Training in progress, step 850, checkpoint

checkpoint-850/README.md

@@ -0,0 +1,202 @@
+---
+base_model: microsoft/Phi-3-mini-4k-instruct
+library_name: peft
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.14.0

checkpoint-850/adapter_config.json

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

checkpoint-850/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd4297aebfd94c1c658683f162b7ca02f1eea775fe99c7ee88c2ef914c2ac511
+size 25200088

checkpoint-850/added_tokens.json

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

checkpoint-850/global_step850/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:05dc2188bb205277dd25544af9d3fd4736b2447f6e1a0b4b4332ddce1b9d7ac1
+size 18881328

checkpoint-850/global_step850/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb7a56c5b271531dd1a6c093235037c22a98c89925510c607357199f141866bd
+size 18881328

checkpoint-850/global_step850/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6066d45bba1674360695aa81fe991799f08ef889b2b6ebf8a780c438de4f8fe2
+size 18881328

checkpoint-850/global_step850/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44245b5d0ac98c26fcae8d5387af87299dae96872b5528161961abfcbc1bdb58
+size 18881392

checkpoint-850/global_step850/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9476925b4cc0bd0434c576680a776d33199eca413a202664e1dcb3173459260
+size 18881392

checkpoint-850/global_step850/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21e583934930498084027cc3e9a33d84031430ac2976d3c4b52f6561afb6beaf
+size 18881392

checkpoint-850/global_step850/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a80466e35bf3e64240ccc355f18a1e65f8d390e54b14a8e54a2c9f1bba71b2a
+size 18881392

checkpoint-850/global_step850/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16e2ac9e09b1aed5a04ffedea66bc841790a45c78c595cee74f769c8882c9808
+size 18881392

checkpoint-850/global_step850/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c95717ea47d0c52c157b716dae51ebdc04f1d210827e8d5cf1f5d14f22afcc95
+size 25379244

checkpoint-850/latest

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

checkpoint-850/rng_state_0.pth

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

checkpoint-850/rng_state_1.pth

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

checkpoint-850/rng_state_2.pth

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

checkpoint-850/rng_state_3.pth

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

checkpoint-850/rng_state_4.pth

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

checkpoint-850/rng_state_5.pth

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

checkpoint-850/rng_state_6.pth

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

checkpoint-850/rng_state_7.pth

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

checkpoint-850/scheduler.pt

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

checkpoint-850/special_tokens_map.json

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

checkpoint-850/tokenizer.model

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

checkpoint-850/tokenizer_config.json

@@ -0,0 +1,133 @@
+{
+  "add_bos_token": false,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": false
+    },
+    "32000": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32001": {
+      "content": "<|assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32002": {
+      "content": "<|placeholder1|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32003": {
+      "content": "<|placeholder2|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32004": {
+      "content": "<|placeholder3|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32005": {
+      "content": "<|placeholder4|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32006": {
+      "content": "<|system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32007": {
+      "content": "<|end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "32008": {
+      "content": "<|placeholder5|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32009": {
+      "content": "<|placeholder6|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    },
+    "32010": {
+      "content": "<|user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": true,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "chat_template": "{% set system_message = 'You are a helpful AI assistant.' %}{% if messages[0]['role'] == 'system' %}{% set system_message = messages[0]['content'] %}{% endif %}{% if system_message is defined %}{{ '<s>' + '<|system|>\n' + system_message + '<|end|>\n' }}{% endif %}{% for message in messages %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ '<|user|>\n' + content + '<|end|>\n<|assistant|>\n' }}{% elif message['role'] == 'assistant' %}{{ content + '<|end|>' + '\n' }}{% endif %}{% endfor %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|end|>",
+  "extra_special_tokens": {},
+  "legacy": false,
+  "model_max_length": 4096,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "split_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

checkpoint-850/trainer_state.json

@@ -0,0 +1,1580 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.730554361839278,
+  "eval_steps": 50,
+  "global_step": 850,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008594757198109154,
+      "grad_norm": 0.06708361208438873,
+      "learning_rate": 4.999451708687114e-06,
+      "logits/chosen": 14.524938583374023,
+      "logits/rejected": 14.82593822479248,
+      "logps/chosen": -0.31433865427970886,
+      "logps/rejected": -0.32406437397003174,
+      "loss": 0.9442,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": -0.4715079367160797,
+      "rewards/margins": 0.014588532969355583,
+      "rewards/rejected": -0.48609647154808044,
+      "step": 10
+    },
+    {
+      "epoch": 0.017189514396218308,
+      "grad_norm": 0.056814808398485184,
+      "learning_rate": 4.997807075247147e-06,
+      "logits/chosen": 14.309213638305664,
+      "logits/rejected": 14.978128433227539,
+      "logps/chosen": -0.31283506751060486,
+      "logps/rejected": -0.3911947011947632,
+      "loss": 0.928,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.46925264596939087,
+      "rewards/margins": 0.1175394207239151,
+      "rewards/rejected": -0.5867919921875,
+      "step": 20
+    },
+    {
+      "epoch": 0.02578427159432746,
+      "grad_norm": 0.061199307441711426,
+      "learning_rate": 4.9950668210706795e-06,
+      "logits/chosen": 14.68384075164795,
+      "logits/rejected": 15.338122367858887,
+      "logps/chosen": -0.3007296621799469,
+      "logps/rejected": -0.3204456865787506,
+      "loss": 0.9439,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.45109447836875916,
+      "rewards/margins": 0.029573997482657433,
+      "rewards/rejected": -0.48066848516464233,
+      "step": 30
+    },
+    {
+      "epoch": 0.034379028792436615,
+      "grad_norm": 0.08423774689435959,
+      "learning_rate": 4.9912321481237616e-06,
+      "logits/chosen": 14.39265251159668,
+      "logits/rejected": 15.059102058410645,
+      "logps/chosen": -0.28216058015823364,
+      "logps/rejected": -0.33495840430259705,
+      "loss": 0.9184,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": -0.42324090003967285,
+      "rewards/margins": 0.07919676601886749,
+      "rewards/rejected": -0.5024376511573792,
+      "step": 40
+    },
+    {
+      "epoch": 0.042973785990545764,
+      "grad_norm": 0.06052614375948906,
+      "learning_rate": 4.986304738420684e-06,
+      "logits/chosen": 14.383735656738281,
+      "logits/rejected": 15.029413223266602,
+      "logps/chosen": -0.27970507740974426,
+      "logps/rejected": -0.33213528990745544,
+      "loss": 0.9317,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.4195576310157776,
+      "rewards/margins": 0.07864536345005035,
+      "rewards/rejected": -0.49820294976234436,
+      "step": 50
+    },
+    {
+      "epoch": 0.042973785990545764,
+      "eval_logits/chosen": 14.424538612365723,
+      "eval_logits/rejected": 15.006633758544922,
+      "eval_logps/chosen": -0.2923925220966339,
+      "eval_logps/rejected": -0.3531996011734009,
+      "eval_loss": 0.9324354529380798,
+      "eval_rewards/accuracies": 0.5052631497383118,
+      "eval_rewards/chosen": -0.43858882784843445,
+      "eval_rewards/margins": 0.09121060371398926,
+      "eval_rewards/rejected": -0.5297994017601013,
+      "eval_runtime": 26.3759,
+      "eval_samples_per_second": 28.549,
+      "eval_steps_per_second": 3.602,
+      "step": 50
+    },
+    {
+      "epoch": 0.05156854318865492,
+      "grad_norm": 0.06899414211511612,
+      "learning_rate": 4.980286753286196e-06,
+      "logits/chosen": 14.888933181762695,
+      "logits/rejected": 15.33955192565918,
+      "logps/chosen": -0.2886829972267151,
+      "logps/rejected": -0.34016504883766174,
+      "loss": 0.9323,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.43302449584007263,
+      "rewards/margins": 0.07722309231758118,
+      "rewards/rejected": -0.5102475881576538,
+      "step": 60
+    },
+    {
+      "epoch": 0.060163300386764075,
+      "grad_norm": 0.06679105013608932,
+      "learning_rate": 4.973180832407471e-06,
+      "logits/chosen": 14.721624374389648,
+      "logits/rejected": 15.614666938781738,
+      "logps/chosen": -0.29435139894485474,
+      "logps/rejected": -0.38699784874916077,
+      "loss": 0.9172,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.4415270686149597,
+      "rewards/margins": 0.13896968960762024,
+      "rewards/rejected": -0.5804967880249023,
+      "step": 70
+    },
+    {
+      "epoch": 0.06875805758487323,
+      "grad_norm": 0.07169903814792633,
+      "learning_rate": 4.964990092676263e-06,
+      "logits/chosen": 13.848808288574219,
+      "logits/rejected": 14.609800338745117,
+      "logps/chosen": -0.26156893372535706,
+      "logps/rejected": -0.33030644059181213,
+      "loss": 0.9245,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.3923533856868744,
+      "rewards/margins": 0.10310628265142441,
+      "rewards/rejected": -0.495459645986557,
+      "step": 80
+    },
+    {
+      "epoch": 0.07735281478298238,
+      "grad_norm": 0.06593246012926102,
+      "learning_rate": 4.9557181268217225e-06,
+      "logits/chosen": 14.603567123413086,
+      "logits/rejected": 14.994171142578125,
+      "logps/chosen": -0.3191321790218353,
+      "logps/rejected": -0.3477073311805725,
+      "loss": 0.9359,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.4786983132362366,
+      "rewards/margins": 0.042862698435783386,
+      "rewards/rejected": -0.5215609669685364,
+      "step": 90
+    },
+    {
+      "epoch": 0.08594757198109153,
+      "grad_norm": 0.0718066617846489,
+      "learning_rate": 4.9453690018345144e-06,
+      "logits/chosen": 13.928094863891602,
+      "logits/rejected": 14.792709350585938,
+      "logps/chosen": -0.24115696549415588,
+      "logps/rejected": -0.3537539839744568,
+      "loss": 0.9066,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.3617354929447174,
+      "rewards/margins": 0.16889554262161255,
+      "rewards/rejected": -0.5306310653686523,
+      "step": 100
+    },
+    {
+      "epoch": 0.08594757198109153,
+      "eval_logits/chosen": 14.40036392211914,
+      "eval_logits/rejected": 14.97786808013916,
+      "eval_logps/chosen": -0.2777771055698395,
+      "eval_logps/rejected": -0.3516874611377716,
+      "eval_loss": 0.9236211180686951,
+      "eval_rewards/accuracies": 0.5052631497383118,
+      "eval_rewards/chosen": -0.4166657328605652,
+      "eval_rewards/margins": 0.11086549609899521,
+      "eval_rewards/rejected": -0.5275312066078186,
+      "eval_runtime": 25.8056,
+      "eval_samples_per_second": 29.18,
+      "eval_steps_per_second": 3.681,
+      "step": 100
+    },
+    {
+      "epoch": 0.09454232917920069,
+      "grad_norm": 0.06681054830551147,
+      "learning_rate": 4.933947257182901e-06,
+      "logits/chosen": 14.76116943359375,
+      "logits/rejected": 15.001077651977539,
+      "logps/chosen": -0.297056645154953,
+      "logps/rejected": -0.3221590518951416,
+      "loss": 0.929,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.4455850124359131,
+      "rewards/margins": 0.03765357658267021,
+      "rewards/rejected": -0.4832385182380676,
+      "step": 110
+    },
+    {
+      "epoch": 0.10313708637730984,
+      "grad_norm": 0.10024584829807281,
+      "learning_rate": 4.921457902821578e-06,
+      "logits/chosen": 14.405306816101074,
+      "logits/rejected": 15.084524154663086,
+      "logps/chosen": -0.2726767361164093,
+      "logps/rejected": -0.3543504774570465,
+      "loss": 0.9299,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.40901508927345276,
+      "rewards/margins": 0.12251058965921402,
+      "rewards/rejected": -0.531525731086731,
+      "step": 120
+    },
+    {
+      "epoch": 0.11173184357541899,
+      "grad_norm": 0.08629737794399261,
+      "learning_rate": 4.907906416994146e-06,
+      "logits/chosen": 14.073992729187012,
+      "logits/rejected": 14.882128715515137,
+      "logps/chosen": -0.2827032506465912,
+      "logps/rejected": -0.369393527507782,
+      "loss": 0.9109,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.42405492067337036,
+      "rewards/margins": 0.13003548979759216,
+      "rewards/rejected": -0.5540903806686401,
+      "step": 130
+    },
+    {
+      "epoch": 0.12032660077352815,
+      "grad_norm": 0.07973086833953857,
+      "learning_rate": 4.893298743830168e-06,
+      "logits/chosen": 13.96656322479248,
+      "logits/rejected": 14.639463424682617,
+      "logps/chosen": -0.28426361083984375,
+      "logps/rejected": -0.3899250030517578,
+      "loss": 0.9138,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.4263954162597656,
+      "rewards/margins": 0.1584920585155487,
+      "rewards/rejected": -0.5848874449729919,
+      "step": 140
+    },
+    {
+      "epoch": 0.1289213579716373,
+      "grad_norm": 0.08767445385456085,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": 13.705177307128906,
+      "logits/rejected": 14.19865608215332,
+      "logps/chosen": -0.26735779643058777,
+      "logps/rejected": -0.34726911783218384,
+      "loss": 0.9157,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.40103667974472046,
+      "rewards/margins": 0.1198669821023941,
+      "rewards/rejected": -0.5209037065505981,
+      "step": 150
+    },
+    {
+      "epoch": 0.1289213579716373,
+      "eval_logits/chosen": 13.20260238647461,
+      "eval_logits/rejected": 13.959339141845703,
+      "eval_logps/chosen": -0.27623170614242554,
+      "eval_logps/rejected": -0.3724917769432068,
+      "eval_loss": 0.909102737903595,
+      "eval_rewards/accuracies": 0.557894766330719,
+      "eval_rewards/chosen": -0.4143475592136383,
+      "eval_rewards/margins": 0.14439010620117188,
+      "eval_rewards/rejected": -0.5587376356124878,
+      "eval_runtime": 25.7839,
+      "eval_samples_per_second": 29.204,
+      "eval_steps_per_second": 3.684,
+      "step": 150
+    },
+    {
+      "epoch": 0.13751611516974646,
+      "grad_norm": 0.09749539196491241,
+      "learning_rate": 4.860940925593703e-06,
+      "logits/chosen": 13.301411628723145,
+      "logits/rejected": 14.054819107055664,
+      "logps/chosen": -0.2808162569999695,
+      "logps/rejected": -0.39500662684440613,
+      "loss": 0.9,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.4212244153022766,
+      "rewards/margins": 0.17128555476665497,
+      "rewards/rejected": -0.592509925365448,
+      "step": 160
+    },
+    {
+      "epoch": 0.1461108723678556,
+      "grad_norm": 0.14965052902698517,
+      "learning_rate": 4.84320497372973e-06,
+      "logits/chosen": 12.261284828186035,
+      "logits/rejected": 13.0617036819458,
+      "logps/chosen": -0.29266461730003357,
+      "logps/rejected": -0.4265298843383789,
+      "loss": 0.896,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.43899694085121155,
+      "rewards/margins": 0.20079784095287323,
+      "rewards/rejected": -0.6397948265075684,
+      "step": 170
+    },
+    {
+      "epoch": 0.15470562956596476,
+      "grad_norm": 0.13044072687625885,
+      "learning_rate": 4.824441214720629e-06,
+      "logits/chosen": 11.509119033813477,
+      "logits/rejected": 12.31033706665039,
+      "logps/chosen": -0.27384257316589355,
+      "logps/rejected": -0.3920982778072357,
+      "loss": 0.8911,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.4107638895511627,
+      "rewards/margins": 0.17738358676433563,
+      "rewards/rejected": -0.5881474018096924,
+      "step": 180
+    },
+    {
+      "epoch": 0.1633003867640739,
+      "grad_norm": 0.16182811558246613,
+      "learning_rate": 4.804657878971252e-06,
+      "logits/chosen": 10.68933391571045,
+      "logits/rejected": 11.632065773010254,
+      "logps/chosen": -0.292975515127182,
+      "logps/rejected": -0.42257896065711975,
+      "loss": 0.9002,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.4394632875919342,
+      "rewards/margins": 0.19440510869026184,
+      "rewards/rejected": -0.633868396282196,
+      "step": 190
+    },
+    {
+      "epoch": 0.17189514396218306,
+      "grad_norm": 0.181160107254982,
+      "learning_rate": 4.783863644106502e-06,
+      "logits/chosen": 10.593437194824219,
+      "logits/rejected": 11.435877799987793,
+      "logps/chosen": -0.32495418190956116,
+      "logps/rejected": -0.4480825364589691,
+      "loss": 0.8773,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.48743128776550293,
+      "rewards/margins": 0.18469250202178955,
+      "rewards/rejected": -0.6721237897872925,
+      "step": 200
+    },
+    {
+      "epoch": 0.17189514396218306,
+      "eval_logits/chosen": 9.299257278442383,
+      "eval_logits/rejected": 10.055145263671875,
+      "eval_logps/chosen": -0.31059205532073975,
+      "eval_logps/rejected": -0.47102925181388855,
+      "eval_loss": 0.8721462488174438,
+      "eval_rewards/accuracies": 0.6105263233184814,
+      "eval_rewards/chosen": -0.4658880829811096,
+      "eval_rewards/margins": 0.24065588414669037,
+      "eval_rewards/rejected": -0.7065439224243164,
+      "eval_runtime": 25.78,
+      "eval_samples_per_second": 29.209,
+      "eval_steps_per_second": 3.685,
+      "step": 200
+    },
+    {
+      "epoch": 0.18048990116029223,
+      "grad_norm": 0.24912959337234497,
+      "learning_rate": 4.762067631165049e-06,
+      "logits/chosen": 8.803088188171387,
+      "logits/rejected": 9.326388359069824,
+      "logps/chosen": -0.3249451816082001,
+      "logps/rejected": -0.44993042945861816,
+      "loss": 0.8484,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.4874177575111389,
+      "rewards/margins": 0.18747788667678833,
+      "rewards/rejected": -0.6748956441879272,
+      "step": 210
+    },
+    {
+      "epoch": 0.18908465835840138,
+      "grad_norm": 0.319579541683197,
+      "learning_rate": 4.7392794005985324e-06,
+      "logits/chosen": 6.257112979888916,
+      "logits/rejected": 7.168400764465332,
+      "logps/chosen": -0.335318386554718,
+      "logps/rejected": -0.5439311265945435,
+      "loss": 0.8499,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.5029775500297546,
+      "rewards/margins": 0.31291908025741577,
+      "rewards/rejected": -0.8158966302871704,
+      "step": 220
+    },
+    {
+      "epoch": 0.19767941555651053,
+      "grad_norm": 0.31494757533073425,
+      "learning_rate": 4.715508948078037e-06,
+      "logits/chosen": 5.725883960723877,
+      "logits/rejected": 5.9254865646362305,
+      "logps/chosen": -0.3735908567905426,
+      "logps/rejected": -0.5729750394821167,
+      "loss": 0.826,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.5603862404823303,
+      "rewards/margins": 0.2990763187408447,
+      "rewards/rejected": -0.859462559223175,
+      "step": 230
+    },
+    {
+      "epoch": 0.20627417275461968,
+      "grad_norm": 0.46439653635025024,
+      "learning_rate": 4.690766700109659e-06,
+      "logits/chosen": 5.059751033782959,
+      "logits/rejected": 5.128623008728027,
+      "logps/chosen": -0.4083784222602844,
+      "logps/rejected": -0.6792675852775574,
+      "loss": 0.7992,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.612567663192749,
+      "rewards/margins": 0.40633392333984375,
+      "rewards/rejected": -1.0189014673233032,
+      "step": 240
+    },
+    {
+      "epoch": 0.21486892995272883,
+      "grad_norm": 0.42406076192855835,
+      "learning_rate": 4.665063509461098e-06,
+      "logits/chosen": 4.128974437713623,
+      "logits/rejected": 4.141166687011719,
+      "logps/chosen": -0.4256651997566223,
+      "logps/rejected": -0.7279168367385864,
+      "loss": 0.7848,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.6384977102279663,
+      "rewards/margins": 0.4533773958683014,
+      "rewards/rejected": -1.0918750762939453,
+      "step": 250
+    },
+    {
+      "epoch": 0.21486892995272883,
+      "eval_logits/chosen": 3.800307512283325,
+      "eval_logits/rejected": 3.1472771167755127,
+      "eval_logps/chosen": -0.4563433527946472,
+      "eval_logps/rejected": -0.8247694373130798,
+      "eval_loss": 0.7728626728057861,
+      "eval_rewards/accuracies": 0.6526315808296204,
+      "eval_rewards/chosen": -0.6845150589942932,
+      "eval_rewards/margins": 0.5526391267776489,
+      "eval_rewards/rejected": -1.237154245376587,
+      "eval_runtime": 25.7836,
+      "eval_samples_per_second": 29.205,
+      "eval_steps_per_second": 3.685,
+      "step": 250
+    },
+    {
+      "epoch": 0.22346368715083798,
+      "grad_norm": 0.4071955680847168,
+      "learning_rate": 4.638410650401267e-06,
+      "logits/chosen": 3.169527530670166,
+      "logits/rejected": 2.603461503982544,
+      "logps/chosen": -0.5029922723770142,
+      "logps/rejected": -0.9469219446182251,
+      "loss": 0.7273,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.7544883489608765,
+      "rewards/margins": 0.6658946871757507,
+      "rewards/rejected": -1.4203828573226929,
+      "step": 260
+    },
+    {
+      "epoch": 0.23205844434894715,
+      "grad_norm": 0.6253886222839355,
+      "learning_rate": 4.610819813755038e-06,
+      "logits/chosen": 3.8718018531799316,
+      "logits/rejected": 2.569753646850586,
+      "logps/chosen": -0.4955294132232666,
+      "logps/rejected": -0.8811863660812378,
+      "loss": 0.7483,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.7432941198348999,
+      "rewards/margins": 0.5784854888916016,
+      "rewards/rejected": -1.321779489517212,
+      "step": 270
+    },
+    {
+      "epoch": 0.2406532015470563,
+      "grad_norm": 0.5592113733291626,
+      "learning_rate": 4.582303101775249e-06,
+      "logits/chosen": 3.4818286895751953,
+      "logits/rejected": 2.428328275680542,
+      "logps/chosen": -0.5700691342353821,
+      "logps/rejected": -1.010145664215088,
+      "loss": 0.7165,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.8551036715507507,
+      "rewards/margins": 0.6601148843765259,
+      "rewards/rejected": -1.5152184963226318,
+      "step": 280
+    },
+    {
+      "epoch": 0.24924795874516545,
+      "grad_norm": 0.8438608050346375,
+      "learning_rate": 4.55287302283426e-06,
+      "logits/chosen": 2.5937914848327637,
+      "logits/rejected": 1.8570162057876587,
+      "logps/chosen": -0.592321515083313,
+      "logps/rejected": -1.1775600910186768,
+      "loss": 0.6685,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.8884822130203247,
+      "rewards/margins": 0.8778578042984009,
+      "rewards/rejected": -1.7663400173187256,
+      "step": 290
+    },
+    {
+      "epoch": 0.2578427159432746,
+      "grad_norm": 2.9559757709503174,
+      "learning_rate": 4.522542485937369e-06,
+      "logits/chosen": 3.2419090270996094,
+      "logits/rejected": 1.9082870483398438,
+      "logps/chosen": -0.6832663416862488,
+      "logps/rejected": -1.5631868839263916,
+      "loss": 0.6009,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -1.0248994827270508,
+      "rewards/margins": 1.3198809623718262,
+      "rewards/rejected": -2.344780445098877,
+      "step": 300
+    },
+    {
+      "epoch": 0.2578427159432746,
+      "eval_logits/chosen": 2.5470504760742188,
+      "eval_logits/rejected": 1.492888331413269,
+      "eval_logps/chosen": -0.7285813689231873,
+      "eval_logps/rejected": -1.8318607807159424,
+      "eval_loss": 0.5855891704559326,
+      "eval_rewards/accuracies": 0.7052631378173828,
+      "eval_rewards/chosen": -1.092872142791748,
+      "eval_rewards/margins": 1.6549187898635864,
+      "eval_rewards/rejected": -2.747790813446045,
+      "eval_runtime": 25.8105,
+      "eval_samples_per_second": 29.174,
+      "eval_steps_per_second": 3.681,
+      "step": 300
+    },
+    {
+      "epoch": 0.2664374731413838,
+      "grad_norm": 1.4503060579299927,
+      "learning_rate": 4.491324795060491e-06,
+      "logits/chosen": 1.6672757863998413,
+      "logits/rejected": 0.7888604402542114,
+      "logps/chosen": -0.769140899181366,
+      "logps/rejected": -2.0822532176971436,
+      "loss": 0.512,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.1537113189697266,
+      "rewards/margins": 1.9696683883666992,
+      "rewards/rejected": -3.123379945755005,
+      "step": 310
+    },
+    {
+      "epoch": 0.2750322303394929,
+      "grad_norm": 0.36741188168525696,
+      "learning_rate": 4.4592336433146e-06,
+      "logits/chosen": 2.6584715843200684,
+      "logits/rejected": 1.835911750793457,
+      "logps/chosen": -0.8400143384933472,
+      "logps/rejected": -1.9262489080429077,
+      "loss": 0.5405,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -1.2600215673446655,
+      "rewards/margins": 1.6293519735336304,
+      "rewards/rejected": -2.889373302459717,
+      "step": 320
+    },
+    {
+      "epoch": 0.28362698753760207,
+      "grad_norm": 0.6233783960342407,
+      "learning_rate": 4.426283106939474e-06,
+      "logits/chosen": 3.2203617095947266,
+      "logits/rejected": 2.3215420246124268,
+      "logps/chosen": -0.7985933423042297,
+      "logps/rejected": -2.4170174598693848,
+      "loss": 0.5335,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -1.197890043258667,
+      "rewards/margins": 2.427635669708252,
+      "rewards/rejected": -3.625525712966919,
+      "step": 330
+    },
+    {
+      "epoch": 0.2922217447357112,
+      "grad_norm": 1.0881849527359009,
+      "learning_rate": 4.3924876391293915e-06,
+      "logits/chosen": 2.229017734527588,
+      "logits/rejected": 1.2251309156417847,
+      "logps/chosen": -0.8058193325996399,
+      "logps/rejected": -2.810622215270996,
+      "loss": 0.4903,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.2087291479110718,
+      "rewards/margins": 3.007204294204712,
+      "rewards/rejected": -4.215933799743652,
+      "step": 340
+    },
+    {
+      "epoch": 0.30081650193382037,
+      "grad_norm": 4.168415069580078,
+      "learning_rate": 4.357862063693486e-06,
+      "logits/chosen": 2.4198296070098877,
+      "logits/rejected": 1.5391919612884521,
+      "logps/chosen": -1.010558843612671,
+      "logps/rejected": -2.2362923622131348,
+      "loss": 0.5249,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -1.515838384628296,
+      "rewards/margins": 1.8385999202728271,
+      "rewards/rejected": -3.3544387817382812,
+      "step": 350
+    },
+    {
+      "epoch": 0.30081650193382037,
+      "eval_logits/chosen": 2.996535539627075,
+      "eval_logits/rejected": 2.064058303833008,
+      "eval_logps/chosen": -0.8687878847122192,
+      "eval_logps/rejected": -2.9790267944335938,
+      "eval_loss": 0.5171241760253906,
+      "eval_rewards/accuracies": 0.7263157963752747,
+      "eval_rewards/chosen": -1.3031818866729736,
+      "eval_rewards/margins": 3.165358781814575,
+      "eval_rewards/rejected": -4.468540668487549,
+      "eval_runtime": 25.8152,
+      "eval_samples_per_second": 29.169,
+      "eval_steps_per_second": 3.68,
+      "step": 350
+    },
+    {
+      "epoch": 0.3094112591319295,
+      "grad_norm": 0.5646592378616333,
+      "learning_rate": 4.322421568553529e-06,
+      "logits/chosen": 3.050445556640625,
+      "logits/rejected": 2.0960793495178223,
+      "logps/chosen": -0.7702202796936035,
+      "logps/rejected": -2.5967533588409424,
+      "loss": 0.5067,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.1553303003311157,
+      "rewards/margins": 2.739799737930298,
+      "rewards/rejected": -3.895130157470703,
+      "step": 360
+    },
+    {
+      "epoch": 0.31800601633003867,
+      "grad_norm": 0.5547713041305542,
+      "learning_rate": 4.286181699082008e-06,
+      "logits/chosen": 2.7148895263671875,
+      "logits/rejected": 1.9958852529525757,
+      "logps/chosen": -0.9548311233520508,
+      "logps/rejected": -3.1348252296447754,
+      "loss": 0.4726,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -1.4322465658187866,
+      "rewards/margins": 3.269991397857666,
+      "rewards/rejected": -4.702237606048584,
+      "step": 370
+    },
+    {
+      "epoch": 0.3266007735281478,
+      "grad_norm": 3.4396660327911377,
+      "learning_rate": 4.249158351283414e-06,
+      "logits/chosen": 2.586766004562378,
+      "logits/rejected": 2.070089340209961,
+      "logps/chosen": -0.9903923273086548,
+      "logps/rejected": -3.0135743618011475,
+      "loss": 0.4801,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -1.485588550567627,
+      "rewards/margins": 3.034773349761963,
+      "rewards/rejected": -4.52036190032959,
+      "step": 380
+    },
+    {
+      "epoch": 0.33519553072625696,
+      "grad_norm": 0.9405317306518555,
+      "learning_rate": 4.211367764821722e-06,
+      "logits/chosen": 4.370789527893066,
+      "logits/rejected": 3.165931224822998,
+      "logps/chosen": -0.7785463929176331,
+      "logps/rejected": -2.456723928451538,
+      "loss": 0.4585,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -1.1678194999694824,
+      "rewards/margins": 2.5172665119171143,
+      "rewards/rejected": -3.6850857734680176,
+      "step": 390
+    },
+    {
+      "epoch": 0.3437902879243661,
+      "grad_norm": 0.7120731472969055,
+      "learning_rate": 4.172826515897146e-06,
+      "logits/chosen": 3.3425400257110596,
+      "logits/rejected": 2.6448545455932617,
+      "logps/chosen": -0.9174768328666687,
+      "logps/rejected": -3.047037124633789,
+      "loss": 0.4771,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.3762153387069702,
+      "rewards/margins": 3.194340229034424,
+      "rewards/rejected": -4.570555686950684,
+      "step": 400
+    },
+    {
+      "epoch": 0.3437902879243661,
+      "eval_logits/chosen": 3.548964262008667,
+      "eval_logits/rejected": 2.7726428508758545,
+      "eval_logps/chosen": -1.0053316354751587,
+      "eval_logps/rejected": -3.487654447555542,
+      "eval_loss": 0.47841358184814453,
+      "eval_rewards/accuracies": 0.7368420958518982,
+      "eval_rewards/chosen": -1.5079973936080933,
+      "eval_rewards/margins": 3.723484992980957,
+      "eval_rewards/rejected": -5.231482028961182,
+      "eval_runtime": 25.8148,
+      "eval_samples_per_second": 29.169,
+      "eval_steps_per_second": 3.68,
+      "step": 400
+    },
+    {
+      "epoch": 0.3523850451224753,
+      "grad_norm": 2.403956651687622,
+      "learning_rate": 4.133551509975264e-06,
+      "logits/chosen": 3.2028489112854004,
+      "logits/rejected": 2.2486982345581055,
+      "logps/chosen": -0.9957242012023926,
+      "logps/rejected": -3.243959426879883,
+      "loss": 0.4449,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -1.4935863018035889,
+      "rewards/margins": 3.3723526000976562,
+      "rewards/rejected": -4.865939140319824,
+      "step": 410
+    },
+    {
+      "epoch": 0.36097980232058446,
+      "grad_norm": 0.39530256390571594,
+      "learning_rate": 4.093559974371725e-06,
+      "logits/chosen": 3.8590214252471924,
+      "logits/rejected": 3.1420931816101074,
+      "logps/chosen": -0.9541120529174805,
+      "logps/rejected": -3.0112829208374023,
+      "loss": 0.4598,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -1.4311680793762207,
+      "rewards/margins": 3.085756301879883,
+      "rewards/rejected": -4.5169243812561035,
+      "step": 420
+    },
+    {
+      "epoch": 0.3695745595186936,
+      "grad_norm": 0.29451707005500793,
+      "learning_rate": 4.052869450695776e-06,
+      "logits/chosen": 4.697268486022949,
+      "logits/rejected": 3.7647697925567627,
+      "logps/chosen": -1.1037578582763672,
+      "logps/rejected": -3.8626160621643066,
+      "loss": 0.4275,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.6556367874145508,
+      "rewards/margins": 4.138287544250488,
+      "rewards/rejected": -5.793923854827881,
+      "step": 430
+    },
+    {
+      "epoch": 0.37816931671680276,
+      "grad_norm": 0.5065125823020935,
+      "learning_rate": 4.011497787155938e-06,
+      "logits/chosen": 3.5233864784240723,
+      "logits/rejected": 2.798567533493042,
+      "logps/chosen": -1.1753087043762207,
+      "logps/rejected": -4.171238899230957,
+      "loss": 0.4132,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.7629629373550415,
+      "rewards/margins": 4.493895053863525,
+      "rewards/rejected": -6.256857872009277,
+      "step": 440
+    },
+    {
+      "epoch": 0.3867640739149119,
+      "grad_norm": 1.414167881011963,
+      "learning_rate": 3.969463130731183e-06,
+      "logits/chosen": 4.733740329742432,
+      "logits/rejected": 4.114102363586426,
+      "logps/chosen": -1.1846634149551392,
+      "logps/rejected": -4.04649543762207,
+      "loss": 0.4266,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -1.776995301246643,
+      "rewards/margins": 4.292747974395752,
+      "rewards/rejected": -6.0697431564331055,
+      "step": 450
+    },
+    {
+      "epoch": 0.3867640739149119,
+      "eval_logits/chosen": 4.25229549407959,
+      "eval_logits/rejected": 3.900564193725586,
+      "eval_logps/chosen": -1.410205602645874,
+      "eval_logps/rejected": -4.276910781860352,
+      "eval_loss": 0.4397798478603363,
+      "eval_rewards/accuracies": 0.800000011920929,
+      "eval_rewards/chosen": -2.1153085231781006,
+      "eval_rewards/margins": 4.3000569343566895,
+      "eval_rewards/rejected": -6.415364742279053,
+      "eval_runtime": 25.7968,
+      "eval_samples_per_second": 29.19,
+      "eval_steps_per_second": 3.683,
+      "step": 450
+    },
+    {
+      "epoch": 0.39535883111302106,
+      "grad_norm": 1.7992101907730103,
+      "learning_rate": 3.92678391921108e-06,
+      "logits/chosen": 4.718934059143066,
+      "logits/rejected": 4.305315971374512,
+      "logps/chosen": -1.423595666885376,
+      "logps/rejected": -3.9873733520507812,
+      "loss": 0.367,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -2.1353936195373535,
+      "rewards/margins": 3.8456661701202393,
+      "rewards/rejected": -5.981060028076172,
+      "step": 460
+    },
+    {
+      "epoch": 0.4039535883111302,
+      "grad_norm": 1.9630879163742065,
+      "learning_rate": 3.88347887310836e-06,
+      "logits/chosen": 3.8371150493621826,
+      "logits/rejected": 3.5719306468963623,
+      "logps/chosen": -2.0386481285095215,
+      "logps/rejected": -4.779314994812012,
+      "loss": 0.3957,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -3.0579724311828613,
+      "rewards/margins": 4.1110005378723145,
+      "rewards/rejected": -7.168972969055176,
+      "step": 470
+    },
+    {
+      "epoch": 0.41254834550923936,
+      "grad_norm": 4.952139854431152,
+      "learning_rate": 3.839566987447492e-06,
+      "logits/chosen": 5.024113655090332,
+      "logits/rejected": 4.880651950836182,
+      "logps/chosen": -2.3612470626831055,
+      "logps/rejected": -4.691690444946289,
+      "loss": 0.3808,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -3.5418708324432373,
+      "rewards/margins": 3.495664596557617,
+      "rewards/rejected": -7.037535190582275,
+      "step": 480
+    },
+    {
+      "epoch": 0.4211431027073485,
+      "grad_norm": 2.832200527191162,
+      "learning_rate": 3.795067523432826e-06,
+      "logits/chosen": 4.689079284667969,
+      "logits/rejected": 4.435003280639648,
+      "logps/chosen": -2.437671184539795,
+      "logps/rejected": -5.003944396972656,
+      "loss": 0.3555,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -3.6565067768096924,
+      "rewards/margins": 3.849409580230713,
+      "rewards/rejected": -7.505916595458984,
+      "step": 490
+    },
+    {
+      "epoch": 0.42973785990545765,
+      "grad_norm": 3.2997682094573975,
+      "learning_rate": 3.7500000000000005e-06,
+      "logits/chosen": 4.286547660827637,
+      "logits/rejected": 4.181652545928955,
+      "logps/chosen": -2.3390612602233887,
+      "logps/rejected": -4.888935565948486,
+      "loss": 0.3211,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -3.508591890335083,
+      "rewards/margins": 3.8248119354248047,
+      "rewards/rejected": -7.333403587341309,
+      "step": 500
+    },
+    {
+      "epoch": 0.42973785990545765,
+      "eval_logits/chosen": 3.5323922634124756,
+      "eval_logits/rejected": 3.5116958618164062,
+      "eval_logps/chosen": -2.679356575012207,
+      "eval_logps/rejected": -5.927057266235352,
+      "eval_loss": 0.37026864290237427,
+      "eval_rewards/accuracies": 0.8736842274665833,
+      "eval_rewards/chosen": -4.019035339355469,
+      "eval_rewards/margins": 4.871551036834717,
+      "eval_rewards/rejected": -8.890586853027344,
+      "eval_runtime": 25.8768,
+      "eval_samples_per_second": 29.099,
+      "eval_steps_per_second": 3.671,
+      "step": 500
+    },
+    {
+      "epoch": 0.4383326171035668,
+      "grad_norm": 2.6194217205047607,
+      "learning_rate": 3.7043841852542884e-06,
+      "logits/chosen": 4.265946388244629,
+      "logits/rejected": 3.7863662242889404,
+      "logps/chosen": -2.279764413833618,
+      "logps/rejected": -4.825397968292236,
+      "loss": 0.3375,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -3.4196460247039795,
+      "rewards/margins": 3.818450927734375,
+      "rewards/rejected": -7.238096714019775,
+      "step": 510
+    },
+    {
+      "epoch": 0.44692737430167595,
+      "grad_norm": 4.434008598327637,
+      "learning_rate": 3.658240087799655e-06,
+      "logits/chosen": 3.4983534812927246,
+      "logits/rejected": 3.3109116554260254,
+      "logps/chosen": -2.7063632011413574,
+      "logps/rejected": -5.9535369873046875,
+      "loss": 0.3388,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.059545040130615,
+      "rewards/margins": 4.870760917663574,
+      "rewards/rejected": -8.930305480957031,
+      "step": 520
+    },
+    {
+      "epoch": 0.45552213149978515,
+      "grad_norm": 2.4193809032440186,
+      "learning_rate": 3.611587947962319e-06,
+      "logits/chosen": 3.4949145317077637,
+      "logits/rejected": 3.402980089187622,
+      "logps/chosen": -2.5146005153656006,
+      "logps/rejected": -5.63289737701416,
+      "loss": 0.3281,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -3.7719013690948486,
+      "rewards/margins": 4.677445411682129,
+      "rewards/rejected": -8.449346542358398,
+      "step": 530
+    },
+    {
+      "epoch": 0.4641168886978943,
+      "grad_norm": 10.788633346557617,
+      "learning_rate": 3.564448228912682e-06,
+      "logits/chosen": 3.3073112964630127,
+      "logits/rejected": 3.163470506668091,
+      "logps/chosen": -2.258653163909912,
+      "logps/rejected": -5.642867088317871,
+      "loss": 0.3565,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -3.387979507446289,
+      "rewards/margins": 5.076320648193359,
+      "rewards/rejected": -8.464300155639648,
+      "step": 540
+    },
+    {
+      "epoch": 0.47271164589600345,
+      "grad_norm": 1.6846323013305664,
+      "learning_rate": 3.516841607689501e-06,
+      "logits/chosen": 3.476361036300659,
+      "logits/rejected": 3.375828504562378,
+      "logps/chosen": -2.5325064659118652,
+      "logps/rejected": -5.799270153045654,
+      "loss": 0.3103,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -3.798759937286377,
+      "rewards/margins": 4.900145530700684,
+      "rewards/rejected": -8.698905944824219,
+      "step": 550
+    },
+    {
+      "epoch": 0.47271164589600345,
+      "eval_logits/chosen": 3.451749563217163,
+      "eval_logits/rejected": 3.3771002292633057,
+      "eval_logps/chosen": -2.9835667610168457,
+      "eval_logps/rejected": -6.5389509201049805,
+      "eval_loss": 0.32732319831848145,
+      "eval_rewards/accuracies": 0.9052631855010986,
+      "eval_rewards/chosen": -4.475350379943848,
+      "eval_rewards/margins": 5.333076000213623,
+      "eval_rewards/rejected": -9.808425903320312,
+      "eval_runtime": 25.8141,
+      "eval_samples_per_second": 29.17,
+      "eval_steps_per_second": 3.68,
+      "step": 550
+    },
+    {
+      "epoch": 0.4813064030941126,
+      "grad_norm": 2.7683331966400146,
+      "learning_rate": 3.4687889661302577e-06,
+      "logits/chosen": 2.468799591064453,
+      "logits/rejected": 2.4257254600524902,
+      "logps/chosen": -2.6801788806915283,
+      "logps/rejected": -6.136897087097168,
+      "loss": 0.3171,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.020268440246582,
+      "rewards/margins": 5.185078144073486,
+      "rewards/rejected": -9.205347061157227,
+      "step": 560
+    },
+    {
+      "epoch": 0.48990116029222175,
+      "grad_norm": 11.559685707092285,
+      "learning_rate": 3.4203113817116955e-06,
+      "logits/chosen": 3.6535427570343018,
+      "logits/rejected": 3.583962917327881,
+      "logps/chosen": -2.7984983921051025,
+      "logps/rejected": -6.468808650970459,
+      "loss": 0.331,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.197747707366943,
+      "rewards/margins": 5.505465984344482,
+      "rewards/rejected": -9.703214645385742,
+      "step": 570
+    },
+    {
+      "epoch": 0.4984959174903309,
+      "grad_norm": 7.691457271575928,
+      "learning_rate": 3.3714301183045382e-06,
+      "logits/chosen": 2.7467286586761475,
+      "logits/rejected": 2.4911317825317383,
+      "logps/chosen": -2.4207634925842285,
+      "logps/rejected": -6.385074138641357,
+      "loss": 0.2557,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.631145477294922,
+      "rewards/margins": 5.946464538574219,
+      "rewards/rejected": -9.577610969543457,
+      "step": 580
+    },
+    {
+      "epoch": 0.50709067468844,
+      "grad_norm": 5.381045341491699,
+      "learning_rate": 3.3221666168464584e-06,
+      "logits/chosen": 3.239227294921875,
+      "logits/rejected": 3.1982555389404297,
+      "logps/chosen": -3.2162883281707764,
+      "logps/rejected": -6.693168640136719,
+      "loss": 0.2921,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.824432373046875,
+      "rewards/margins": 5.215321063995361,
+      "rewards/rejected": -10.039752960205078,
+      "step": 590
+    },
+    {
+      "epoch": 0.5156854318865493,
+      "grad_norm": 2.616710901260376,
+      "learning_rate": 3.272542485937369e-06,
+      "logits/chosen": 2.7326784133911133,
+      "logits/rejected": 2.4248623847961426,
+      "logps/chosen": -3.001952648162842,
+      "logps/rejected": -6.597250461578369,
+      "loss": 0.3225,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.502928733825684,
+      "rewards/margins": 5.392947196960449,
+      "rewards/rejected": -9.89587688446045,
+      "step": 600
+    },
+    {
+      "epoch": 0.5156854318865493,
+      "eval_logits/chosen": 3.1995701789855957,
+      "eval_logits/rejected": 3.2761952877044678,
+      "eval_logps/chosen": -3.166325330734253,
+      "eval_logps/rejected": -7.062953472137451,
+      "eval_loss": 0.29786577820777893,
+      "eval_rewards/accuracies": 0.9263157844543457,
+      "eval_rewards/chosen": -4.74948787689209,
+      "eval_rewards/margins": 5.84494161605835,
+      "eval_rewards/rejected": -10.594429969787598,
+      "eval_runtime": 25.8074,
+      "eval_samples_per_second": 29.178,
+      "eval_steps_per_second": 3.681,
+      "step": 600
+    },
+    {
+      "epoch": 0.5242801890846583,
+      "grad_norm": 3.5550060272216797,
+      "learning_rate": 3.222579492361179e-06,
+      "logits/chosen": 2.8619818687438965,
+      "logits/rejected": 3.014125347137451,
+      "logps/chosen": -2.9217543601989746,
+      "logps/rejected": -6.349586486816406,
+      "loss": 0.2516,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.382631301879883,
+      "rewards/margins": 5.141747951507568,
+      "rewards/rejected": -9.52437973022461,
+      "step": 610
+    },
+    {
+      "epoch": 0.5328749462827675,
+      "grad_norm": 3.6020898818969727,
+      "learning_rate": 3.1722995515381644e-06,
+      "logits/chosen": 2.770552635192871,
+      "logits/rejected": 2.9711716175079346,
+      "logps/chosen": -2.8700039386749268,
+      "logps/rejected": -6.527164459228516,
+      "loss": 0.3062,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.305006504058838,
+      "rewards/margins": 5.485739707946777,
+      "rewards/rejected": -9.790745735168457,
+      "step": 620
+    },
+    {
+      "epoch": 0.5414697034808766,
+      "grad_norm": 2.2209339141845703,
+      "learning_rate": 3.121724717912138e-06,
+      "logits/chosen": 3.1150155067443848,
+      "logits/rejected": 3.038687229156494,
+      "logps/chosen": -2.9821603298187256,
+      "logps/rejected": -6.432187080383301,
+      "loss": 0.2726,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.473240852355957,
+      "rewards/margins": 5.175040245056152,
+      "rewards/rejected": -9.648280143737793,
+      "step": 630
+    },
+    {
+      "epoch": 0.5500644606789858,
+      "grad_norm": 2.148709297180176,
+      "learning_rate": 3.0708771752766397e-06,
+      "logits/chosen": 3.256204605102539,
+      "logits/rejected": 2.9765384197235107,
+      "logps/chosen": -2.7979862689971924,
+      "logps/rejected": -6.7102813720703125,
+      "loss": 0.2861,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.196979522705078,
+      "rewards/margins": 5.868442535400391,
+      "rewards/rejected": -10.065422058105469,
+      "step": 640
+    },
+    {
+      "epoch": 0.5586592178770949,
+      "grad_norm": 7.988170623779297,
+      "learning_rate": 3.019779227044398e-06,
+      "logits/chosen": 2.7312607765197754,
+      "logits/rejected": 2.705409526824951,
+      "logps/chosen": -2.8119194507598877,
+      "logps/rejected": -6.627874851226807,
+      "loss": 0.2507,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.217879295349121,
+      "rewards/margins": 5.723933219909668,
+      "rewards/rejected": -9.941811561584473,
+      "step": 650
+    },
+    {
+      "epoch": 0.5586592178770949,
+      "eval_logits/chosen": 2.9934144020080566,
+      "eval_logits/rejected": 3.124950647354126,
+      "eval_logps/chosen": -3.1934540271759033,
+      "eval_logps/rejected": -7.429007053375244,
+      "eval_loss": 0.28581172227859497,
+      "eval_rewards/accuracies": 0.9368420839309692,
+      "eval_rewards/chosen": -4.7901811599731445,
+      "eval_rewards/margins": 6.353330135345459,
+      "eval_rewards/rejected": -11.143510818481445,
+      "eval_runtime": 25.8075,
+      "eval_samples_per_second": 29.178,
+      "eval_steps_per_second": 3.681,
+      "step": 650
+    },
+    {
+      "epoch": 0.5672539750752041,
+      "grad_norm": 3.133023500442505,
+      "learning_rate": 2.9684532864643123e-06,
+      "logits/chosen": 3.3388328552246094,
+      "logits/rejected": 3.3477108478546143,
+      "logps/chosen": -3.3007023334503174,
+      "logps/rejected": -6.496421813964844,
+      "loss": 0.2635,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.951053619384766,
+      "rewards/margins": 4.793579578399658,
+      "rewards/rejected": -9.744632720947266,
+      "step": 660
+    },
+    {
+      "epoch": 0.5758487322733132,
+      "grad_norm": 3.6694588661193848,
+      "learning_rate": 2.9169218667902562e-06,
+      "logits/chosen": 3.1316323280334473,
+      "logits/rejected": 2.8977527618408203,
+      "logps/chosen": -2.8769032955169678,
+      "logps/rejected": -5.834782123565674,
+      "loss": 0.2854,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.31535530090332,
+      "rewards/margins": 4.4368181228637695,
+      "rewards/rejected": -8.752172470092773,
+      "step": 670
+    },
+    {
+      "epoch": 0.5844434894714224,
+      "grad_norm": 3.409086227416992,
+      "learning_rate": 2.8652075714060296e-06,
+      "logits/chosen": 2.726431369781494,
+      "logits/rejected": 2.9928715229034424,
+      "logps/chosen": -3.0180654525756836,
+      "logps/rejected": -7.160694122314453,
+      "loss": 0.2461,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.527098178863525,
+      "rewards/margins": 6.213942050933838,
+      "rewards/rejected": -10.741040229797363,
+      "step": 680
+    },
+    {
+      "epoch": 0.5930382466695315,
+      "grad_norm": 5.1549153327941895,
+      "learning_rate": 2.813333083910761e-06,
+      "logits/chosen": 3.371833086013794,
+      "logits/rejected": 3.317675828933716,
+      "logps/chosen": -3.1700775623321533,
+      "logps/rejected": -6.8295745849609375,
+      "loss": 0.2458,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.755115985870361,
+      "rewards/margins": 5.489245891571045,
+      "rewards/rejected": -10.244361877441406,
+      "step": 690
+    },
+    {
+      "epoch": 0.6016330038676407,
+      "grad_norm": 4.8320112228393555,
+      "learning_rate": 2.761321158169134e-06,
+      "logits/chosen": 2.5933046340942383,
+      "logits/rejected": 2.7114923000335693,
+      "logps/chosen": -3.2017345428466797,
+      "logps/rejected": -7.420884609222412,
+      "loss": 0.2464,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.8026018142700195,
+      "rewards/margins": 6.328725814819336,
+      "rewards/rejected": -11.131326675415039,
+      "step": 700
+    },
+    {
+      "epoch": 0.6016330038676407,
+      "eval_logits/chosen": 2.7770564556121826,
+      "eval_logits/rejected": 2.9842188358306885,
+      "eval_logps/chosen": -3.5835421085357666,
+      "eval_logps/rejected": -8.199637413024902,
+      "eval_loss": 0.26223084330558777,
+      "eval_rewards/accuracies": 0.9263157844543457,
+      "eval_rewards/chosen": -5.375312805175781,
+      "eval_rewards/margins": 6.924142360687256,
+      "eval_rewards/rejected": -12.299455642700195,
+      "eval_runtime": 25.8482,
+      "eval_samples_per_second": 29.132,
+      "eval_steps_per_second": 3.675,
+      "step": 700
+    },
+    {
+      "epoch": 0.6102277610657499,
+      "grad_norm": 3.6024580001831055,
+      "learning_rate": 2.70919460833079e-06,
+      "logits/chosen": 2.4713566303253174,
+      "logits/rejected": 2.523773193359375,
+      "logps/chosen": -2.962407112121582,
+      "logps/rejected": -7.556809902191162,
+      "loss": 0.2785,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -4.443611145019531,
+      "rewards/margins": 6.891604423522949,
+      "rewards/rejected": -11.33521556854248,
+      "step": 710
+    },
+    {
+      "epoch": 0.618822518263859,
+      "grad_norm": 3.6402506828308105,
+      "learning_rate": 2.6569762988232838e-06,
+      "logits/chosen": 3.112736940383911,
+      "logits/rejected": 2.901930809020996,
+      "logps/chosen": -3.1669626235961914,
+      "logps/rejected": -7.049059867858887,
+      "loss": 0.2623,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.750443935394287,
+      "rewards/margins": 5.823145866394043,
+      "rewards/rejected": -10.573590278625488,
+      "step": 720
+    },
+    {
+      "epoch": 0.6274172754619682,
+      "grad_norm": 9.418655395507812,
+      "learning_rate": 2.604689134322999e-06,
+      "logits/chosen": 2.8671321868896484,
+      "logits/rejected": 2.763396739959717,
+      "logps/chosen": -3.1475024223327637,
+      "logps/rejected": -7.338767051696777,
+      "loss": 0.2702,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.721253871917725,
+      "rewards/margins": 6.286896705627441,
+      "rewards/rejected": -11.008151054382324,
+      "step": 730
+    },
+    {
+      "epoch": 0.6360120326600773,
+      "grad_norm": 6.016907215118408,
+      "learning_rate": 2.5523560497083927e-06,
+      "logits/chosen": 2.8831398487091064,
+      "logits/rejected": 2.9516844749450684,
+      "logps/chosen": -3.055087089538574,
+      "logps/rejected": -7.092196464538574,
+      "loss": 0.2103,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.5826311111450195,
+      "rewards/margins": 6.0556640625,
+      "rewards/rejected": -10.638293266296387,
+      "step": 740
+    },
+    {
+      "epoch": 0.6446067898581865,
+      "grad_norm": 2.791388988494873,
+      "learning_rate": 2.5e-06,
+      "logits/chosen": 3.5676627159118652,
+      "logits/rejected": 3.4635086059570312,
+      "logps/chosen": -2.8061039447784424,
+      "logps/rejected": -6.826286315917969,
+      "loss": 0.2545,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.209155082702637,
+      "rewards/margins": 6.030273914337158,
+      "rewards/rejected": -10.239428520202637,
+      "step": 750
+    },
+    {
+      "epoch": 0.6446067898581865,
+      "eval_logits/chosen": 2.94565749168396,
+      "eval_logits/rejected": 3.170260190963745,
+      "eval_logps/chosen": -3.455902338027954,
+      "eval_logps/rejected": -8.303979873657227,
+      "eval_loss": 0.2537091076374054,
+      "eval_rewards/accuracies": 0.9368420839309692,
+      "eval_rewards/chosen": -5.1838531494140625,
+      "eval_rewards/margins": 7.272115707397461,
+      "eval_rewards/rejected": -12.455968856811523,
+      "eval_runtime": 25.8128,
+      "eval_samples_per_second": 29.172,
+      "eval_steps_per_second": 3.68,
+      "step": 750
+    },
+    {
+      "epoch": 0.6532015470562956,
+      "grad_norm": 2.978318452835083,
+      "learning_rate": 2.447643950291608e-06,
+      "logits/chosen": 3.1550345420837402,
+      "logits/rejected": 2.9587855339050293,
+      "logps/chosen": -3.403642177581787,
+      "logps/rejected": -7.627197265625,
+      "loss": 0.2261,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.10546350479126,
+      "rewards/margins": 6.335333824157715,
+      "rewards/rejected": -11.440796852111816,
+      "step": 760
+    },
+    {
+      "epoch": 0.6617963042544048,
+      "grad_norm": 4.7188801765441895,
+      "learning_rate": 2.3953108656770018e-06,
+      "logits/chosen": 3.7232565879821777,
+      "logits/rejected": 3.5991597175598145,
+      "logps/chosen": -3.6658260822296143,
+      "logps/rejected": -7.148935794830322,
+      "loss": 0.2703,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.498739242553711,
+      "rewards/margins": 5.22466516494751,
+      "rewards/rejected": -10.723405838012695,
+      "step": 770
+    },
+    {
+      "epoch": 0.6703910614525139,
+      "grad_norm": 4.847439289093018,
+      "learning_rate": 2.3430237011767166e-06,
+      "logits/chosen": 2.366446018218994,
+      "logits/rejected": 2.4268651008605957,
+      "logps/chosen": -3.6872811317443848,
+      "logps/rejected": -8.306299209594727,
+      "loss": 0.2493,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -5.530921459197998,
+      "rewards/margins": 6.928528785705566,
+      "rewards/rejected": -12.459450721740723,
+      "step": 780
+    },
+    {
+      "epoch": 0.6789858186506231,
+      "grad_norm": 3.72248911857605,
+      "learning_rate": 2.290805391669212e-06,
+      "logits/chosen": 2.7826573848724365,
+      "logits/rejected": 2.9013209342956543,
+      "logps/chosen": -3.5557899475097656,
+      "logps/rejected": -7.959009647369385,
+      "loss": 0.2461,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.33368444442749,
+      "rewards/margins": 6.604828834533691,
+      "rewards/rejected": -11.938512802124023,
+      "step": 790
+    },
+    {
+      "epoch": 0.6875805758487322,
+      "grad_norm": 4.7869343757629395,
+      "learning_rate": 2.238678841830867e-06,
+      "logits/chosen": 2.5417182445526123,
+      "logits/rejected": 2.938063621520996,
+      "logps/chosen": -3.5474331378936768,
+      "logps/rejected": -7.563382148742676,
+      "loss": 0.2483,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.3211493492126465,
+      "rewards/margins": 6.023923873901367,
+      "rewards/rejected": -11.345073699951172,
+      "step": 800
+    },
+    {
+      "epoch": 0.6875805758487322,
+      "eval_logits/chosen": 2.9645943641662598,
+      "eval_logits/rejected": 3.2736430168151855,
+      "eval_logps/chosen": -3.4982783794403076,
+      "eval_logps/rejected": -8.614095687866211,
+      "eval_loss": 0.24035032093524933,
+      "eval_rewards/accuracies": 0.9263157844543457,
+      "eval_rewards/chosen": -5.24741792678833,
+      "eval_rewards/margins": 7.673725128173828,
+      "eval_rewards/rejected": -12.921142578125,
+      "eval_runtime": 25.8098,
+      "eval_samples_per_second": 29.175,
+      "eval_steps_per_second": 3.681,
+      "step": 800
+    },
+    {
+      "epoch": 0.6961753330468414,
+      "grad_norm": 5.635983943939209,
+      "learning_rate": 2.186666916089239e-06,
+      "logits/chosen": 2.7499587535858154,
+      "logits/rejected": 2.6932997703552246,
+      "logps/chosen": -3.2679648399353027,
+      "logps/rejected": -8.038375854492188,
+      "loss": 0.2647,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.901947975158691,
+      "rewards/margins": 7.155615329742432,
+      "rewards/rejected": -12.057561874389648,
+      "step": 810
+    },
+    {
+      "epoch": 0.7047700902449506,
+      "grad_norm": 4.1279215812683105,
+      "learning_rate": 2.134792428593971e-06,
+      "logits/chosen": 3.1286826133728027,
+      "logits/rejected": 3.3689827919006348,
+      "logps/chosen": -3.5489754676818848,
+      "logps/rejected": -8.034095764160156,
+      "loss": 0.1888,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.323462963104248,
+      "rewards/margins": 6.727679252624512,
+      "rewards/rejected": -12.051143646240234,
+      "step": 820
+    },
+    {
+      "epoch": 0.7133648474430597,
+      "grad_norm": 4.562918186187744,
+      "learning_rate": 2.0830781332097446e-06,
+      "logits/chosen": 2.635140895843506,
+      "logits/rejected": 2.909487247467041,
+      "logps/chosen": -3.598254442214966,
+      "logps/rejected": -8.546786308288574,
+      "loss": 0.2141,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.39738130569458,
+      "rewards/margins": 7.422798156738281,
+      "rewards/rejected": -12.820180892944336,
+      "step": 830
+    },
+    {
+      "epoch": 0.7219596046411689,
+      "grad_norm": 5.229101657867432,
+      "learning_rate": 2.031546713535688e-06,
+      "logits/chosen": 3.0485472679138184,
+      "logits/rejected": 3.2676749229431152,
+      "logps/chosen": -3.5765304565429688,
+      "logps/rejected": -8.731932640075684,
+      "loss": 0.1953,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.364795207977295,
+      "rewards/margins": 7.7331037521362305,
+      "rewards/rejected": -13.097898483276367,
+      "step": 840
+    },
+    {
+      "epoch": 0.730554361839278,
+      "grad_norm": 3.0395517349243164,
+      "learning_rate": 1.9802207729556023e-06,
+      "logits/chosen": 3.242750644683838,
+      "logits/rejected": 3.410076141357422,
+      "logps/chosen": -3.5732295513153076,
+      "logps/rejected": -7.774885654449463,
+      "loss": 0.2395,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.359843730926514,
+      "rewards/margins": 6.302483558654785,
+      "rewards/rejected": -11.662328720092773,
+      "step": 850
+    },
+    {
+      "epoch": 0.730554361839278,
+      "eval_logits/chosen": 2.9314279556274414,
+      "eval_logits/rejected": 3.276207208633423,
+      "eval_logps/chosen": -3.8869926929473877,
+      "eval_logps/rejected": -9.050567626953125,
+      "eval_loss": 0.2363433688879013,
+      "eval_rewards/accuracies": 0.9263157844543457,
+      "eval_rewards/chosen": -5.830489158630371,
+      "eval_rewards/margins": 7.745361328125,
+      "eval_rewards/rejected": -13.575852394104004,
+      "eval_runtime": 25.8152,
+      "eval_samples_per_second": 29.169,
+      "eval_steps_per_second": 3.68,
+      "step": 850
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 1500,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 50,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.934492070852952e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-850/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cc4cd97bd7e66b9f8e8e61852db449abcbbbd1c049a5ad37ec259f9f39597e8
+size 7224

checkpoint-850/zero_to_fp32.py

@@ -0,0 +1,674 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example:
+#   python zero_to_fp32.py . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import json
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in tqdm(param_shapes.items(), desc='Gathering Sharded Weights'):
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        state_dict_split = split_torch_state_dict_into_shards(state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard = {tensor: state_dict[tensor].contiguous() for tensor in tensors}
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard, output_path)
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument("output_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+                                               args.output_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)