+      "logps/rejected": -1.1232259273529053,
+      "loss": 0.7635,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.5153006911277771,
+      "rewards/margins": 1.1695382595062256,
+      "rewards/rejected": -1.684838891029358,
+      "step": 270
+    },
+    {
+      "epoch": 0.35555555555555557,
+      "grad_norm": 0.1740872859954834,
+      "learning_rate": 4.582303101775249e-06,
+      "logits/chosen": 13.705289840698242,
+      "logits/rejected": 13.493337631225586,
+      "logps/chosen": -0.35998308658599854,
+      "logps/rejected": -1.3943986892700195,
+      "loss": 0.7678,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.539974570274353,
+      "rewards/margins": 1.5516235828399658,
+      "rewards/rejected": -2.0915980339050293,
+      "step": 280
+    },
+    {
+      "epoch": 0.3682539682539683,
+      "grad_norm": 0.22531260550022125,
+      "learning_rate": 4.55287302283426e-06,
+      "logits/chosen": 13.186914443969727,
+      "logits/rejected": 13.140413284301758,
+      "logps/chosen": -0.3548193573951721,
+      "logps/rejected": -1.2321991920471191,
+      "loss": 0.7719,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.5322290062904358,
+      "rewards/margins": 1.3160697221755981,
+      "rewards/rejected": -1.8482987880706787,
+      "step": 290
+    },
+    {
+      "epoch": 0.38095238095238093,
+      "grad_norm": 0.5229180455207825,
+      "learning_rate": 4.522542485937369e-06,
+      "logits/chosen": 13.780011177062988,
+      "logits/rejected": 13.604715347290039,
+      "logps/chosen": -0.4277075231075287,
+      "logps/rejected": -1.415838599205017,
+      "loss": 0.7644,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.6415613293647766,
+      "rewards/margins": 1.482196569442749,
+      "rewards/rejected": -2.123757839202881,
+      "step": 300
+    },
+    {
+      "epoch": 0.38095238095238093,
+      "eval_logits/chosen": 13.13498592376709,
+      "eval_logits/rejected": 13.13513469696045,
+      "eval_logps/chosen": -0.476482629776001,
+      "eval_logps/rejected": -1.4701811075210571,
+      "eval_loss": 0.7514793872833252,
+      "eval_rewards/accuracies": 0.546875,
+      "eval_rewards/chosen": -0.7147239446640015,
+      "eval_rewards/margins": 1.4905478954315186,
+      "eval_rewards/rejected": -2.2052717208862305,
+      "eval_runtime": 18.6093,
+      "eval_samples_per_second": 27.406,
+      "eval_steps_per_second": 3.439,
+      "step": 300
+    },
+    {
+      "epoch": 0.39365079365079364,
+      "grad_norm": 0.7556002736091614,
+      "learning_rate": 4.491324795060491e-06,
+      "logits/chosen": 13.26073932647705,
+      "logits/rejected": 13.28388786315918,
+      "logps/chosen": -0.39378833770751953,
+      "logps/rejected": -1.4242979288101196,
+      "loss": 0.7092,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.5906823873519897,
+      "rewards/margins": 1.5457642078399658,
+      "rewards/rejected": -2.136446714401245,
+      "step": 310
+    },
+    {
+      "epoch": 0.40634920634920635,
+      "grad_norm": 0.8167753219604492,
+      "learning_rate": 4.4592336433146e-06,
+      "logits/chosen": 12.952977180480957,
+      "logits/rejected": 12.89118480682373,
+      "logps/chosen": -0.4167153835296631,
+      "logps/rejected": -1.1332799196243286,
+      "loss": 0.7364,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.6250730752944946,
+      "rewards/margins": 1.074846625328064,
+      "rewards/rejected": -1.6999199390411377,
+      "step": 320
+    },
+    {
+      "epoch": 0.41904761904761906,
+      "grad_norm": 2.9806692600250244,
+      "learning_rate": 4.426283106939474e-06,
+      "logits/chosen": 12.890368461608887,
+      "logits/rejected": 12.765925407409668,
+      "logps/chosen": -0.5025959014892578,
+      "logps/rejected": -1.2589428424835205,
+      "loss": 0.748,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.7538937926292419,
+      "rewards/margins": 1.1345205307006836,
+      "rewards/rejected": -1.8884143829345703,
+      "step": 330
+    },
+    {
+      "epoch": 0.43174603174603177,
+      "grad_norm": 1.568097710609436,
+      "learning_rate": 4.3924876391293915e-06,
+      "logits/chosen": 13.032608032226562,
+      "logits/rejected": 12.877195358276367,
+      "logps/chosen": -0.5034081935882568,
+      "logps/rejected": -1.585137963294983,
+      "loss": 0.7516,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.7551122903823853,
+      "rewards/margins": 1.6225944757461548,
+      "rewards/rejected": -2.37770676612854,
+      "step": 340
+    },
+    {
+      "epoch": 0.4444444444444444,
+      "grad_norm": 0.5996735095977783,
+      "learning_rate": 4.357862063693486e-06,
+      "logits/chosen": 12.859817504882812,
+      "logits/rejected": 12.712678909301758,
+      "logps/chosen": -0.5043476819992065,
+      "logps/rejected": -1.525444507598877,
+      "loss": 0.7497,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.7565216422080994,
+      "rewards/margins": 1.5316450595855713,
+      "rewards/rejected": -2.2881667613983154,
+      "step": 350
+    },
+    {
+      "epoch": 0.4444444444444444,
+      "eval_logits/chosen": 12.228137969970703,
+      "eval_logits/rejected": 12.226001739501953,
+      "eval_logps/chosen": -0.5828607082366943,
+      "eval_logps/rejected": -1.6681612730026245,
+      "eval_loss": 0.7238383889198303,
+      "eval_rewards/accuracies": 0.6875,
+      "eval_rewards/chosen": -0.8742910623550415,
+      "eval_rewards/margins": 1.6279507875442505,
+      "eval_rewards/rejected": -2.502241611480713,
+      "eval_runtime": 18.6038,
+      "eval_samples_per_second": 27.414,
+      "eval_steps_per_second": 3.44,
+      "step": 350
+    },
+    {
+      "epoch": 0.45714285714285713,
+      "grad_norm": 0.605993926525116,
+      "learning_rate": 4.322421568553529e-06,
+      "logits/chosen": 11.993739128112793,
+      "logits/rejected": 11.75650691986084,
+      "logps/chosen": -0.5674928426742554,
+      "logps/rejected": -1.7509374618530273,
+      "loss": 0.7241,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.8512393236160278,
+      "rewards/margins": 1.7751665115356445,
+      "rewards/rejected": -2.626405954360962,
+      "step": 360
+    },
+    {
+      "epoch": 0.46984126984126984,
+      "grad_norm": 0.931057870388031,
+      "learning_rate": 4.286181699082008e-06,
+      "logits/chosen": 11.784490585327148,
+      "logits/rejected": 12.052295684814453,
+      "logps/chosen": -0.5866945385932922,
+      "logps/rejected": -1.8955312967300415,
+      "loss": 0.7141,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.8800417184829712,
+      "rewards/margins": 1.9632551670074463,
+      "rewards/rejected": -2.843297243118286,
+      "step": 370
+    },
+    {
+      "epoch": 0.48253968253968255,
+      "grad_norm": 1.3936405181884766,
+      "learning_rate": 4.249158351283414e-06,
+      "logits/chosen": 11.842119216918945,
+      "logits/rejected": 11.340182304382324,
+      "logps/chosen": -0.7804869413375854,
+      "logps/rejected": -1.8759396076202393,
+      "loss": 0.6654,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -1.1707303524017334,
+      "rewards/margins": 1.6431787014007568,
+      "rewards/rejected": -2.8139090538024902,
+      "step": 380
+    },
+    {
+      "epoch": 0.49523809523809526,
+      "grad_norm": 1.737855076789856,
+      "learning_rate": 4.211367764821722e-06,
+      "logits/chosen": 11.465726852416992,
+      "logits/rejected": 11.05290699005127,
+      "logps/chosen": -1.3201282024383545,
+      "logps/rejected": -2.3962795734405518,
+      "loss": 0.6301,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -1.9801921844482422,
+      "rewards/margins": 1.6142269372940063,
+      "rewards/rejected": -3.594419002532959,
+      "step": 390
+    },
+    {
+      "epoch": 0.5079365079365079,
+      "grad_norm": 1.6870065927505493,
+      "learning_rate": 4.172826515897146e-06,
+      "logits/chosen": 10.830609321594238,
+      "logits/rejected": 10.663077354431152,
+      "logps/chosen": -2.5642189979553223,
+      "logps/rejected": -3.535013198852539,
+      "loss": 0.6274,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -3.8463282585144043,
+      "rewards/margins": 1.4561914205551147,
+      "rewards/rejected": -5.302519798278809,
+      "step": 400
+    },
+    {
+      "epoch": 0.5079365079365079,
+      "eval_logits/chosen": 9.964677810668945,
+      "eval_logits/rejected": 9.858954429626465,
+      "eval_logps/chosen": -2.578787088394165,
+      "eval_logps/rejected": -3.803541898727417,
+      "eval_loss": 0.590033233165741,
+      "eval_rewards/accuracies": 0.796875,
+      "eval_rewards/chosen": -3.868180513381958,
+      "eval_rewards/margins": 1.837132453918457,
+      "eval_rewards/rejected": -5.705312728881836,
+      "eval_runtime": 18.5898,
+      "eval_samples_per_second": 27.434,
+      "eval_steps_per_second": 3.443,
+      "step": 400
+    },
+    {
+      "epoch": 0.5206349206349207,
+      "grad_norm": 2.0417189598083496,
+      "learning_rate": 4.133551509975264e-06,
+      "logits/chosen": 9.34677505493164,
+      "logits/rejected": 9.576300621032715,
+      "logps/chosen": -2.1631455421447754,
+      "logps/rejected": -3.024636745452881,
+      "loss": 0.6309,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -3.244718074798584,
+      "rewards/margins": 1.2922370433807373,
+      "rewards/rejected": -4.536954879760742,
+      "step": 410
+    },
+    {
+      "epoch": 0.5333333333333333,
+      "grad_norm": 2.9144859313964844,
+      "learning_rate": 4.093559974371725e-06,
+      "logits/chosen": 9.20117473602295,
+      "logits/rejected": 9.481060028076172,
+      "logps/chosen": -2.7644081115722656,
+      "logps/rejected": -3.905733585357666,
+      "loss": 0.5839,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -4.14661169052124,
+      "rewards/margins": 1.7119888067245483,
+      "rewards/rejected": -5.858600616455078,
+      "step": 420
+    },
+    {
+      "epoch": 0.546031746031746,
+      "grad_norm": 2.4795055389404297,
+      "learning_rate": 4.052869450695776e-06,
+      "logits/chosen": 9.674077987670898,
+      "logits/rejected": 9.64409065246582,
+      "logps/chosen": -2.562514066696167,
+      "logps/rejected": -3.8458714485168457,
+      "loss": 0.5266,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -3.843771457672119,
+      "rewards/margins": 1.9250361919403076,
+      "rewards/rejected": -5.768807411193848,
+      "step": 430
+    },
+    {
+      "epoch": 0.5587301587301587,
+      "grad_norm": 1.897057056427002,
+      "learning_rate": 4.011497787155938e-06,
+      "logits/chosen": 8.887510299682617,
+      "logits/rejected": 8.687074661254883,
+      "logps/chosen": -3.2801125049591064,
+      "logps/rejected": -4.673043251037598,
+      "loss": 0.5665,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -4.920168876647949,
+      "rewards/margins": 2.089395523071289,
+      "rewards/rejected": -7.009564399719238,
+      "step": 440
+    },
+    {
+      "epoch": 0.5714285714285714,
+      "grad_norm": 2.1004762649536133,
+      "learning_rate": 3.969463130731183e-06,
+      "logits/chosen": 8.606618881225586,
+      "logits/rejected": 8.299476623535156,
+      "logps/chosen": -3.3771705627441406,
+      "logps/rejected": -4.999676704406738,
+      "loss": 0.5118,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -5.0657548904418945,
+      "rewards/margins": 2.433760166168213,
+      "rewards/rejected": -7.499515533447266,
+      "step": 450
+    },
+    {
+      "epoch": 0.5714285714285714,
+      "eval_logits/chosen": 8.417736053466797,
+      "eval_logits/rejected": 8.183667182922363,
+      "eval_logps/chosen": -3.3982253074645996,
+      "eval_logps/rejected": -5.010103225708008,
+      "eval_loss": 0.5201926827430725,
+      "eval_rewards/accuracies": 0.796875,
+      "eval_rewards/chosen": -5.0973381996154785,
+      "eval_rewards/margins": 2.4178173542022705,
+      "eval_rewards/rejected": -7.515154838562012,
+      "eval_runtime": 18.6069,
+      "eval_samples_per_second": 27.409,
+      "eval_steps_per_second": 3.44,
+      "step": 450
+    },
+    {
+      "epoch": 0.5841269841269842,
+      "grad_norm": 2.016348123550415,
+      "learning_rate": 3.92678391921108e-06,
+      "logits/chosen": 9.348031997680664,
+      "logits/rejected": 8.707467079162598,
+      "logps/chosen": -3.687103271484375,
+      "logps/rejected": -5.414787769317627,
+      "loss": 0.4783,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.5306549072265625,
+      "rewards/margins": 2.591526508331299,
+      "rewards/rejected": -8.12218189239502,
+      "step": 460
+    },
+    {
+      "epoch": 0.5968253968253968,
+      "grad_norm": 2.9189071655273438,
+      "learning_rate": 3.88347887310836e-06,
+      "logits/chosen": 8.294754028320312,
+      "logits/rejected": 7.891358852386475,
+      "logps/chosen": -3.737588405609131,
+      "logps/rejected": -5.187026023864746,
+      "loss": 0.534,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.606382846832275,
+      "rewards/margins": 2.1741559505462646,
+      "rewards/rejected": -7.780538082122803,
+      "step": 470
+    },
+    {
+      "epoch": 0.6095238095238096,
+      "grad_norm": 2.306356906890869,
+      "learning_rate": 3.839566987447492e-06,
+      "logits/chosen": 8.484542846679688,
+      "logits/rejected": 8.39714241027832,
+      "logps/chosen": -3.494408369064331,
+      "logps/rejected": -5.29335880279541,
+      "loss": 0.4829,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.241612911224365,
+      "rewards/margins": 2.6984262466430664,
+      "rewards/rejected": -7.940038204193115,
+      "step": 480
+    },
+    {
+      "epoch": 0.6222222222222222,
+      "grad_norm": 2.2581255435943604,
+      "learning_rate": 3.795067523432826e-06,
+      "logits/chosen": 8.763944625854492,
+      "logits/rejected": 8.361797332763672,
+      "logps/chosen": -3.9699864387512207,
+      "logps/rejected": -6.275031566619873,
+      "loss": 0.4437,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.954979419708252,
+      "rewards/margins": 3.4575679302215576,
+      "rewards/rejected": -9.412548065185547,
+      "step": 490
+    },
+    {
+      "epoch": 0.6349206349206349,
+      "grad_norm": 2.2695906162261963,
+      "learning_rate": 3.7500000000000005e-06,
+      "logits/chosen": 7.8348212242126465,
+      "logits/rejected": 7.482022285461426,
+      "logps/chosen": -3.7032783031463623,
+      "logps/rejected": -5.9005255699157715,
+      "loss": 0.4411,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.554916858673096,
+      "rewards/margins": 3.2958710193634033,
+      "rewards/rejected": -8.850788116455078,
+      "step": 500
+    },
+    {
+      "epoch": 0.6349206349206349,
+      "eval_logits/chosen": 7.651264190673828,
+      "eval_logits/rejected": 7.331784248352051,
+      "eval_logps/chosen": -3.6274948120117188,
+      "eval_logps/rejected": -5.359984874725342,
+      "eval_loss": 0.4819534122943878,
+      "eval_rewards/accuracies": 0.796875,
+      "eval_rewards/chosen": -5.441242694854736,
+      "eval_rewards/margins": 2.5987353324890137,
+      "eval_rewards/rejected": -8.03997802734375,
+      "eval_runtime": 18.6061,
+      "eval_samples_per_second": 27.41,
+      "eval_steps_per_second": 3.44,
+      "step": 500
+    },
+    {
+      "epoch": 0.6476190476190476,
+      "grad_norm": 4.491377353668213,
+      "learning_rate": 3.7043841852542884e-06,
+      "logits/chosen": 8.346134185791016,
+      "logits/rejected": 7.885122776031494,
+      "logps/chosen": -3.1610424518585205,
+      "logps/rejected": -4.682709217071533,
+      "loss": 0.5284,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.7415642738342285,
+      "rewards/margins": 2.2825000286102295,
+      "rewards/rejected": -7.0240631103515625,
+      "step": 510
+    },
+    {
+      "epoch": 0.6603174603174603,
+      "grad_norm": 2.4325833320617676,
+      "learning_rate": 3.658240087799655e-06,
+      "logits/chosen": 7.592903137207031,
+      "logits/rejected": 7.519273281097412,
+      "logps/chosen": -3.836571455001831,
+      "logps/rejected": -5.284958839416504,
+      "loss": 0.5357,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -5.754857063293457,
+      "rewards/margins": 2.1725804805755615,
+      "rewards/rejected": -7.927438259124756,
+      "step": 520
+    },
+    {
+      "epoch": 0.6730158730158731,
+      "grad_norm": 3.2697038650512695,
+      "learning_rate": 3.611587947962319e-06,
+      "logits/chosen": 8.396175384521484,
+      "logits/rejected": 8.135710716247559,
+      "logps/chosen": -3.09209942817688,
+      "logps/rejected": -4.844483375549316,
+      "loss": 0.5079,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.638149261474609,
+      "rewards/margins": 2.6285760402679443,
+      "rewards/rejected": -7.266725063323975,
+      "step": 530
+    },
+    {
+      "epoch": 0.6857142857142857,
+      "grad_norm": 3.9283788204193115,
+      "learning_rate": 3.564448228912682e-06,
+      "logits/chosen": 7.992170810699463,
+      "logits/rejected": 7.327617645263672,
+      "logps/chosen": -3.6388747692108154,
+      "logps/rejected": -5.46083927154541,
+      "loss": 0.4713,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.458312511444092,
+      "rewards/margins": 2.7329471111297607,
+      "rewards/rejected": -8.191259384155273,
+      "step": 540
+    },
+    {
+      "epoch": 0.6984126984126984,
+      "grad_norm": 5.740615367889404,
+      "learning_rate": 3.516841607689501e-06,
+      "logits/chosen": 7.8473920822143555,
+      "logits/rejected": 7.583371162414551,
+      "logps/chosen": -3.4951674938201904,
+      "logps/rejected": -5.054746150970459,
+      "loss": 0.5543,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.242751121520996,
+      "rewards/margins": 2.3393683433532715,
+      "rewards/rejected": -7.582118988037109,
+      "step": 550
+    },
+    {
+      "epoch": 0.6984126984126984,
+      "eval_logits/chosen": 7.201205253601074,
+      "eval_logits/rejected": 6.811280727386475,
+      "eval_logps/chosen": -3.9208078384399414,
+      "eval_logps/rejected": -5.703615665435791,
+      "eval_loss": 0.458388090133667,
+      "eval_rewards/accuracies": 0.84375,
+      "eval_rewards/chosen": -5.881211757659912,
+      "eval_rewards/margins": 2.674211025238037,
+      "eval_rewards/rejected": -8.555423736572266,
+      "eval_runtime": 18.6149,
+      "eval_samples_per_second": 27.397,
+      "eval_steps_per_second": 3.438,
+      "step": 550
+    },
+    {
+      "epoch": 0.7111111111111111,
+      "grad_norm": 2.257397413253784,
+      "learning_rate": 3.4687889661302577e-06,
+      "logits/chosen": 7.826712608337402,
+      "logits/rejected": 7.213566780090332,
+      "logps/chosen": -4.419582843780518,
+      "logps/rejected": -6.427667140960693,
+      "loss": 0.4126,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -6.629374027252197,
+      "rewards/margins": 3.0121266841888428,
+      "rewards/rejected": -9.641500473022461,
+      "step": 560
+    },
+    {
+      "epoch": 0.7238095238095238,
+      "grad_norm": 3.2082905769348145,
+      "learning_rate": 3.4203113817116955e-06,
+      "logits/chosen": 6.789637565612793,
+      "logits/rejected": 6.794576168060303,
+      "logps/chosen": -3.8381495475769043,
+      "logps/rejected": -5.592730522155762,
+      "loss": 0.4379,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.757224082946777,
+      "rewards/margins": 2.6318702697753906,
+      "rewards/rejected": -8.389094352722168,
+      "step": 570
+    },
+    {
+      "epoch": 0.7365079365079366,
+      "grad_norm": 1.4739922285079956,
+      "learning_rate": 3.3714301183045382e-06,
+      "logits/chosen": 6.731423854827881,
+      "logits/rejected": 6.7872209548950195,
+      "logps/chosen": -2.9625697135925293,
+      "logps/rejected": -4.8117570877075195,
+      "loss": 0.4483,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.443854808807373,
+      "rewards/margins": 2.773780345916748,
+      "rewards/rejected": -7.217635154724121,
+      "step": 580
+    },
+    {
+      "epoch": 0.7492063492063492,
+      "grad_norm": 2.3974905014038086,
+      "learning_rate": 3.3221666168464584e-06,
+      "logits/chosen": 7.254199981689453,
+      "logits/rejected": 6.7625627517700195,
+      "logps/chosen": -3.562770366668701,
+      "logps/rejected": -5.307286262512207,
+      "loss": 0.4229,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.344155788421631,
+      "rewards/margins": 2.616774082183838,
+      "rewards/rejected": -7.960929870605469,
+      "step": 590
+    },
+    {
+      "epoch": 0.7619047619047619,
+      "grad_norm": 2.3709428310394287,
+      "learning_rate": 3.272542485937369e-06,
+      "logits/chosen": 7.226127624511719,
+      "logits/rejected": 6.73668909072876,
+      "logps/chosen": -3.8387343883514404,
+      "logps/rejected": -5.65582275390625,
+      "loss": 0.4304,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.7581024169921875,
+      "rewards/margins": 2.725632429122925,
+      "rewards/rejected": -8.483735084533691,
+      "step": 600
+    },
+    {
+      "epoch": 0.7619047619047619,
+      "eval_logits/chosen": 6.816856861114502,
+      "eval_logits/rejected": 6.3922905921936035,
+      "eval_logps/chosen": -3.8147382736206055,
+      "eval_logps/rejected": -5.733050346374512,
+      "eval_loss": 0.43386051058769226,
+      "eval_rewards/accuracies": 0.875,
+      "eval_rewards/chosen": -5.722107410430908,
+      "eval_rewards/margins": 2.877467393875122,
+      "eval_rewards/rejected": -8.59957504272461,
+      "eval_runtime": 18.5848,
+      "eval_samples_per_second": 27.442,
+      "eval_steps_per_second": 3.444,
+      "step": 600
+    },
+    {
+      "epoch": 0.7746031746031746,
+      "grad_norm": 8.223690032958984,
+      "learning_rate": 3.222579492361179e-06,
+      "logits/chosen": 6.017186164855957,
+      "logits/rejected": 6.098165988922119,
+      "logps/chosen": -4.070714950561523,
+      "logps/rejected": -5.999022006988525,
+      "loss": 0.4752,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -6.106072902679443,
+      "rewards/margins": 2.8924598693847656,
+      "rewards/rejected": -8.998533248901367,
+      "step": 610
+    },
+    {
+      "epoch": 0.7873015873015873,
+      "grad_norm": 2.494403839111328,
+      "learning_rate": 3.1722995515381644e-06,
+      "logits/chosen": 5.941250801086426,
+      "logits/rejected": 5.958924770355225,
+      "logps/chosen": -3.984475612640381,
+      "logps/rejected": -5.801859378814697,
+      "loss": 0.4245,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.97671365737915,
+      "rewards/margins": 2.726074695587158,
+      "rewards/rejected": -8.702788352966309,
+      "step": 620
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 1.906548023223877,
+      "learning_rate": 3.121724717912138e-06,
+      "logits/chosen": 6.673853874206543,
+      "logits/rejected": 6.525673866271973,
+      "logps/chosen": -3.4477615356445312,
+      "logps/rejected": -5.480903148651123,
+      "loss": 0.3671,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.171643257141113,
+      "rewards/margins": 3.0497121810913086,
+      "rewards/rejected": -8.221354484558105,
+      "step": 630
+    },
+    {
+      "epoch": 0.8126984126984127,
+      "grad_norm": 3.656189441680908,
+      "learning_rate": 3.0708771752766397e-06,
+      "logits/chosen": 6.097588062286377,
+      "logits/rejected": 6.049706935882568,
+      "logps/chosen": -3.8336167335510254,
+      "logps/rejected": -5.6922125816345215,
+      "loss": 0.4461,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.750425338745117,
+      "rewards/margins": 2.787893772125244,
+      "rewards/rejected": -8.53831958770752,
+      "step": 640
+    },
+    {
+      "epoch": 0.8253968253968254,
+      "grad_norm": 4.886603355407715,
+      "learning_rate": 3.019779227044398e-06,
+      "logits/chosen": 7.281914710998535,
+      "logits/rejected": 7.21079158782959,
+      "logps/chosen": -3.3890902996063232,
+      "logps/rejected": -5.331442356109619,
+      "loss": 0.4231,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.083634853363037,
+      "rewards/margins": 2.9135282039642334,
+      "rewards/rejected": -7.99716329574585,
+      "step": 650
+    },
+    {
+      "epoch": 0.8253968253968254,
+      "eval_logits/chosen": 7.003323554992676,
+      "eval_logits/rejected": 6.476384162902832,
+      "eval_logps/chosen": -3.6840155124664307,
+      "eval_logps/rejected": -5.79384708404541,
+      "eval_loss": 0.4155474007129669,
+      "eval_rewards/accuracies": 0.875,
+      "eval_rewards/chosen": -5.5260233879089355,
+      "eval_rewards/margins": 3.164747476577759,
+      "eval_rewards/rejected": -8.690771102905273,
+      "eval_runtime": 18.6081,
+      "eval_samples_per_second": 27.407,
+      "eval_steps_per_second": 3.439,
+      "step": 650
+    },
+    {
+      "epoch": 0.8380952380952381,
+      "grad_norm": 4.255698204040527,
+      "learning_rate": 2.9684532864643123e-06,
+      "logits/chosen": 7.213356018066406,
+      "logits/rejected": 6.562827110290527,
+      "logps/chosen": -4.553462505340576,
+      "logps/rejected": -7.089639186859131,
+      "loss": 0.4137,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.830193996429443,
+      "rewards/margins": 3.804264783859253,
+      "rewards/rejected": -10.6344575881958,
+      "step": 660
+    },
+    {
+      "epoch": 0.8507936507936508,
+      "grad_norm": 2.784740924835205,
+      "learning_rate": 2.9169218667902562e-06,
+      "logits/chosen": 7.139246463775635,
+      "logits/rejected": 7.061759948730469,
+      "logps/chosen": -3.2805087566375732,
+      "logps/rejected": -5.112305641174316,
+      "loss": 0.4138,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.920762538909912,
+      "rewards/margins": 2.7476963996887207,
+      "rewards/rejected": -7.668459415435791,
+      "step": 670
+    },
+    {
+      "epoch": 0.8634920634920635,
+      "grad_norm": 2.966514825820923,
+      "learning_rate": 2.8652075714060296e-06,
+      "logits/chosen": 7.687324523925781,
+      "logits/rejected": 6.8130292892456055,
+      "logps/chosen": -3.9359238147735596,
+      "logps/rejected": -6.452606201171875,
+      "loss": 0.4267,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.903885841369629,
+      "rewards/margins": 3.7750232219696045,
+      "rewards/rejected": -9.678911209106445,
+      "step": 680
+    },
+    {
+      "epoch": 0.8761904761904762,
+      "grad_norm": 2.919829845428467,
+      "learning_rate": 2.813333083910761e-06,
+      "logits/chosen": 7.361714839935303,
+      "logits/rejected": 6.8447113037109375,
+      "logps/chosen": -4.811502456665039,
+      "logps/rejected": -7.118119716644287,
+      "loss": 0.3531,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -7.217253684997559,
+      "rewards/margins": 3.4599266052246094,
+      "rewards/rejected": -10.677180290222168,
+      "step": 690
+    },
+    {
+      "epoch": 0.8888888888888888,
+      "grad_norm": 4.350555896759033,
+      "learning_rate": 2.761321158169134e-06,
+      "logits/chosen": 7.161977291107178,
+      "logits/rejected": 6.46688985824585,
+      "logps/chosen": -5.0110602378845215,
+      "logps/rejected": -7.346589088439941,
+      "loss": 0.4327,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -7.516589164733887,
+      "rewards/margins": 3.503293514251709,
+      "rewards/rejected": -11.01988410949707,
+      "step": 700
+    },
+    {
+      "epoch": 0.8888888888888888,
+      "eval_logits/chosen": 6.070926666259766,
+      "eval_logits/rejected": 5.528128147125244,
+      "eval_logps/chosen": -4.264724254608154,
+      "eval_logps/rejected": -6.332221031188965,
+      "eval_loss": 0.40030673146247864,
+      "eval_rewards/accuracies": 0.875,
+      "eval_rewards/chosen": -6.3970866203308105,
+      "eval_rewards/margins": 3.1012446880340576,
+      "eval_rewards/rejected": -9.498331069946289,
+      "eval_runtime": 18.5969,
+      "eval_samples_per_second": 27.424,
+      "eval_steps_per_second": 3.441,
+      "step": 700
+    },
+    {
+      "epoch": 0.9015873015873016,
+      "grad_norm": 2.253688335418701,
+      "learning_rate": 2.70919460833079e-06,
+      "logits/chosen": 6.234388828277588,
+      "logits/rejected": 5.7127604484558105,
+      "logps/chosen": -4.306221008300781,
+      "logps/rejected": -6.937534332275391,
+      "loss": 0.3878,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.459331512451172,
+      "rewards/margins": 3.946969509124756,
+      "rewards/rejected": -10.406301498413086,
+      "step": 710
+    },
+    {
+      "epoch": 0.9142857142857143,
+      "grad_norm": 5.499817371368408,
+      "learning_rate": 2.6569762988232838e-06,
+      "logits/chosen": 6.5680670738220215,
+      "logits/rejected": 5.752740383148193,
+      "logps/chosen": -4.220452308654785,
+      "logps/rejected": -6.862195014953613,
+      "loss": 0.3841,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.330678462982178,
+      "rewards/margins": 3.9626128673553467,
+      "rewards/rejected": -10.293292045593262,
+      "step": 720
+    },
+    {
+      "epoch": 0.926984126984127,
+      "grad_norm": 4.771604061126709,
+      "learning_rate": 2.604689134322999e-06,
+      "logits/chosen": 6.312363624572754,
+      "logits/rejected": 5.750403881072998,
+      "logps/chosen": -4.694201946258545,
+      "logps/rejected": -6.822142601013184,
+      "loss": 0.4237,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -7.041302680969238,
+      "rewards/margins": 3.1919119358062744,
+      "rewards/rejected": -10.233213424682617,
+      "step": 730
+    },
+    {
+      "epoch": 0.9396825396825397,
+      "grad_norm": 2.910656452178955,
+      "learning_rate": 2.5523560497083927e-06,
+      "logits/chosen": 5.823894500732422,
+      "logits/rejected": 5.497160911560059,
+      "logps/chosen": -4.014594078063965,
+      "logps/rejected": -6.6623029708862305,
+      "loss": 0.3566,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.021890640258789,
+      "rewards/margins": 3.9715638160705566,
+      "rewards/rejected": -9.99345588684082,
+      "step": 740
+    },
+    {
+      "epoch": 0.9523809523809523,
+      "grad_norm": 2.6968085765838623,
+      "learning_rate": 2.5e-06,
+      "logits/chosen": 5.813364028930664,
+      "logits/rejected": 5.676685333251953,
+      "logps/chosen": -3.678469181060791,
+      "logps/rejected": -6.177823066711426,
+      "loss": 0.3494,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.517704486846924,
+      "rewards/margins": 3.749030351638794,
+      "rewards/rejected": -9.26673412322998,
+      "step": 750
+    },
+    {
+      "epoch": 0.9523809523809523,
+      "eval_logits/chosen": 6.183505535125732,
+      "eval_logits/rejected": 5.580399036407471,
+      "eval_logps/chosen": -3.959765911102295,
+      "eval_logps/rejected": -6.2080864906311035,
+      "eval_loss": 0.37998583912849426,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -5.939648628234863,
+      "eval_rewards/margins": 3.372481107711792,
+      "eval_rewards/rejected": -9.312129974365234,
+      "eval_runtime": 18.6645,
+      "eval_samples_per_second": 27.325,
+      "eval_steps_per_second": 3.429,
+      "step": 750
+    },
+    {
+      "epoch": 0.9650793650793651,
+      "grad_norm": 2.2613444328308105,
+      "learning_rate": 2.447643950291608e-06,
+      "logits/chosen": 6.018897533416748,
+      "logits/rejected": 5.643963813781738,
+      "logps/chosen": -3.9108989238739014,
+      "logps/rejected": -6.244706630706787,
+      "loss": 0.3948,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.866348743438721,
+      "rewards/margins": 3.500711441040039,
+      "rewards/rejected": -9.367059707641602,
+      "step": 760
+    },
+    {
+      "epoch": 0.9777777777777777,
+      "grad_norm": 6.015537261962891,
+      "learning_rate": 2.3953108656770018e-06,
+      "logits/chosen": 5.021273136138916,
+      "logits/rejected": 4.748461723327637,
+      "logps/chosen": -3.692396640777588,
+      "logps/rejected": -6.072467803955078,
+      "loss": 0.3751,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.538595199584961,
+      "rewards/margins": 3.570105791091919,
+      "rewards/rejected": -9.108701705932617,
+      "step": 770
+    },
+    {
+      "epoch": 0.9904761904761905,
+      "grad_norm": 4.150258541107178,
+      "learning_rate": 2.3430237011767166e-06,
+      "logits/chosen": 6.515981197357178,
+      "logits/rejected": 5.819984436035156,
+      "logps/chosen": -4.485560894012451,
+      "logps/rejected": -6.801682949066162,
+      "loss": 0.3818,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.728341579437256,
+      "rewards/margins": 3.4741833209991455,
+      "rewards/rejected": -10.20252513885498,
+      "step": 780
+    },
+    {
+      "epoch": 1.0025396825396826,
+      "grad_norm": 2.078934907913208,
+      "learning_rate": 2.290805391669212e-06,
+      "logits/chosen": 7.137444972991943,
+      "logits/rejected": 6.565527439117432,
+      "logps/chosen": -4.413013935089111,
+      "logps/rejected": -7.346111297607422,
+      "loss": 0.3495,
+      "rewards/accuracies": 0.9473684430122375,
+      "rewards/chosen": -6.619520664215088,
+      "rewards/margins": 4.399644374847412,
+      "rewards/rejected": -11.019165992736816,
+      "step": 790
+    },
+    {
+      "epoch": 1.0152380952380953,
+      "grad_norm": 3.736583709716797,
+      "learning_rate": 2.238678841830867e-06,
+      "logits/chosen": 6.755683898925781,
+      "logits/rejected": 6.001730442047119,
+      "logps/chosen": -4.394175052642822,
+      "logps/rejected": -6.807177543640137,
+      "loss": 0.386,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.591261386871338,
+      "rewards/margins": 3.61950421333313,
+      "rewards/rejected": -10.210765838623047,
+      "step": 800
+    },
+    {
+      "epoch": 1.0152380952380953,
+      "eval_logits/chosen": 6.0334320068359375,
+      "eval_logits/rejected": 5.392582416534424,
+      "eval_logps/chosen": -3.8732340335845947,
+      "eval_logps/rejected": -6.17905855178833,
+      "eval_loss": 0.3669988214969635,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -5.809851169586182,
+      "eval_rewards/margins": 3.4587368965148926,
+      "eval_rewards/rejected": -9.26858901977539,
+      "eval_runtime": 18.6021,
+      "eval_samples_per_second": 27.416,
+      "eval_steps_per_second": 3.44,
+      "step": 800
+    },
+    {
+      "epoch": 1.027936507936508,
+      "grad_norm": 4.095701694488525,
+      "learning_rate": 2.186666916089239e-06,
+      "logits/chosen": 6.593466281890869,
+      "logits/rejected": 6.300512790679932,
+      "logps/chosen": -3.7561659812927246,
+      "logps/rejected": -5.825091361999512,
+      "loss": 0.3869,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.63424825668335,
+      "rewards/margins": 3.103389024734497,
+      "rewards/rejected": -8.73763656616211,
+      "step": 810
+    },
+    {
+      "epoch": 1.0406349206349206,
+      "grad_norm": 2.803724765777588,
+      "learning_rate": 2.134792428593971e-06,
+      "logits/chosen": 4.933786392211914,
+      "logits/rejected": 4.718127727508545,
+      "logps/chosen": -3.9171605110168457,
+      "logps/rejected": -6.475947380065918,
+      "loss": 0.3528,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.875741004943848,
+      "rewards/margins": 3.8381810188293457,
+      "rewards/rejected": -9.713921546936035,
+      "step": 820
+    },
+    {
+      "epoch": 1.0533333333333332,
+      "grad_norm": 3.8399882316589355,
+      "learning_rate": 2.0830781332097446e-06,
+      "logits/chosen": 5.394495487213135,
+      "logits/rejected": 5.215309143066406,
+      "logps/chosen": -4.407529354095459,
+      "logps/rejected": -6.585978031158447,
+      "loss": 0.3686,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -6.611294746398926,
+      "rewards/margins": 3.2676727771759033,
+      "rewards/rejected": -9.87896728515625,
+      "step": 830
+    },
+    {
+      "epoch": 1.066031746031746,
+      "grad_norm": 4.447514533996582,
+      "learning_rate": 2.031546713535688e-06,
+      "logits/chosen": 6.456748962402344,
+      "logits/rejected": 6.44448709487915,
+      "logps/chosen": -4.091554164886475,
+      "logps/rejected": -6.2741193771362305,
+      "loss": 0.3309,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.137330532073975,
+      "rewards/margins": 3.273848056793213,
+      "rewards/rejected": -9.411178588867188,
+      "step": 840
+    },
+    {
+      "epoch": 1.0787301587301588,
+      "grad_norm": 4.561574459075928,
+      "learning_rate": 1.9802207729556023e-06,
+      "logits/chosen": 5.541195869445801,
+      "logits/rejected": 5.277142524719238,
+      "logps/chosen": -4.099520683288574,
+      "logps/rejected": -6.2324628829956055,
+      "loss": 0.3944,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -6.1492815017700195,
+      "rewards/margins": 3.1994125843048096,
+      "rewards/rejected": -9.34869384765625,
+      "step": 850
+    },
+    {
+      "epoch": 1.0787301587301588,
+      "eval_logits/chosen": 5.802575588226318,
+      "eval_logits/rejected": 5.148407936096191,
+      "eval_logps/chosen": -3.714815139770508,
+      "eval_logps/rejected": -6.146266460418701,
+      "eval_loss": 0.3534550666809082,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -5.572222709655762,
+      "eval_rewards/margins": 3.647176742553711,
+      "eval_rewards/rejected": -9.219400405883789,
+      "eval_runtime": 18.9204,
+      "eval_samples_per_second": 26.955,
+      "eval_steps_per_second": 3.383,
+      "step": 850
+    },
+    {
+      "epoch": 1.0914285714285714,
+      "grad_norm": 2.9834394454956055,
+      "learning_rate": 1.9291228247233607e-06,
+      "logits/chosen": 6.26169490814209,
+      "logits/rejected": 5.704649925231934,
+      "logps/chosen": -3.854257583618164,
+      "logps/rejected": -5.967762470245361,
+      "loss": 0.335,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.781386852264404,
+      "rewards/margins": 3.1702563762664795,
+      "rewards/rejected": -8.951642990112305,
+      "step": 860
+    },
+    {
+      "epoch": 1.104126984126984,
+      "grad_norm": 2.181910514831543,
+      "learning_rate": 1.8782752820878636e-06,
+      "logits/chosen": 6.014175891876221,
+      "logits/rejected": 6.0545759201049805,
+      "logps/chosen": -3.297023296356201,
+      "logps/rejected": -5.658031463623047,
+      "loss": 0.3228,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.945534706115723,
+      "rewards/margins": 3.5415127277374268,
+      "rewards/rejected": -8.48704719543457,
+      "step": 870
+    },
+    {
+      "epoch": 1.116825396825397,
+      "grad_norm": 3.857339382171631,
+      "learning_rate": 1.827700448461836e-06,
+      "logits/chosen": 5.903520107269287,
+      "logits/rejected": 5.670694828033447,
+      "logps/chosen": -3.792879104614258,
+      "logps/rejected": -6.34661340713501,
+      "loss": 0.3311,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.689318656921387,
+      "rewards/margins": 3.8306007385253906,
+      "rewards/rejected": -9.519918441772461,
+      "step": 880
+    },
+    {
+      "epoch": 1.1295238095238096,
+      "grad_norm": 3.8721022605895996,
+      "learning_rate": 1.7774205076388207e-06,
+      "logits/chosen": 5.741324424743652,
+      "logits/rejected": 5.332767009735107,
+      "logps/chosen": -4.255464553833008,
+      "logps/rejected": -7.147697448730469,
+      "loss": 0.3063,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -6.383196830749512,
+      "rewards/margins": 4.33834981918335,
+      "rewards/rejected": -10.721546173095703,
+      "step": 890
+    },
+    {
+      "epoch": 1.1422222222222222,
+      "grad_norm": 2.701751708984375,
+      "learning_rate": 1.7274575140626318e-06,
+      "logits/chosen": 5.502126693725586,
+      "logits/rejected": 4.810341835021973,
+      "logps/chosen": -4.429174900054932,
+      "logps/rejected": -6.926125526428223,
+      "loss": 0.3479,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.643762111663818,
+      "rewards/margins": 3.7454261779785156,
+      "rewards/rejected": -10.389187812805176,
+      "step": 900
+    },
+    {
+      "epoch": 1.1422222222222222,
+      "eval_logits/chosen": 5.738080978393555,
+      "eval_logits/rejected": 5.0560622215271,
+      "eval_logps/chosen": -3.8086652755737305,
+      "eval_logps/rejected": -6.326536178588867,
+      "eval_loss": 0.3475661873817444,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -5.7129974365234375,
+      "eval_rewards/margins": 3.7768068313598633,
+      "eval_rewards/rejected": -9.4898042678833,
+      "eval_runtime": 18.797,
+      "eval_samples_per_second": 27.132,
+      "eval_steps_per_second": 3.405,
+      "step": 900
+    },
+    {
+      "epoch": 1.154920634920635,
+      "grad_norm": 3.521653175354004,
+      "learning_rate": 1.677833383153542e-06,
+      "logits/chosen": 5.684510231018066,
+      "logits/rejected": 5.133933067321777,
+      "logps/chosen": -3.8667006492614746,
+      "logps/rejected": -6.714064121246338,
+      "loss": 0.3325,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.800051212310791,
+      "rewards/margins": 4.271045207977295,
+      "rewards/rejected": -10.071096420288086,
+      "step": 910
+    },
+    {
+      "epoch": 1.1676190476190476,
+      "grad_norm": 4.271011829376221,
+      "learning_rate": 1.6285698816954626e-06,
+      "logits/chosen": 5.627573013305664,
+      "logits/rejected": 5.377304553985596,
+      "logps/chosen": -3.229398727416992,
+      "logps/rejected": -6.157980918884277,
+      "loss": 0.3216,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.844098091125488,
+      "rewards/margins": 4.3928728103637695,
+      "rewards/rejected": -9.236970901489258,
+      "step": 920
+    },
+    {
+      "epoch": 1.1803174603174602,
+      "grad_norm": 5.995442867279053,
+      "learning_rate": 1.5796886182883053e-06,
+      "logits/chosen": 5.5010247230529785,
+      "logits/rejected": 5.061415672302246,
+      "logps/chosen": -4.375211715698242,
+      "logps/rejected": -7.426985263824463,
+      "loss": 0.372,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -6.562816619873047,
+      "rewards/margins": 4.577658653259277,
+      "rewards/rejected": -11.14047622680664,
+      "step": 930
+    },
+    {
+      "epoch": 1.193015873015873,
+      "grad_norm": 3.330029249191284,
+      "learning_rate": 1.5312110338697427e-06,
+      "logits/chosen": 5.348600387573242,
+      "logits/rejected": 4.949542045593262,
+      "logps/chosen": -3.677006483078003,
+      "logps/rejected": -6.103867530822754,
+      "loss": 0.3265,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.515509605407715,
+      "rewards/margins": 3.640291929244995,
+      "rewards/rejected": -9.155801773071289,
+      "step": 940
+    },
+    {
+      "epoch": 1.2057142857142857,
+      "grad_norm": 6.799316883087158,
+      "learning_rate": 1.4831583923105e-06,
+      "logits/chosen": 5.847796440124512,
+      "logits/rejected": 5.91001033782959,
+      "logps/chosen": -3.563372850418091,
+      "logps/rejected": -6.072541236877441,
+      "loss": 0.3288,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.345059394836426,
+      "rewards/margins": 3.7637526988983154,
+      "rewards/rejected": -9.108811378479004,
+      "step": 950
+    },
+    {
+      "epoch": 1.2057142857142857,
+      "eval_logits/chosen": 5.582961082458496,
+      "eval_logits/rejected": 4.875361442565918,
+      "eval_logps/chosen": -3.9592418670654297,
+      "eval_logps/rejected": -6.523550033569336,
+      "eval_loss": 0.3351227343082428,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -5.938862323760986,
+      "eval_rewards/margins": 3.846461772918701,
+      "eval_rewards/rejected": -9.785325050354004,
+      "eval_runtime": 18.5918,
+      "eval_samples_per_second": 27.431,
+      "eval_steps_per_second": 3.442,
+      "step": 950
+    },
+    {
+      "epoch": 1.2184126984126984,
+      "grad_norm": 8.878515243530273,
+      "learning_rate": 1.4355517710873184e-06,
+      "logits/chosen": 5.0982561111450195,
+      "logits/rejected": 4.761422634124756,
+      "logps/chosen": -4.501910209655762,
+      "logps/rejected": -6.980757236480713,
+      "loss": 0.3757,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.752864837646484,
+      "rewards/margins": 3.718270778656006,
+      "rewards/rejected": -10.471136093139648,
+      "step": 960
+    },
+    {
+      "epoch": 1.231111111111111,
+      "grad_norm": 1.8070721626281738,
+      "learning_rate": 1.388412052037682e-06,
+      "logits/chosen": 5.758985996246338,
+      "logits/rejected": 5.578696250915527,
+      "logps/chosen": -3.597564697265625,
+      "logps/rejected": -6.114877223968506,
+      "loss": 0.2717,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.396347999572754,
+      "rewards/margins": 3.775968074798584,
+      "rewards/rejected": -9.17231559753418,
+      "step": 970
+    },
+    {
+      "epoch": 1.243809523809524,
+      "grad_norm": 4.821048259735107,
+      "learning_rate": 1.3417599122003464e-06,
+      "logits/chosen": 6.082779884338379,
+      "logits/rejected": 5.357623100280762,
+      "logps/chosen": -4.046582221984863,
+      "logps/rejected": -6.718660831451416,
+      "loss": 0.3108,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.069873332977295,
+      "rewards/margins": 4.008117198944092,
+      "rewards/rejected": -10.077990531921387,
+      "step": 980
+    },
+    {
+      "epoch": 1.2565079365079366,
+      "grad_norm": 3.886730670928955,
+      "learning_rate": 1.2956158147457116e-06,
+      "logits/chosen": 5.890940189361572,
+      "logits/rejected": 5.407658576965332,
+      "logps/chosen": -4.195148468017578,
+      "logps/rejected": -6.932916164398193,
+      "loss": 0.303,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.292722225189209,
+      "rewards/margins": 4.1066508293151855,
+      "rewards/rejected": -10.399374008178711,
+      "step": 990
+    },
+    {
+      "epoch": 1.2692063492063492,
+      "grad_norm": 4.833825588226318,
+      "learning_rate": 1.2500000000000007e-06,
+      "logits/chosen": 4.744889259338379,
+      "logits/rejected": 4.313926696777344,
+      "logps/chosen": -4.499282360076904,
+      "logps/rejected": -7.198727607727051,
+      "loss": 0.3254,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -6.748922824859619,
+      "rewards/margins": 4.049168586730957,
+      "rewards/rejected": -10.798090934753418,
+      "step": 1000
+    },
+    {
+      "epoch": 1.2692063492063492,
+      "eval_logits/chosen": 5.406946182250977,
+      "eval_logits/rejected": 4.719847202301025,
+      "eval_logps/chosen": -4.074473857879639,
+      "eval_logps/rejected": -6.686357498168945,
+      "eval_loss": 0.3262788951396942,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -6.111710548400879,
+      "eval_rewards/margins": 3.9178245067596436,
+      "eval_rewards/rejected": -10.029535293579102,
+      "eval_runtime": 18.6049,
+      "eval_samples_per_second": 27.412,
+      "eval_steps_per_second": 3.44,
+      "step": 1000
+    },
+    {
+      "epoch": 1.2819047619047619,
+      "grad_norm": 3.368656635284424,
+      "learning_rate": 1.204932476567175e-06,
+      "logits/chosen": 4.968677520751953,
+      "logits/rejected": 4.569239139556885,
+      "logps/chosen": -4.589535713195801,
+      "logps/rejected": -7.349358558654785,
+      "loss": 0.3144,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -6.884303092956543,
+      "rewards/margins": 4.139734745025635,
+      "rewards/rejected": -11.024038314819336,
+      "step": 1010
+    },
+    {
+      "epoch": 1.2946031746031745,
+      "grad_norm": 3.3188092708587646,
+      "learning_rate": 1.160433012552508e-06,
+      "logits/chosen": 5.417339324951172,
+      "logits/rejected": 4.767195701599121,
+      "logps/chosen": -3.8293652534484863,
+      "logps/rejected": -6.963759422302246,
+      "loss": 0.327,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.744047164916992,
+      "rewards/margins": 4.701592445373535,
+      "rewards/rejected": -10.445640563964844,
+      "step": 1020
+    },
+    {
+      "epoch": 1.3073015873015872,
+      "grad_norm": 3.1887564659118652,
+      "learning_rate": 1.11652112689164e-06,
+      "logits/chosen": 5.339169025421143,
+      "logits/rejected": 5.1165924072265625,
+      "logps/chosen": -4.203000068664551,
+      "logps/rejected": -6.630669593811035,
+      "loss": 0.3242,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.304499626159668,
+      "rewards/margins": 3.6415042877197266,
+      "rewards/rejected": -9.946003913879395,
+      "step": 1030
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 2.607088565826416,
+      "learning_rate": 1.073216080788921e-06,
+      "logits/chosen": 6.265632629394531,
+      "logits/rejected": 5.395622253417969,
+      "logps/chosen": -4.786427974700928,
+      "logps/rejected": -8.071474075317383,
+      "loss": 0.3077,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.179642677307129,
+      "rewards/margins": 4.92756986618042,
+      "rewards/rejected": -12.107213020324707,
+      "step": 1040
+    },
+    {
+      "epoch": 1.3326984126984127,
+      "grad_norm": 3.955307722091675,
+      "learning_rate": 1.0305368692688175e-06,
+      "logits/chosen": 5.4535136222839355,
+      "logits/rejected": 5.403381824493408,
+      "logps/chosen": -4.037718296051025,
+      "logps/rejected": -6.1900153160095215,
+      "loss": 0.3369,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -6.056577682495117,
+      "rewards/margins": 3.2284457683563232,
+      "rewards/rejected": -9.28502368927002,
+      "step": 1050
+    },
+    {
+      "epoch": 1.3326984126984127,
+      "eval_logits/chosen": 5.332888126373291,
+      "eval_logits/rejected": 4.61313533782959,
+      "eval_logps/chosen": -4.032596588134766,
+      "eval_logps/rejected": -6.697214126586914,
+      "eval_loss": 0.32348182797431946,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -6.04889440536499,
+      "eval_rewards/margins": 3.9969258308410645,
+      "eval_rewards/rejected": -10.045821189880371,
+      "eval_runtime": 18.6029,
+      "eval_samples_per_second": 27.415,
+      "eval_steps_per_second": 3.44,
+      "step": 1050
+    },
+    {
+      "epoch": 1.3453968253968254,
+      "grad_norm": 3.041147232055664,
+      "learning_rate": 9.88502212844063e-07,
+      "logits/chosen": 5.934795379638672,
+      "logits/rejected": 5.3202738761901855,
+      "logps/chosen": -4.406095027923584,
+      "logps/rejected": -7.17871618270874,
+      "loss": 0.3061,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -6.609143257141113,
+      "rewards/margins": 4.158930778503418,
+      "rewards/rejected": -10.768074035644531,
+      "step": 1060
+    },
+    {
+      "epoch": 1.358095238095238,
+      "grad_norm": 2.5167226791381836,
+      "learning_rate": 9.471305493042243e-07,
+      "logits/chosen": 5.5874810218811035,
+      "logits/rejected": 5.0323638916015625,
+      "logps/chosen": -4.143420219421387,
+      "logps/rejected": -7.082531929016113,
+      "loss": 0.3183,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.21513032913208,
+      "rewards/margins": 4.408668041229248,
+      "rewards/rejected": -10.623798370361328,
+      "step": 1070
+    },
+    {
+      "epoch": 1.370793650793651,
+      "grad_norm": 4.380863666534424,
+      "learning_rate": 9.064400256282757e-07,
+      "logits/chosen": 4.3980302810668945,
+      "logits/rejected": 4.2179107666015625,
+      "logps/chosen": -3.5643622875213623,
+      "logps/rejected": -6.895630836486816,
+      "loss": 0.3187,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.346543312072754,
+      "rewards/margins": 4.996903419494629,
+      "rewards/rejected": -10.343446731567383,
+      "step": 1080
+    },
+    {
+      "epoch": 1.3834920634920636,
+      "grad_norm": 2.758718967437744,
+      "learning_rate": 8.664484900247363e-07,
+      "logits/chosen": 5.32845401763916,
+      "logits/rejected": 4.726469993591309,
+      "logps/chosen": -4.442277908325195,
+      "logps/rejected": -7.003743648529053,
+      "loss": 0.3229,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.663416862487793,
+      "rewards/margins": 3.8421969413757324,
+      "rewards/rejected": -10.505613327026367,
+      "step": 1090
+    },
+    {
+      "epoch": 1.3961904761904762,
+      "grad_norm": 13.110466003417969,
+      "learning_rate": 8.271734841028553e-07,
+      "logits/chosen": 4.463423728942871,
+      "logits/rejected": 4.455904006958008,
+      "logps/chosen": -3.9567787647247314,
+      "logps/rejected": -7.14736795425415,
+      "loss": 0.3538,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.935168743133545,
+      "rewards/margins": 4.78588342666626,
+      "rewards/rejected": -10.721052169799805,
+      "step": 1100
+    },
+    {
+      "epoch": 1.3961904761904762,
+      "eval_logits/chosen": 5.36574125289917,
+      "eval_logits/rejected": 4.6298828125,
+      "eval_logps/chosen": -4.231103897094727,
+      "eval_logps/rejected": -6.961848735809326,
+      "eval_loss": 0.3140927851200104,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -6.346655368804932,
+      "eval_rewards/margins": 4.09611701965332,
+      "eval_rewards/rejected": -10.44277286529541,
+      "eval_runtime": 18.6091,
+      "eval_samples_per_second": 27.406,
+      "eval_steps_per_second": 3.439,
+      "step": 1100
+    },
+    {
+      "epoch": 1.4088888888888889,
+      "grad_norm": 4.922638893127441,
+      "learning_rate": 7.886322351782782e-07,
+      "logits/chosen": 5.046323299407959,
+      "logits/rejected": 4.895082950592041,
+      "logps/chosen": -3.7289280891418457,
+      "logps/rejected": -6.19178581237793,
+      "loss": 0.2895,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.593391418457031,
+      "rewards/margins": 3.6942875385284424,
+      "rewards/rejected": -9.287678718566895,
+      "step": 1110
+    },
+    {
+      "epoch": 1.4215873015873015,
+      "grad_norm": 3.6287381649017334,
+      "learning_rate": 7.508416487165862e-07,
+      "logits/chosen": 4.906771659851074,
+      "logits/rejected": 4.588590145111084,
+      "logps/chosen": -4.353616714477539,
+      "logps/rejected": -7.00437068939209,
+      "loss": 0.2854,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.53042459487915,
+      "rewards/margins": 3.976130723953247,
+      "rewards/rejected": -10.506555557250977,
+      "step": 1120
+    },
+    {
+      "epoch": 1.4342857142857142,
+      "grad_norm": 6.361443519592285,
+      "learning_rate": 7.138183009179922e-07,
+      "logits/chosen": 5.715910911560059,
+      "logits/rejected": 5.303093910217285,
+      "logps/chosen": -4.322514533996582,
+      "logps/rejected": -6.951764106750488,
+      "loss": 0.3345,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.483772277832031,
+      "rewards/margins": 3.943873167037964,
+      "rewards/rejected": -10.427645683288574,
+      "step": 1130
+    },
+    {
+      "epoch": 1.446984126984127,
+      "grad_norm": 3.708784818649292,
+      "learning_rate": 6.775784314464717e-07,
+      "logits/chosen": 5.768750190734863,
+      "logits/rejected": 5.022824764251709,
+      "logps/chosen": -4.504899501800537,
+      "logps/rejected": -7.219842433929443,
+      "loss": 0.3077,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.757348537445068,
+      "rewards/margins": 4.072414875030518,
+      "rewards/rejected": -10.829763412475586,
+      "step": 1140
+    },
+    {
+      "epoch": 1.4596825396825397,
+      "grad_norm": 3.36335825920105,
+      "learning_rate": 6.421379363065142e-07,
+      "logits/chosen": 6.26393985748291,
+      "logits/rejected": 5.9589128494262695,
+      "logps/chosen": -4.327595233917236,
+      "logps/rejected": -7.032060146331787,
+      "loss": 0.3086,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.491393089294434,
+      "rewards/margins": 4.056697845458984,
+      "rewards/rejected": -10.548090934753418,
+      "step": 1150
+    },
+    {
+      "epoch": 1.4596825396825397,
+      "eval_logits/chosen": 5.456152439117432,
+      "eval_logits/rejected": 4.708561897277832,
+      "eval_logps/chosen": -4.160144805908203,
+      "eval_logps/rejected": -6.92811393737793,
+      "eval_loss": 0.3138086199760437,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -6.2402167320251465,
+      "eval_rewards/margins": 4.151954650878906,
+      "eval_rewards/rejected": -10.392171859741211,
+      "eval_runtime": 18.6031,
+      "eval_samples_per_second": 27.415,
+      "eval_steps_per_second": 3.44,
+      "step": 1150
+    },
+    {
+      "epoch": 1.4723809523809523,
+      "grad_norm": 4.188318729400635,
+      "learning_rate": 6.075123608706093e-07,
+      "logits/chosen": 5.52956485748291,
+      "logits/rejected": 5.087782859802246,
+      "logps/chosen": -3.985388994216919,
+      "logps/rejected": -6.644972324371338,
+      "loss": 0.3143,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.978083610534668,
+      "rewards/margins": 3.9893736839294434,
+      "rewards/rejected": -9.96745777130127,
+      "step": 1160
+    },
+    {
+      "epoch": 1.485079365079365,
+      "grad_norm": 3.731320381164551,
+      "learning_rate": 5.737168930605272e-07,
+      "logits/chosen": 4.43437385559082,
+      "logits/rejected": 4.397299766540527,
+      "logps/chosen": -4.047328948974609,
+      "logps/rejected": -6.8785400390625,
+      "loss": 0.2796,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.070993423461914,
+      "rewards/margins": 4.246816635131836,
+      "rewards/rejected": -10.31781005859375,
+      "step": 1170
+    },
+    {
+      "epoch": 1.4977777777777779,
+      "grad_norm": 4.786388397216797,
+      "learning_rate": 5.407663566854008e-07,
+      "logits/chosen": 5.6721954345703125,
+      "logits/rejected": 5.462382793426514,
+      "logps/chosen": -3.7892093658447266,
+      "logps/rejected": -6.402613162994385,
+      "loss": 0.3145,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.68381404876709,
+      "rewards/margins": 3.9201061725616455,
+      "rewards/rejected": -9.603919982910156,
+      "step": 1180
+    },
+    {
+      "epoch": 1.5104761904761905,
+      "grad_norm": 2.9629695415496826,
+      "learning_rate": 5.086752049395094e-07,
+      "logits/chosen": 5.122151851654053,
+      "logits/rejected": 4.882320404052734,
+      "logps/chosen": -4.3091840744018555,
+      "logps/rejected": -6.654644966125488,
+      "loss": 0.2832,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.463775634765625,
+      "rewards/margins": 3.5181918144226074,
+      "rewards/rejected": -9.981966972351074,
+      "step": 1190
+    },
+    {
+      "epoch": 1.5231746031746032,
+      "grad_norm": 2.9508302211761475,
+      "learning_rate": 4.774575140626317e-07,
+      "logits/chosen": 6.055984020233154,
+      "logits/rejected": 5.055758953094482,
+      "logps/chosen": -5.162354469299316,
+      "logps/rejected": -8.137951850891113,
+      "loss": 0.2911,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.743532657623291,
+      "rewards/margins": 4.4633941650390625,
+      "rewards/rejected": -12.206927299499512,
+      "step": 1200
+    },
+    {
+      "epoch": 1.5231746031746032,
+      "eval_logits/chosen": 5.3908843994140625,
+      "eval_logits/rejected": 4.647245407104492,
+      "eval_logps/chosen": -4.132325649261475,
+      "eval_logps/rejected": -6.900644302368164,
+      "eval_loss": 0.3120686113834381,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -6.198488235473633,
+      "eval_rewards/margins": 4.152477741241455,
+      "eval_rewards/rejected": -10.35096549987793,
+      "eval_runtime": 18.6071,
+      "eval_samples_per_second": 27.409,
+      "eval_steps_per_second": 3.44,
+      "step": 1200
+    },
+    {
+      "epoch": 1.5358730158730158,
+      "grad_norm": 4.485844135284424,
+      "learning_rate": 4.4712697716573994e-07,
+      "logits/chosen": 5.793521881103516,
+      "logits/rejected": 5.4918904304504395,
+      "logps/chosen": -4.251668930053711,
+      "logps/rejected": -6.8831281661987305,
+      "loss": 0.2888,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.377503395080566,
+      "rewards/margins": 3.9471893310546875,
+      "rewards/rejected": -10.324692726135254,
+      "step": 1210
+    },
+    {
+      "epoch": 1.5485714285714285,
+      "grad_norm": 9.882649421691895,
+      "learning_rate": 4.1769689822475147e-07,
+      "logits/chosen": 5.395316123962402,
+      "logits/rejected": 4.986631870269775,
+      "logps/chosen": -3.989335298538208,
+      "logps/rejected": -6.505860805511475,
+      "loss": 0.2933,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.984002113342285,
+      "rewards/margins": 3.7747879028320312,
+      "rewards/rejected": -9.758790016174316,
+      "step": 1220
+    },
+    {
+      "epoch": 1.5612698412698411,
+      "grad_norm": 3.3249247074127197,
+      "learning_rate": 3.891801862449629e-07,
+      "logits/chosen": 5.716710090637207,
+      "logits/rejected": 5.00654411315918,
+      "logps/chosen": -3.812396287918091,
+      "logps/rejected": -6.605644226074219,
+      "loss": 0.2754,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.718594551086426,
+      "rewards/margins": 4.189871788024902,
+      "rewards/rejected": -9.908466339111328,
+      "step": 1230
+    },
+    {
+      "epoch": 1.573968253968254,
+      "grad_norm": 4.766965389251709,
+      "learning_rate": 3.615893495987335e-07,
+      "logits/chosen": 5.129265308380127,
+      "logits/rejected": 4.630931377410889,
+      "logps/chosen": -3.9612507820129395,
+      "logps/rejected": -6.8835625648498535,
+      "loss": 0.3342,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.941876411437988,
+      "rewards/margins": 4.383467674255371,
+      "rewards/rejected": -10.325343132019043,
+      "step": 1240
+    },
+    {
+      "epoch": 1.5866666666666667,
+      "grad_norm": 4.599531650543213,
+      "learning_rate": 3.3493649053890325e-07,
+      "logits/chosen": 5.172502040863037,
+      "logits/rejected": 4.7626752853393555,
+      "logps/chosen": -3.961320161819458,
+      "logps/rejected": -6.46725606918335,
+      "loss": 0.2882,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.94197940826416,
+      "rewards/margins": 3.7589058876037598,
+      "rewards/rejected": -9.700885772705078,
+      "step": 1250
+    },
+    {
+      "epoch": 1.5866666666666667,
+      "eval_logits/chosen": 5.349624156951904,
+      "eval_logits/rejected": 4.600058555603027,
+      "eval_logps/chosen": -4.137331008911133,
+      "eval_logps/rejected": -6.9393310546875,
+      "eval_loss": 0.3089294135570526,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -6.205997467041016,
+      "eval_rewards/margins": 4.202999114990234,
+      "eval_rewards/rejected": -10.408995628356934,
+      "eval_runtime": 18.5963,
+      "eval_samples_per_second": 27.425,
+      "eval_steps_per_second": 3.442,
+      "step": 1250
+    },
+    {
+      "epoch": 1.5993650793650793,
+      "grad_norm": 2.7759478092193604,
+      "learning_rate": 3.092332998903416e-07,
+      "logits/chosen": 4.7535295486450195,
+      "logits/rejected": 4.923503398895264,
+      "logps/chosen": -3.9828543663024902,
+      "logps/rejected": -6.226633071899414,
+      "loss": 0.3187,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.974281311035156,
+      "rewards/margins": 3.3656678199768066,
+      "rewards/rejected": -9.339948654174805,
+      "step": 1260
+    },
+    {
+      "epoch": 1.6120634920634922,
+      "grad_norm": 3.4311914443969727,
+      "learning_rate": 2.844910519219632e-07,
+      "logits/chosen": 5.836272239685059,
+      "logits/rejected": 5.564979076385498,
+      "logps/chosen": -3.938349485397339,
+      "logps/rejected": -6.979580879211426,
+      "loss": 0.2769,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.907524585723877,
+      "rewards/margins": 4.561846733093262,
+      "rewards/rejected": -10.469371795654297,
+      "step": 1270
+    },
+    {
+      "epoch": 1.6247619047619049,
+      "grad_norm": 5.06914758682251,
+      "learning_rate": 2.6072059940146775e-07,
+      "logits/chosen": 5.980331897735596,
+      "logits/rejected": 5.410573482513428,
+      "logps/chosen": -4.529973030090332,
+      "logps/rejected": -7.559577941894531,
+      "loss": 0.2804,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.794958591461182,
+      "rewards/margins": 4.544407844543457,
+      "rewards/rejected": -11.33936595916748,
+      "step": 1280
+    },
+    {
+      "epoch": 1.6374603174603175,
+      "grad_norm": 3.5914173126220703,
+      "learning_rate": 2.3793236883495164e-07,
+      "logits/chosen": 5.733067512512207,
+      "logits/rejected": 5.0557990074157715,
+      "logps/chosen": -4.857331275939941,
+      "logps/rejected": -7.6449127197265625,
+      "loss": 0.3558,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -7.285996913909912,
+      "rewards/margins": 4.181371688842773,
+      "rewards/rejected": -11.467369079589844,
+      "step": 1290
+    },
+    {
+      "epoch": 1.6501587301587302,
+      "grad_norm": 4.237303256988525,
+      "learning_rate": 2.1613635589349756e-07,
+      "logits/chosen": 4.954525947570801,
+      "logits/rejected": 4.65056848526001,
+      "logps/chosen": -4.077624320983887,
+      "logps/rejected": -7.1667280197143555,
+      "loss": 0.3516,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.116436958312988,
+      "rewards/margins": 4.633655548095703,
+      "rewards/rejected": -10.750093460083008,
+      "step": 1300
+    },
+    {
+      "epoch": 1.6501587301587302,
+      "eval_logits/chosen": 5.3195085525512695,
+      "eval_logits/rejected": 4.565763473510742,
+      "eval_logps/chosen": -4.004814147949219,
+      "eval_logps/rejected": -6.822023391723633,
+      "eval_loss": 0.30641260743141174,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -6.00722074508667,
+      "eval_rewards/margins": 4.2258148193359375,
+      "eval_rewards/rejected": -10.23303508758545,
+      "eval_runtime": 18.6032,
+      "eval_samples_per_second": 27.415,
+      "eval_steps_per_second": 3.44,
+      "step": 1300
+    },
+    {
+      "epoch": 1.6628571428571428,
+      "grad_norm": 4.412789821624756,
+      "learning_rate": 1.95342121028749e-07,
+      "logits/chosen": 5.839550971984863,
+      "logits/rejected": 4.976950168609619,
+      "logps/chosen": -4.390134334564209,
+      "logps/rejected": -7.591056823730469,
+      "loss": 0.2899,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -6.585201263427734,
+      "rewards/margins": 4.801384449005127,
+      "rewards/rejected": -11.386585235595703,
+      "step": 1310
+    },
+    {
+      "epoch": 1.6755555555555555,
+      "grad_norm": 2.8668479919433594,
+      "learning_rate": 1.7555878527937164e-07,
+      "logits/chosen": 5.967339992523193,
+      "logits/rejected": 5.515614986419678,
+      "logps/chosen": -4.793093204498291,
+      "logps/rejected": -7.345193386077881,
+      "loss": 0.3251,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -7.189639091491699,
+      "rewards/margins": 3.8281517028808594,
+      "rewards/rejected": -11.017791748046875,
+      "step": 1320
+    },
+    {
+      "epoch": 1.6882539682539681,
+      "grad_norm": 3.966052293777466,
+      "learning_rate": 1.567950262702714e-07,
+      "logits/chosen": 6.094046592712402,
+      "logits/rejected": 5.1793212890625,
+      "logps/chosen": -4.440655708312988,
+      "logps/rejected": -7.3250579833984375,
+      "loss": 0.2771,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.660984039306641,
+      "rewards/margins": 4.326603412628174,
+      "rewards/rejected": -10.987586975097656,
+      "step": 1330
+    },
+    {
+      "epoch": 1.700952380952381,
+      "grad_norm": 7.9511566162109375,
+      "learning_rate": 1.3905907440629752e-07,
+      "logits/chosen": 5.2374444007873535,
+      "logits/rejected": 4.61051607131958,
+      "logps/chosen": -3.8392302989959717,
+      "logps/rejected": -6.958086967468262,
+      "loss": 0.3232,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.75884485244751,
+      "rewards/margins": 4.678285121917725,
+      "rewards/rejected": -10.437129020690918,
+      "step": 1340
+    },
+    {
+      "epoch": 1.7136507936507936,
+      "grad_norm": 4.378050804138184,
+      "learning_rate": 1.223587092621162e-07,
+      "logits/chosen": 5.540999412536621,
+      "logits/rejected": 4.931067943572998,
+      "logps/chosen": -4.15987491607666,
+      "logps/rejected": -7.16122579574585,
+      "loss": 0.2938,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.239811897277832,
+      "rewards/margins": 4.502026557922363,
+      "rewards/rejected": -10.741838455200195,
+      "step": 1350
+    },
+    {
+      "epoch": 1.7136507936507936,
+      "eval_logits/chosen": 5.3179707527160645,
+      "eval_logits/rejected": 4.558828353881836,
+      "eval_logps/chosen": -3.97402286529541,
+      "eval_logps/rejected": -6.798065662384033,
+      "eval_loss": 0.30637431144714355,
+      "eval_rewards/accuracies": 0.890625,
+      "eval_rewards/chosen": -5.961033821105957,
+      "eval_rewards/margins": 4.236064434051514,
+      "eval_rewards/rejected": -10.197098731994629,
+      "eval_runtime": 18.6057,
+      "eval_samples_per_second": 27.411,
+      "eval_steps_per_second": 3.44,
+      "step": 1350
+    }
+  ],
+  "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": 3.2794440242802196e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-1350/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5596c575b3ac9598d22b9072a75c78c8a21b1561e291767fb4db4d3d0245393
+size 7224

checkpoint-1350/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)