Training in progress, step 800, checkpoint

checkpoint-800/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-800/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": [
+    "gate_up_proj",
+    "qkv_proj",
+    "down_proj",
+    "o_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-800/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:864182eeff2d5d13882233a48c798fb85fa04ee78aa6ea4bbc63d231c1085498
+size 25200088

checkpoint-800/global_step800/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:e7d5089cdaa17f2abd703632add10c12182fe949c6b705b0b4f803e5e8f15a55
+size 18881328

checkpoint-800/global_step800/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75957ccc7677b00f26e11c60ba1943a2f3e7b40b48578d31392b773a70548120
+size 18881328

checkpoint-800/global_step800/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8652bc8d27693520a5250d873ec853ab4376443e6a8bb8080dd01d767a077d20
+size 18881328

checkpoint-800/global_step800/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3118ff73c06c720a57be5a01ba62e6641c3aaf452eb2dd6efef1955737380851
+size 18881392

checkpoint-800/global_step800/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:27e4c8215868e790b53ccae9680e217bffb967d6432a6456d4b37eb9d145c072
+size 18881392

checkpoint-800/global_step800/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9dc62363829f47349cdd821eddccab95c0b060be8f196ff303dd13ef7a0a0837
+size 18881392

checkpoint-800/global_step800/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:a94f707f21c3a0f62136b45f0da4a21cce810691c2bd7222cb4cc103ee86d591
+size 18881392

checkpoint-800/global_step800/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:184d9545798a33e07a3e698d989cdd8b0d4a14412169345cb946b15f2f300976
+size 18881392

checkpoint-800/global_step800/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1054ad8086b8bec6af92baf2cbec97156dcc30f1076cba2be8a0bea98b511b92
+size 25379244

checkpoint-800/latest

@@ -0,0 +1 @@
+global_step800

checkpoint-800/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3bae17ea03fef9b7a71d175e609746f5f3fe2a60f3ed13635f6dc665b11f354
+size 15984

checkpoint-800/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:923a360532bddf10ba1645b451f7759402dad170be3620bc17d5f6f4fb00de36
+size 15984

checkpoint-800/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d71813f450c82f8af72c2cd844617dfe5f76fb4aa51ff57fece979d10d0c8170
+size 15984

checkpoint-800/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d5298c6ce71cac9a3de67ce5987382569aedae7042216a68b8a5d6589481a48
+size 15984

checkpoint-800/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:216e2aae4ea0e03b6c9709171221e51c73a1ca974c74a698d09a6e0e90e7a102
+size 15984

checkpoint-800/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:549ec918a3957d0f1b3143c1d9ff03581199729925aaa57987967cad5ce3913e
+size 15984

checkpoint-800/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eacf66f46690cb657e6ccb8105ac57343b666f504a27c9e8486c483c383012b9
+size 15984

checkpoint-800/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb94ba11b1ba7011123e588c395190bbe8fd47f934d7816c934d6dc7d11065ba
+size 15984

checkpoint-800/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70738be5abb38e8cfb91cc830ac49dd9920dd3cf34dfffcb1efb26bab828bcb1
+size 1064

checkpoint-800/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-800/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-800/trainer_state.json

@@ -0,0 +1,1489 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.60882800608828,
+  "eval_steps": 50,
+  "global_step": 800,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0076103500761035,
+      "grad_norm": 0.058339186012744904,
+      "learning_rate": 4.999451708687114e-06,
+      "logits/chosen": 14.268467903137207,
+      "logits/rejected": 14.600369453430176,
+      "logps/chosen": -0.2669850289821625,
+      "logps/rejected": -0.3412467837333679,
+      "loss": 0.9049,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.4004775583744049,
+      "rewards/margins": 0.11139259487390518,
+      "rewards/rejected": -0.5118702054023743,
+      "step": 10
+    },
+    {
+      "epoch": 0.015220700152207,
+      "grad_norm": 0.049545690417289734,
+      "learning_rate": 4.997807075247147e-06,
+      "logits/chosen": 14.14539623260498,
+      "logits/rejected": 15.191584587097168,
+      "logps/chosen": -0.25579872727394104,
+      "logps/rejected": -0.3931494653224945,
+      "loss": 0.8989,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.38369807600975037,
+      "rewards/margins": 0.2060261219739914,
+      "rewards/rejected": -0.5897241830825806,
+      "step": 20
+    },
+    {
+      "epoch": 0.0228310502283105,
+      "grad_norm": 0.061699289828538895,
+      "learning_rate": 4.9950668210706795e-06,
+      "logits/chosen": 14.284139633178711,
+      "logits/rejected": 15.006326675415039,
+      "logps/chosen": -0.275672048330307,
+      "logps/rejected": -0.3603581488132477,
+      "loss": 0.9004,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.4135080873966217,
+      "rewards/margins": 0.12702910602092743,
+      "rewards/rejected": -0.5405372381210327,
+      "step": 30
+    },
+    {
+      "epoch": 0.030441400304414,
+      "grad_norm": 0.05706426501274109,
+      "learning_rate": 4.9912321481237616e-06,
+      "logits/chosen": 14.275796890258789,
+      "logits/rejected": 14.935521125793457,
+      "logps/chosen": -0.2802076041698456,
+      "logps/rejected": -0.38278770446777344,
+      "loss": 0.9138,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.42031145095825195,
+      "rewards/margins": 0.15387018024921417,
+      "rewards/rejected": -0.5741815567016602,
+      "step": 40
+    },
+    {
+      "epoch": 0.0380517503805175,
+      "grad_norm": 0.05318514257669449,
+      "learning_rate": 4.986304738420684e-06,
+      "logits/chosen": 14.433627128601074,
+      "logits/rejected": 15.458297729492188,
+      "logps/chosen": -0.2581387162208557,
+      "logps/rejected": -0.38208404183387756,
+      "loss": 0.914,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.38720807433128357,
+      "rewards/margins": 0.18591801822185516,
+      "rewards/rejected": -0.5731261372566223,
+      "step": 50
+    },
+    {
+      "epoch": 0.0380517503805175,
+      "eval_logits/chosen": 14.396967887878418,
+      "eval_logits/rejected": 15.221076965332031,
+      "eval_logps/chosen": -0.27519574761390686,
+      "eval_logps/rejected": -0.3709692656993866,
+      "eval_loss": 0.9084128141403198,
+      "eval_rewards/accuracies": 0.5981308221817017,
+      "eval_rewards/chosen": -0.4127936065196991,
+      "eval_rewards/margins": 0.14366032183170319,
+      "eval_rewards/rejected": -0.5564539432525635,
+      "eval_runtime": 30.773,
+      "eval_samples_per_second": 27.622,
+      "eval_steps_per_second": 3.477,
+      "step": 50
+    },
+    {
+      "epoch": 0.045662100456621,
+      "grad_norm": 0.06310460716485977,
+      "learning_rate": 4.980286753286196e-06,
+      "logits/chosen": 14.548416137695312,
+      "logits/rejected": 15.526041030883789,
+      "logps/chosen": -0.29403647780418396,
+      "logps/rejected": -0.40682005882263184,
+      "loss": 0.9082,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.44105473160743713,
+      "rewards/margins": 0.1691754311323166,
+      "rewards/rejected": -0.6102300882339478,
+      "step": 60
+    },
+    {
+      "epoch": 0.0532724505327245,
+      "grad_norm": 0.1258806735277176,
+      "learning_rate": 4.973180832407471e-06,
+      "logits/chosen": 14.390210151672363,
+      "logits/rejected": 14.817584037780762,
+      "logps/chosen": -0.25258123874664307,
+      "logps/rejected": -0.36392712593078613,
+      "loss": 0.896,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.3788718581199646,
+      "rewards/margins": 0.1670188158750534,
+      "rewards/rejected": -0.5458906888961792,
+      "step": 70
+    },
+    {
+      "epoch": 0.060882800608828,
+      "grad_norm": 0.09006265550851822,
+      "learning_rate": 4.964990092676263e-06,
+      "logits/chosen": 13.844560623168945,
+      "logits/rejected": 14.811120986938477,
+      "logps/chosen": -0.2630843222141266,
+      "logps/rejected": -0.3794577717781067,
+      "loss": 0.8977,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.3946264684200287,
+      "rewards/margins": 0.17456015944480896,
+      "rewards/rejected": -0.5691865682601929,
+      "step": 80
+    },
+    {
+      "epoch": 0.0684931506849315,
+      "grad_norm": 0.07123688608407974,
+      "learning_rate": 4.9557181268217225e-06,
+      "logits/chosen": 13.927327156066895,
+      "logits/rejected": 14.746416091918945,
+      "logps/chosen": -0.25282323360443115,
+      "logps/rejected": -0.3279832601547241,
+      "loss": 0.9092,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.37923485040664673,
+      "rewards/margins": 0.11274002492427826,
+      "rewards/rejected": -0.4919748902320862,
+      "step": 90
+    },
+    {
+      "epoch": 0.076103500761035,
+      "grad_norm": 0.08333446085453033,
+      "learning_rate": 4.9453690018345144e-06,
+      "logits/chosen": 14.406118392944336,
+      "logits/rejected": 14.770090103149414,
+      "logps/chosen": -0.28569403290748596,
+      "logps/rejected": -0.3596845269203186,
+      "loss": 0.8932,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.42854103446006775,
+      "rewards/margins": 0.11098580062389374,
+      "rewards/rejected": -0.5395268201828003,
+      "step": 100
+    },
+    {
+      "epoch": 0.076103500761035,
+      "eval_logits/chosen": 13.925265312194824,
+      "eval_logits/rejected": 14.808513641357422,
+      "eval_logps/chosen": -0.2667020559310913,
+      "eval_logps/rejected": -0.3739235997200012,
+      "eval_loss": 0.8984279036521912,
+      "eval_rewards/accuracies": 0.5981308221817017,
+      "eval_rewards/chosen": -0.40005311369895935,
+      "eval_rewards/margins": 0.16083234548568726,
+      "eval_rewards/rejected": -0.5608854293823242,
+      "eval_runtime": 30.7791,
+      "eval_samples_per_second": 27.616,
+      "eval_steps_per_second": 3.476,
+      "step": 100
+    },
+    {
+      "epoch": 0.0837138508371385,
+      "grad_norm": 0.08474570512771606,
+      "learning_rate": 4.933947257182901e-06,
+      "logits/chosen": 13.641456604003906,
+      "logits/rejected": 14.799921035766602,
+      "logps/chosen": -0.2721528708934784,
+      "logps/rejected": -0.38378894329071045,
+      "loss": 0.8995,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.40822935104370117,
+      "rewards/margins": 0.1674540936946869,
+      "rewards/rejected": -0.5756834149360657,
+      "step": 110
+    },
+    {
+      "epoch": 0.091324200913242,
+      "grad_norm": 0.1004580408334732,
+      "learning_rate": 4.921457902821578e-06,
+      "logits/chosen": 13.835454940795898,
+      "logits/rejected": 14.882522583007812,
+      "logps/chosen": -0.28507837653160095,
+      "logps/rejected": -0.39737468957901,
+      "loss": 0.8795,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.42761754989624023,
+      "rewards/margins": 0.16844449937343597,
+      "rewards/rejected": -0.5960620641708374,
+      "step": 120
+    },
+    {
+      "epoch": 0.0989345509893455,
+      "grad_norm": 0.09537151455879211,
+      "learning_rate": 4.907906416994146e-06,
+      "logits/chosen": 13.607874870300293,
+      "logits/rejected": 14.091131210327148,
+      "logps/chosen": -0.2739318013191223,
+      "logps/rejected": -0.36800479888916016,
+      "loss": 0.8912,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.4108976721763611,
+      "rewards/margins": 0.14110951125621796,
+      "rewards/rejected": -0.5520071983337402,
+      "step": 130
+    },
+    {
+      "epoch": 0.106544901065449,
+      "grad_norm": 0.10281535238027573,
+      "learning_rate": 4.893298743830168e-06,
+      "logits/chosen": 12.017224311828613,
+      "logits/rejected": 13.04835319519043,
+      "logps/chosen": -0.24072685837745667,
+      "logps/rejected": -0.36906492710113525,
+      "loss": 0.8908,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.3610902428627014,
+      "rewards/margins": 0.19250717759132385,
+      "rewards/rejected": -0.5535974502563477,
+      "step": 140
+    },
+    {
+      "epoch": 0.1141552511415525,
+      "grad_norm": 0.707987368106842,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": 12.522550582885742,
+      "logits/rejected": 13.272679328918457,
+      "logps/chosen": -0.2583540081977844,
+      "logps/rejected": -0.3796755075454712,
+      "loss": 0.8867,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.38753098249435425,
+      "rewards/margins": 0.18198221921920776,
+      "rewards/rejected": -0.569513201713562,
+      "step": 150
+    },
+    {
+      "epoch": 0.1141552511415525,
+      "eval_logits/chosen": 11.989100456237793,
+      "eval_logits/rejected": 12.92872142791748,
+      "eval_logps/chosen": -0.27158522605895996,
+      "eval_logps/rejected": -0.40521273016929626,
+      "eval_loss": 0.8765817284584045,
+      "eval_rewards/accuracies": 0.5981308221817017,
+      "eval_rewards/chosen": -0.40737783908843994,
+      "eval_rewards/margins": 0.20044119656085968,
+      "eval_rewards/rejected": -0.6078190803527832,
+      "eval_runtime": 30.7739,
+      "eval_samples_per_second": 27.621,
+      "eval_steps_per_second": 3.477,
+      "step": 150
+    },
+    {
+      "epoch": 0.121765601217656,
+      "grad_norm": 0.19342070817947388,
+      "learning_rate": 4.860940925593703e-06,
+      "logits/chosen": 11.095940589904785,
+      "logits/rejected": 12.351040840148926,
+      "logps/chosen": -0.24749942123889923,
+      "logps/rejected": -0.43422192335128784,
+      "loss": 0.8762,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.37124913930892944,
+      "rewards/margins": 0.2800838053226471,
+      "rewards/rejected": -0.6513329744338989,
+      "step": 160
+    },
+    {
+      "epoch": 0.1293759512937595,
+      "grad_norm": 0.19374576210975647,
+      "learning_rate": 4.84320497372973e-06,
+      "logits/chosen": 10.510068893432617,
+      "logits/rejected": 11.507593154907227,
+      "logps/chosen": -0.26223134994506836,
+      "logps/rejected": -0.43635931611061096,
+      "loss": 0.8581,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.39334696531295776,
+      "rewards/margins": 0.2611919641494751,
+      "rewards/rejected": -0.6545389294624329,
+      "step": 170
+    },
+    {
+      "epoch": 0.136986301369863,
+      "grad_norm": 0.20330430567264557,
+      "learning_rate": 4.824441214720629e-06,
+      "logits/chosen": 9.89570140838623,
+      "logits/rejected": 10.669364929199219,
+      "logps/chosen": -0.3143860101699829,
+      "logps/rejected": -0.46989941596984863,
+      "loss": 0.8558,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.47157901525497437,
+      "rewards/margins": 0.23327013850212097,
+      "rewards/rejected": -0.704849123954773,
+      "step": 180
+    },
+    {
+      "epoch": 0.1445966514459665,
+      "grad_norm": 0.22942212224006653,
+      "learning_rate": 4.804657878971252e-06,
+      "logits/chosen": 8.887057304382324,
+      "logits/rejected": 9.542157173156738,
+      "logps/chosen": -0.2906036972999573,
+      "logps/rejected": -0.4810206890106201,
+      "loss": 0.8554,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.4359055459499359,
+      "rewards/margins": 0.28562551736831665,
+      "rewards/rejected": -0.7215310335159302,
+      "step": 190
+    },
+    {
+      "epoch": 0.15220700152207,
+      "grad_norm": 0.29071903228759766,
+      "learning_rate": 4.783863644106502e-06,
+      "logits/chosen": 6.791537284851074,
+      "logits/rejected": 7.366445064544678,
+      "logps/chosen": -0.31382033228874207,
+      "logps/rejected": -0.5417486429214478,
+      "loss": 0.838,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.4707304835319519,
+      "rewards/margins": 0.34189245104789734,
+      "rewards/rejected": -0.8126228451728821,
+      "step": 200
+    },
+    {
+      "epoch": 0.15220700152207,
+      "eval_logits/chosen": 7.050150394439697,
+      "eval_logits/rejected": 7.516275405883789,
+      "eval_logps/chosen": -0.3289315402507782,
+      "eval_logps/rejected": -0.5481724143028259,
+      "eval_loss": 0.813983678817749,
+      "eval_rewards/accuracies": 0.6168224215507507,
+      "eval_rewards/chosen": -0.4933973252773285,
+      "eval_rewards/margins": 0.3288613557815552,
+      "eval_rewards/rejected": -0.8222586512565613,
+      "eval_runtime": 30.7734,
+      "eval_samples_per_second": 27.621,
+      "eval_steps_per_second": 3.477,
+      "step": 200
+    },
+    {
+      "epoch": 0.1598173515981735,
+      "grad_norm": 0.23101097345352173,
+      "learning_rate": 4.762067631165049e-06,
+      "logits/chosen": 5.132790565490723,
+      "logits/rejected": 5.848537445068359,
+      "logps/chosen": -0.33372369408607483,
+      "logps/rejected": -0.5993582010269165,
+      "loss": 0.8212,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.5005855560302734,
+      "rewards/margins": 0.3984517455101013,
+      "rewards/rejected": -0.8990373611450195,
+      "step": 210
+    },
+    {
+      "epoch": 0.167427701674277,
+      "grad_norm": 0.5136363506317139,
+      "learning_rate": 4.7392794005985324e-06,
+      "logits/chosen": 3.807554244995117,
+      "logits/rejected": 4.600871562957764,
+      "logps/chosen": -0.32092416286468506,
+      "logps/rejected": -0.651642918586731,
+      "loss": 0.7851,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.4813862442970276,
+      "rewards/margins": 0.4960783123970032,
+      "rewards/rejected": -0.977464497089386,
+      "step": 220
+    },
+    {
+      "epoch": 0.1750380517503805,
+      "grad_norm": 0.4106898009777069,
+      "learning_rate": 4.715508948078037e-06,
+      "logits/chosen": 2.760650396347046,
+      "logits/rejected": 2.1608071327209473,
+      "logps/chosen": -0.43665003776550293,
+      "logps/rejected": -0.8352751731872559,
+      "loss": 0.7685,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.6549750566482544,
+      "rewards/margins": 0.5979377627372742,
+      "rewards/rejected": -1.2529128789901733,
+      "step": 230
+    },
+    {
+      "epoch": 0.182648401826484,
+      "grad_norm": 0.4719419479370117,
+      "learning_rate": 4.690766700109659e-06,
+      "logits/chosen": 3.1216347217559814,
+      "logits/rejected": 2.7202537059783936,
+      "logps/chosen": -0.444007933139801,
+      "logps/rejected": -0.7697597742080688,
+      "loss": 0.7474,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.6660118699073792,
+      "rewards/margins": 0.4886276125907898,
+      "rewards/rejected": -1.154639482498169,
+      "step": 240
+    },
+    {
+      "epoch": 0.1902587519025875,
+      "grad_norm": 0.548523485660553,
+      "learning_rate": 4.665063509461098e-06,
+      "logits/chosen": 1.3678622245788574,
+      "logits/rejected": 0.46835970878601074,
+      "logps/chosen": -0.48227253556251526,
+      "logps/rejected": -0.997289776802063,
+      "loss": 0.7017,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.7234088182449341,
+      "rewards/margins": 0.7725256681442261,
+      "rewards/rejected": -1.4959346055984497,
+      "step": 250
+    },
+    {
+      "epoch": 0.1902587519025875,
+      "eval_logits/chosen": 2.1362831592559814,
+      "eval_logits/rejected": 1.1932121515274048,
+      "eval_logps/chosen": -0.500978946685791,
+      "eval_logps/rejected": -1.0073517560958862,
+      "eval_loss": 0.6914573907852173,
+      "eval_rewards/accuracies": 0.6542056202888489,
+      "eval_rewards/chosen": -0.7514683604240417,
+      "eval_rewards/margins": 0.7595593929290771,
+      "eval_rewards/rejected": -1.5110276937484741,
+      "eval_runtime": 30.7706,
+      "eval_samples_per_second": 27.624,
+      "eval_steps_per_second": 3.477,
+      "step": 250
+    },
+    {
+      "epoch": 0.197869101978691,
+      "grad_norm": 0.700670063495636,
+      "learning_rate": 4.638410650401267e-06,
+      "logits/chosen": 2.537666082382202,
+      "logits/rejected": 1.3070740699768066,
+      "logps/chosen": -0.59038907289505,
+      "logps/rejected": -1.0600087642669678,
+      "loss": 0.6908,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.8855836987495422,
+      "rewards/margins": 0.7044296264648438,
+      "rewards/rejected": -1.5900132656097412,
+      "step": 260
+    },
+    {
+      "epoch": 0.2054794520547945,
+      "grad_norm": 0.6454456448554993,
+      "learning_rate": 4.610819813755038e-06,
+      "logits/chosen": 2.312289237976074,
+      "logits/rejected": 1.6705052852630615,
+      "logps/chosen": -0.601074755191803,
+      "logps/rejected": -1.12887442111969,
+      "loss": 0.6868,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.9016121029853821,
+      "rewards/margins": 0.7916995286941528,
+      "rewards/rejected": -1.6933116912841797,
+      "step": 270
+    },
+    {
+      "epoch": 0.213089802130898,
+      "grad_norm": 0.8001136183738708,
+      "learning_rate": 4.582303101775249e-06,
+      "logits/chosen": 1.6213299036026,
+      "logits/rejected": 0.9048928022384644,
+      "logps/chosen": -0.6731385588645935,
+      "logps/rejected": -1.3181935548782349,
+      "loss": 0.632,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -1.0097079277038574,
+      "rewards/margins": 0.9675822257995605,
+      "rewards/rejected": -1.977290153503418,
+      "step": 280
+    },
+    {
+      "epoch": 0.2207001522070015,
+      "grad_norm": 0.45858490467071533,
+      "learning_rate": 4.55287302283426e-06,
+      "logits/chosen": 1.0463030338287354,
+      "logits/rejected": 0.05798797681927681,
+      "logps/chosen": -0.677167534828186,
+      "logps/rejected": -1.4764039516448975,
+      "loss": 0.6447,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -1.0157512426376343,
+      "rewards/margins": 1.1988548040390015,
+      "rewards/rejected": -2.2146058082580566,
+      "step": 290
+    },
+    {
+      "epoch": 0.228310502283105,
+      "grad_norm": 0.5778977870941162,
+      "learning_rate": 4.522542485937369e-06,
+      "logits/chosen": 2.3259291648864746,
+      "logits/rejected": 1.6117414236068726,
+      "logps/chosen": -0.7591919302940369,
+      "logps/rejected": -1.5995824337005615,
+      "loss": 0.5702,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -1.1387879848480225,
+      "rewards/margins": 1.2605856657028198,
+      "rewards/rejected": -2.3993735313415527,
+      "step": 300
+    },
+    {
+      "epoch": 0.228310502283105,
+      "eval_logits/chosen": 1.9625831842422485,
+      "eval_logits/rejected": 1.028193473815918,
+      "eval_logps/chosen": -0.7516441941261292,
+      "eval_logps/rejected": -1.771378517150879,
+      "eval_loss": 0.5786539912223816,
+      "eval_rewards/accuracies": 0.6915887594223022,
+      "eval_rewards/chosen": -1.1274662017822266,
+      "eval_rewards/margins": 1.5296014547348022,
+      "eval_rewards/rejected": -2.6570677757263184,
+      "eval_runtime": 30.7716,
+      "eval_samples_per_second": 27.623,
+      "eval_steps_per_second": 3.477,
+      "step": 300
+    },
+    {
+      "epoch": 0.2359208523592085,
+      "grad_norm": 0.5383133292198181,
+      "learning_rate": 4.491324795060491e-06,
+      "logits/chosen": 1.2824015617370605,
+      "logits/rejected": 0.7073851823806763,
+      "logps/chosen": -0.8315173387527466,
+      "logps/rejected": -1.9733762741088867,
+      "loss": 0.587,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.2472760677337646,
+      "rewards/margins": 1.7127883434295654,
+      "rewards/rejected": -2.96006441116333,
+      "step": 310
+    },
+    {
+      "epoch": 0.243531202435312,
+      "grad_norm": 3.721909284591675,
+      "learning_rate": 4.4592336433146e-06,
+      "logits/chosen": 1.993947982788086,
+      "logits/rejected": 1.192871332168579,
+      "logps/chosen": -0.9074883460998535,
+      "logps/rejected": -1.9389015436172485,
+      "loss": 0.5194,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -1.3612326383590698,
+      "rewards/margins": 1.5471194982528687,
+      "rewards/rejected": -2.9083518981933594,
+      "step": 320
+    },
+    {
+      "epoch": 0.2511415525114155,
+      "grad_norm": 0.9611485004425049,
+      "learning_rate": 4.426283106939474e-06,
+      "logits/chosen": 0.607239305973053,
+      "logits/rejected": 0.040740929543972015,
+      "logps/chosen": -0.9696615934371948,
+      "logps/rejected": -2.3865818977355957,
+      "loss": 0.4715,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -1.4544923305511475,
+      "rewards/margins": 2.1253809928894043,
+      "rewards/rejected": -3.5798733234405518,
+      "step": 330
+    },
+    {
+      "epoch": 0.258751902587519,
+      "grad_norm": 3.716665744781494,
+      "learning_rate": 4.3924876391293915e-06,
+      "logits/chosen": 1.486352801322937,
+      "logits/rejected": 0.860406756401062,
+      "logps/chosen": -0.9488881826400757,
+      "logps/rejected": -2.771193027496338,
+      "loss": 0.4584,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -1.4233323335647583,
+      "rewards/margins": 2.733457088470459,
+      "rewards/rejected": -4.156789302825928,
+      "step": 340
+    },
+    {
+      "epoch": 0.2663622526636225,
+      "grad_norm": 2.496544361114502,
+      "learning_rate": 4.357862063693486e-06,
+      "logits/chosen": 2.1065332889556885,
+      "logits/rejected": 1.4116215705871582,
+      "logps/chosen": -0.9290377497673035,
+      "logps/rejected": -2.717181444168091,
+      "loss": 0.4126,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.393556833267212,
+      "rewards/margins": 2.682215452194214,
+      "rewards/rejected": -4.075772285461426,
+      "step": 350
+    },
+    {
+      "epoch": 0.2663622526636225,
+      "eval_logits/chosen": 2.3063719272613525,
+      "eval_logits/rejected": 1.7392665147781372,
+      "eval_logps/chosen": -0.9553582072257996,
+      "eval_logps/rejected": -2.8578038215637207,
+      "eval_loss": 0.43925610184669495,
+      "eval_rewards/accuracies": 0.7196261882781982,
+      "eval_rewards/chosen": -1.433037281036377,
+      "eval_rewards/margins": 2.853668212890625,
+      "eval_rewards/rejected": -4.286705493927002,
+      "eval_runtime": 30.7732,
+      "eval_samples_per_second": 27.621,
+      "eval_steps_per_second": 3.477,
+      "step": 350
+    },
+    {
+      "epoch": 0.273972602739726,
+      "grad_norm": 1.0364434719085693,
+      "learning_rate": 4.322421568553529e-06,
+      "logits/chosen": 3.5145366191864014,
+      "logits/rejected": 2.562318801879883,
+      "logps/chosen": -0.9316509366035461,
+      "logps/rejected": -2.7451562881469727,
+      "loss": 0.4566,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.3974764347076416,
+      "rewards/margins": 2.7202582359313965,
+      "rewards/rejected": -4.117734432220459,
+      "step": 360
+    },
+    {
+      "epoch": 0.2815829528158295,
+      "grad_norm": 0.7246320843696594,
+      "learning_rate": 4.286181699082008e-06,
+      "logits/chosen": 1.6608537435531616,
+      "logits/rejected": 1.27449631690979,
+      "logps/chosen": -1.0797128677368164,
+      "logps/rejected": -3.467390537261963,
+      "loss": 0.4299,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -1.6195694208145142,
+      "rewards/margins": 3.5815162658691406,
+      "rewards/rejected": -5.201085567474365,
+      "step": 370
+    },
+    {
+      "epoch": 0.289193302891933,
+      "grad_norm": 0.942298173904419,
+      "learning_rate": 4.249158351283414e-06,
+      "logits/chosen": 2.1106579303741455,
+      "logits/rejected": 1.5492799282073975,
+      "logps/chosen": -1.2671682834625244,
+      "logps/rejected": -3.201054811477661,
+      "loss": 0.4114,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.900752305984497,
+      "rewards/margins": 2.900829792022705,
+      "rewards/rejected": -4.801582336425781,
+      "step": 380
+    },
+    {
+      "epoch": 0.2968036529680365,
+      "grad_norm": 0.4278697371482849,
+      "learning_rate": 4.211367764821722e-06,
+      "logits/chosen": 3.2620933055877686,
+      "logits/rejected": 2.7777600288391113,
+      "logps/chosen": -1.0661684274673462,
+      "logps/rejected": -3.025578022003174,
+      "loss": 0.4259,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -1.599252462387085,
+      "rewards/margins": 2.9391140937805176,
+      "rewards/rejected": -4.53836727142334,
+      "step": 390
+    },
+    {
+      "epoch": 0.30441400304414,
+      "grad_norm": 0.6019588708877563,
+      "learning_rate": 4.172826515897146e-06,
+      "logits/chosen": 3.057295560836792,
+      "logits/rejected": 2.397916078567505,
+      "logps/chosen": -1.0584070682525635,
+      "logps/rejected": -3.479670286178589,
+      "loss": 0.4167,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.5876106023788452,
+      "rewards/margins": 3.631894588470459,
+      "rewards/rejected": -5.219505310058594,
+      "step": 400
+    },
+    {
+      "epoch": 0.30441400304414,
+      "eval_logits/chosen": 3.2906479835510254,
+      "eval_logits/rejected": 2.9191884994506836,
+      "eval_logps/chosen": -1.1546303033828735,
+      "eval_logps/rejected": -3.499722957611084,
+      "eval_loss": 0.4080003499984741,
+      "eval_rewards/accuracies": 0.7102803587913513,
+      "eval_rewards/chosen": -1.731945514678955,
+      "eval_rewards/margins": 3.517639398574829,
+      "eval_rewards/rejected": -5.249584674835205,
+      "eval_runtime": 30.7819,
+      "eval_samples_per_second": 27.614,
+      "eval_steps_per_second": 3.476,
+      "step": 400
+    },
+    {
+      "epoch": 0.3120243531202435,
+      "grad_norm": 0.610105037689209,
+      "learning_rate": 4.133551509975264e-06,
+      "logits/chosen": 2.7366433143615723,
+      "logits/rejected": 2.350151538848877,
+      "logps/chosen": -1.3425816297531128,
+      "logps/rejected": -4.451743125915527,
+      "loss": 0.3834,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -2.0138726234436035,
+      "rewards/margins": 4.663742542266846,
+      "rewards/rejected": -6.677615165710449,
+      "step": 410
+    },
+    {
+      "epoch": 0.319634703196347,
+      "grad_norm": 0.9136129021644592,
+      "learning_rate": 4.093559974371725e-06,
+      "logits/chosen": 3.8271331787109375,
+      "logits/rejected": 3.666091203689575,
+      "logps/chosen": -1.3493579626083374,
+      "logps/rejected": -3.8569908142089844,
+      "loss": 0.397,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -2.0240368843078613,
+      "rewards/margins": 3.7614493370056152,
+      "rewards/rejected": -5.785486221313477,
+      "step": 420
+    },
+    {
+      "epoch": 0.3272450532724505,
+      "grad_norm": 0.6076493859291077,
+      "learning_rate": 4.052869450695776e-06,
+      "logits/chosen": 3.027143955230713,
+      "logits/rejected": 2.2761549949645996,
+      "logps/chosen": -1.2890465259552002,
+      "logps/rejected": -4.363173961639404,
+      "loss": 0.3623,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -1.9335696697235107,
+      "rewards/margins": 4.6111907958984375,
+      "rewards/rejected": -6.544760704040527,
+      "step": 430
+    },
+    {
+      "epoch": 0.334855403348554,
+      "grad_norm": 1.0694931745529175,
+      "learning_rate": 4.011497787155938e-06,
+      "logits/chosen": 3.989302158355713,
+      "logits/rejected": 3.3767571449279785,
+      "logps/chosen": -1.3799726963043213,
+      "logps/rejected": -4.611227512359619,
+      "loss": 0.3786,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -2.0699591636657715,
+      "rewards/margins": 4.846882343292236,
+      "rewards/rejected": -6.916840553283691,
+      "step": 440
+    },
+    {
+      "epoch": 0.3424657534246575,
+      "grad_norm": 2.3523929119110107,
+      "learning_rate": 3.969463130731183e-06,
+      "logits/chosen": 3.046278953552246,
+      "logits/rejected": 2.7509286403656006,
+      "logps/chosen": -1.577859878540039,
+      "logps/rejected": -4.554004669189453,
+      "loss": 0.3948,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -2.3667898178100586,
+      "rewards/margins": 4.464217185974121,
+      "rewards/rejected": -6.831006050109863,
+      "step": 450
+    },
+    {
+      "epoch": 0.3424657534246575,
+      "eval_logits/chosen": 3.555213451385498,
+      "eval_logits/rejected": 3.359722375869751,
+      "eval_logps/chosen": -1.6125141382217407,
+      "eval_logps/rejected": -4.374329566955566,
+      "eval_loss": 0.3748260736465454,
+      "eval_rewards/accuracies": 0.7943925261497498,
+      "eval_rewards/chosen": -2.418771266937256,
+      "eval_rewards/margins": 4.142723560333252,
+      "eval_rewards/rejected": -6.56149435043335,
+      "eval_runtime": 30.7795,
+      "eval_samples_per_second": 27.616,
+      "eval_steps_per_second": 3.476,
+      "step": 450
+    },
+    {
+      "epoch": 0.350076103500761,
+      "grad_norm": 1.113964557647705,
+      "learning_rate": 3.92678391921108e-06,
+      "logits/chosen": 3.61175274848938,
+      "logits/rejected": 3.547903537750244,
+      "logps/chosen": -1.7464786767959595,
+      "logps/rejected": -5.045803070068359,
+      "loss": 0.3717,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -2.619718074798584,
+      "rewards/margins": 4.948986053466797,
+      "rewards/rejected": -7.568705081939697,
+      "step": 460
+    },
+    {
+      "epoch": 0.3576864535768645,
+      "grad_norm": 1.5195355415344238,
+      "learning_rate": 3.88347887310836e-06,
+      "logits/chosen": 3.0807926654815674,
+      "logits/rejected": 3.012016773223877,
+      "logps/chosen": -2.164515733718872,
+      "logps/rejected": -5.039651393890381,
+      "loss": 0.3507,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -3.2467732429504395,
+      "rewards/margins": 4.312704086303711,
+      "rewards/rejected": -7.55947732925415,
+      "step": 470
+    },
+    {
+      "epoch": 0.365296803652968,
+      "grad_norm": 2.3880045413970947,
+      "learning_rate": 3.839566987447492e-06,
+      "logits/chosen": 2.4990105628967285,
+      "logits/rejected": 2.5192058086395264,
+      "logps/chosen": -2.5131685733795166,
+      "logps/rejected": -5.811826705932617,
+      "loss": 0.3326,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -3.7697532176971436,
+      "rewards/margins": 4.947987079620361,
+      "rewards/rejected": -8.717740058898926,
+      "step": 480
+    },
+    {
+      "epoch": 0.3729071537290715,
+      "grad_norm": 4.61068868637085,
+      "learning_rate": 3.795067523432826e-06,
+      "logits/chosen": 2.1001622676849365,
+      "logits/rejected": 2.0562539100646973,
+      "logps/chosen": -2.7572569847106934,
+      "logps/rejected": -6.228929042816162,
+      "loss": 0.3227,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.135885715484619,
+      "rewards/margins": 5.207508563995361,
+      "rewards/rejected": -9.343393325805664,
+      "step": 490
+    },
+    {
+      "epoch": 0.380517503805175,
+      "grad_norm": 8.403047561645508,
+      "learning_rate": 3.7500000000000005e-06,
+      "logits/chosen": 3.1137287616729736,
+      "logits/rejected": 2.6646764278411865,
+      "logps/chosen": -2.8061861991882324,
+      "logps/rejected": -6.236757755279541,
+      "loss": 0.3422,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.2092790603637695,
+      "rewards/margins": 5.1458563804626465,
+      "rewards/rejected": -9.355135917663574,
+      "step": 500
+    },
+    {
+      "epoch": 0.380517503805175,
+      "eval_logits/chosen": 3.3083701133728027,
+      "eval_logits/rejected": 3.13222336769104,
+      "eval_logps/chosen": -2.6677865982055664,
+      "eval_logps/rejected": -5.843282222747803,
+      "eval_loss": 0.30595287680625916,
+      "eval_rewards/accuracies": 0.8878504633903503,
+      "eval_rewards/chosen": -4.001679420471191,
+      "eval_rewards/margins": 4.763244152069092,
+      "eval_rewards/rejected": -8.764924049377441,
+      "eval_runtime": 30.7793,
+      "eval_samples_per_second": 27.616,
+      "eval_steps_per_second": 3.476,
+      "step": 500
+    },
+    {
+      "epoch": 0.3881278538812785,
+      "grad_norm": 2.3582851886749268,
+      "learning_rate": 3.7043841852542884e-06,
+      "logits/chosen": 2.7116522789001465,
+      "logits/rejected": 2.5776076316833496,
+      "logps/chosen": -2.7367191314697266,
+      "logps/rejected": -6.1324052810668945,
+      "loss": 0.2649,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -4.10507869720459,
+      "rewards/margins": 5.093530178070068,
+      "rewards/rejected": -9.1986083984375,
+      "step": 510
+    },
+    {
+      "epoch": 0.395738203957382,
+      "grad_norm": 2.8183226585388184,
+      "learning_rate": 3.658240087799655e-06,
+      "logits/chosen": 2.327544689178467,
+      "logits/rejected": 2.3745343685150146,
+      "logps/chosen": -2.6957223415374756,
+      "logps/rejected": -6.291537284851074,
+      "loss": 0.295,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.043583393096924,
+      "rewards/margins": 5.393722057342529,
+      "rewards/rejected": -9.437305450439453,
+      "step": 520
+    },
+    {
+      "epoch": 0.4033485540334855,
+      "grad_norm": 1.8313360214233398,
+      "learning_rate": 3.611587947962319e-06,
+      "logits/chosen": 2.4468109607696533,
+      "logits/rejected": 2.4551472663879395,
+      "logps/chosen": -2.7839953899383545,
+      "logps/rejected": -6.5379180908203125,
+      "loss": 0.2618,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.175992965698242,
+      "rewards/margins": 5.63088321685791,
+      "rewards/rejected": -9.806875228881836,
+      "step": 530
+    },
+    {
+      "epoch": 0.410958904109589,
+      "grad_norm": 2.2132411003112793,
+      "learning_rate": 3.564448228912682e-06,
+      "logits/chosen": 3.125279664993286,
+      "logits/rejected": 2.7795650959014893,
+      "logps/chosen": -3.349208116531372,
+      "logps/rejected": -6.923414707183838,
+      "loss": 0.2592,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.023811340332031,
+      "rewards/margins": 5.361310005187988,
+      "rewards/rejected": -10.385122299194336,
+      "step": 540
+    },
+    {
+      "epoch": 0.4185692541856925,
+      "grad_norm": 6.05848503112793,
+      "learning_rate": 3.516841607689501e-06,
+      "logits/chosen": 2.841399669647217,
+      "logits/rejected": 2.997351884841919,
+      "logps/chosen": -3.256176710128784,
+      "logps/rejected": -7.098822593688965,
+      "loss": 0.2603,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.884264945983887,
+      "rewards/margins": 5.763968467712402,
+      "rewards/rejected": -10.648235321044922,
+      "step": 550
+    },
+    {
+      "epoch": 0.4185692541856925,
+      "eval_logits/chosen": 3.109469413757324,
+      "eval_logits/rejected": 3.010756492614746,
+      "eval_logps/chosen": -3.216036558151245,
+      "eval_logps/rejected": -6.825747013092041,
+      "eval_loss": 0.27887609601020813,
+      "eval_rewards/accuracies": 0.8878504633903503,
+      "eval_rewards/chosen": -4.824055194854736,
+      "eval_rewards/margins": 5.414565563201904,
+      "eval_rewards/rejected": -10.23862075805664,
+      "eval_runtime": 30.773,
+      "eval_samples_per_second": 27.622,
+      "eval_steps_per_second": 3.477,
+      "step": 550
+    },
+    {
+      "epoch": 0.426179604261796,
+      "grad_norm": 1.8734403848648071,
+      "learning_rate": 3.4687889661302577e-06,
+      "logits/chosen": 1.8899682760238647,
+      "logits/rejected": 1.7766664028167725,
+      "logps/chosen": -3.1907763481140137,
+      "logps/rejected": -7.273028373718262,
+      "loss": 0.2669,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.7861647605896,
+      "rewards/margins": 6.123377323150635,
+      "rewards/rejected": -10.909541130065918,
+      "step": 560
+    },
+    {
+      "epoch": 0.4337899543378995,
+      "grad_norm": 3.2115261554718018,
+      "learning_rate": 3.4203113817116955e-06,
+      "logits/chosen": 2.563091278076172,
+      "logits/rejected": 2.530696392059326,
+      "logps/chosen": -3.620448589324951,
+      "logps/rejected": -7.546849250793457,
+      "loss": 0.2683,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.430673599243164,
+      "rewards/margins": 5.8896002769470215,
+      "rewards/rejected": -11.320273399353027,
+      "step": 570
+    },
+    {
+      "epoch": 0.441400304414003,
+      "grad_norm": 3.684910297393799,
+      "learning_rate": 3.3714301183045382e-06,
+      "logits/chosen": 2.873882293701172,
+      "logits/rejected": 3.193092107772827,
+      "logps/chosen": -3.386859178543091,
+      "logps/rejected": -7.514338493347168,
+      "loss": 0.2715,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -5.080288887023926,
+      "rewards/margins": 6.191219329833984,
+      "rewards/rejected": -11.271509170532227,
+      "step": 580
+    },
+    {
+      "epoch": 0.4490106544901065,
+      "grad_norm": 2.661367416381836,
+      "learning_rate": 3.3221666168464584e-06,
+      "logits/chosen": 2.5157277584075928,
+      "logits/rejected": 2.5739080905914307,
+      "logps/chosen": -3.658534526824951,
+      "logps/rejected": -7.9988884925842285,
+      "loss": 0.2454,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.487801551818848,
+      "rewards/margins": 6.510530948638916,
+      "rewards/rejected": -11.998331069946289,
+      "step": 590
+    },
+    {
+      "epoch": 0.45662100456621,
+      "grad_norm": 1.8180292844772339,
+      "learning_rate": 3.272542485937369e-06,
+      "logits/chosen": 2.6391870975494385,
+      "logits/rejected": 2.72003173828125,
+      "logps/chosen": -3.382587432861328,
+      "logps/rejected": -8.08546257019043,
+      "loss": 0.2706,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.073880672454834,
+      "rewards/margins": 7.054312229156494,
+      "rewards/rejected": -12.128194808959961,
+      "step": 600
+    },
+    {
+      "epoch": 0.45662100456621,
+      "eval_logits/chosen": 3.235776901245117,
+      "eval_logits/rejected": 3.2310192584991455,
+      "eval_logps/chosen": -3.201641082763672,
+      "eval_logps/rejected": -7.113856315612793,
+      "eval_loss": 0.26078400015830994,
+      "eval_rewards/accuracies": 0.8878504633903503,
+      "eval_rewards/chosen": -4.802461624145508,
+      "eval_rewards/margins": 5.8683247566223145,
+      "eval_rewards/rejected": -10.67078685760498,
+      "eval_runtime": 30.7777,
+      "eval_samples_per_second": 27.617,
+      "eval_steps_per_second": 3.477,
+      "step": 600
+    },
+    {
+      "epoch": 0.4642313546423135,
+      "grad_norm": 3.1237454414367676,
+      "learning_rate": 3.222579492361179e-06,
+      "logits/chosen": 3.097729444503784,
+      "logits/rejected": 2.8835232257843018,
+      "logps/chosen": -3.2986862659454346,
+      "logps/rejected": -7.824951171875,
+      "loss": 0.2455,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.948029518127441,
+      "rewards/margins": 6.789399147033691,
+      "rewards/rejected": -11.7374267578125,
+      "step": 610
+    },
+    {
+      "epoch": 0.471841704718417,
+      "grad_norm": 2.250023365020752,
+      "learning_rate": 3.1722995515381644e-06,
+      "logits/chosen": 3.303495407104492,
+      "logits/rejected": 3.124060869216919,
+      "logps/chosen": -2.9880855083465576,
+      "logps/rejected": -6.581275939941406,
+      "loss": 0.2754,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.482128143310547,
+      "rewards/margins": 5.389786243438721,
+      "rewards/rejected": -9.87191390991211,
+      "step": 620
+    },
+    {
+      "epoch": 0.4794520547945205,
+      "grad_norm": 4.364448547363281,
+      "learning_rate": 3.121724717912138e-06,
+      "logits/chosen": 2.8994319438934326,
+      "logits/rejected": 2.593780755996704,
+      "logps/chosen": -3.4450290203094482,
+      "logps/rejected": -7.1797990798950195,
+      "loss": 0.2316,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.167544364929199,
+      "rewards/margins": 5.602154731750488,
+      "rewards/rejected": -10.769698143005371,
+      "step": 630
+    },
+    {
+      "epoch": 0.487062404870624,
+      "grad_norm": 3.65561580657959,
+      "learning_rate": 3.0708771752766397e-06,
+      "logits/chosen": 3.1075518131256104,
+      "logits/rejected": 2.8703231811523438,
+      "logps/chosen": -3.26599383354187,
+      "logps/rejected": -7.536534786224365,
+      "loss": 0.2549,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.898990631103516,
+      "rewards/margins": 6.4058122634887695,
+      "rewards/rejected": -11.304803848266602,
+      "step": 640
+    },
+    {
+      "epoch": 0.4946727549467275,
+      "grad_norm": 3.1211891174316406,
+      "learning_rate": 3.019779227044398e-06,
+      "logits/chosen": 3.2364888191223145,
+      "logits/rejected": 3.3938751220703125,
+      "logps/chosen": -3.538849353790283,
+      "logps/rejected": -7.827691555023193,
+      "loss": 0.2571,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.308274269104004,
+      "rewards/margins": 6.433261871337891,
+      "rewards/rejected": -11.741537094116211,
+      "step": 650
+    },
+    {
+      "epoch": 0.4946727549467275,
+      "eval_logits/chosen": 3.091616153717041,
+      "eval_logits/rejected": 3.0459396839141846,
+      "eval_logps/chosen": -3.361125946044922,
+      "eval_logps/rejected": -7.390212535858154,
+      "eval_loss": 0.2536354660987854,
+      "eval_rewards/accuracies": 0.8971962332725525,
+      "eval_rewards/chosen": -5.041689395904541,
+      "eval_rewards/margins": 6.043630599975586,
+      "eval_rewards/rejected": -11.085319519042969,
+      "eval_runtime": 30.7751,
+      "eval_samples_per_second": 27.62,
+      "eval_steps_per_second": 3.477,
+      "step": 650
+    },
+    {
+      "epoch": 0.502283105022831,
+      "grad_norm": 4.375415802001953,
+      "learning_rate": 2.9684532864643123e-06,
+      "logits/chosen": 1.8274192810058594,
+      "logits/rejected": 1.9628839492797852,
+      "logps/chosen": -3.083608627319336,
+      "logps/rejected": -8.370513916015625,
+      "loss": 0.2166,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -4.625412940979004,
+      "rewards/margins": 7.930357456207275,
+      "rewards/rejected": -12.555770874023438,
+      "step": 660
+    },
+    {
+      "epoch": 0.5098934550989346,
+      "grad_norm": 3.6583638191223145,
+      "learning_rate": 2.9169218667902562e-06,
+      "logits/chosen": 2.5409281253814697,
+      "logits/rejected": 2.46968150138855,
+      "logps/chosen": -3.252990245819092,
+      "logps/rejected": -7.6487884521484375,
+      "loss": 0.2103,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.879485607147217,
+      "rewards/margins": 6.593697547912598,
+      "rewards/rejected": -11.473182678222656,
+      "step": 670
+    },
+    {
+      "epoch": 0.517503805175038,
+      "grad_norm": 2.009876251220703,
+      "learning_rate": 2.8652075714060296e-06,
+      "logits/chosen": 3.553900957107544,
+      "logits/rejected": 3.4104526042938232,
+      "logps/chosen": -2.9901890754699707,
+      "logps/rejected": -7.472288608551025,
+      "loss": 0.2405,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.485283374786377,
+      "rewards/margins": 6.723149299621582,
+      "rewards/rejected": -11.208433151245117,
+      "step": 680
+    },
+    {
+      "epoch": 0.5251141552511416,
+      "grad_norm": 2.9065611362457275,
+      "learning_rate": 2.813333083910761e-06,
+      "logits/chosen": 2.323111057281494,
+      "logits/rejected": 2.0086140632629395,
+      "logps/chosen": -3.430807590484619,
+      "logps/rejected": -8.105338096618652,
+      "loss": 0.2182,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.146210670471191,
+      "rewards/margins": 7.011796474456787,
+      "rewards/rejected": -12.15800666809082,
+      "step": 690
+    },
+    {
+      "epoch": 0.532724505327245,
+      "grad_norm": 4.097139358520508,
+      "learning_rate": 2.761321158169134e-06,
+      "logits/chosen": 1.9292926788330078,
+      "logits/rejected": 2.0105385780334473,
+      "logps/chosen": -3.337139129638672,
+      "logps/rejected": -7.7645721435546875,
+      "loss": 0.245,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.005709171295166,
+      "rewards/margins": 6.641148567199707,
+      "rewards/rejected": -11.646858215332031,
+      "step": 700
+    },
+    {
+      "epoch": 0.532724505327245,
+      "eval_logits/chosen": 2.946713924407959,
+      "eval_logits/rejected": 2.912729501724243,
+      "eval_logps/chosen": -3.3646414279937744,
+      "eval_logps/rejected": -7.633481979370117,
+      "eval_loss": 0.2412233203649521,
+      "eval_rewards/accuracies": 0.9065420627593994,
+      "eval_rewards/chosen": -5.046962261199951,
+      "eval_rewards/margins": 6.403261184692383,
+      "eval_rewards/rejected": -11.450223922729492,
+      "eval_runtime": 30.7742,
+      "eval_samples_per_second": 27.621,
+      "eval_steps_per_second": 3.477,
+      "step": 700
+    },
+    {
+      "epoch": 0.5403348554033486,
+      "grad_norm": 4.534071922302246,
+      "learning_rate": 2.70919460833079e-06,
+      "logits/chosen": 2.4406304359436035,
+      "logits/rejected": 2.713369607925415,
+      "logps/chosen": -3.4392433166503906,
+      "logps/rejected": -8.228872299194336,
+      "loss": 0.2245,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.158864498138428,
+      "rewards/margins": 7.184444427490234,
+      "rewards/rejected": -12.34330940246582,
+      "step": 710
+    },
+    {
+      "epoch": 0.547945205479452,
+      "grad_norm": 1.5941667556762695,
+      "learning_rate": 2.6569762988232838e-06,
+      "logits/chosen": 3.3259758949279785,
+      "logits/rejected": 3.3186354637145996,
+      "logps/chosen": -3.7198212146759033,
+      "logps/rejected": -7.6400909423828125,
+      "loss": 0.2339,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.5797319412231445,
+      "rewards/margins": 5.8804030418396,
+      "rewards/rejected": -11.460134506225586,
+      "step": 720
+    },
+    {
+      "epoch": 0.5555555555555556,
+      "grad_norm": 5.972750186920166,
+      "learning_rate": 2.604689134322999e-06,
+      "logits/chosen": 2.455244302749634,
+      "logits/rejected": 2.5398240089416504,
+      "logps/chosen": -3.892965316772461,
+      "logps/rejected": -8.59666633605957,
+      "loss": 0.2041,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.83944845199585,
+      "rewards/margins": 7.055548667907715,
+      "rewards/rejected": -12.894998550415039,
+      "step": 730
+    },
+    {
+      "epoch": 0.563165905631659,
+      "grad_norm": 3.1441280841827393,
+      "learning_rate": 2.5523560497083927e-06,
+      "logits/chosen": 2.3029608726501465,
+      "logits/rejected": 2.3662524223327637,
+      "logps/chosen": -3.682513475418091,
+      "logps/rejected": -8.340951919555664,
+      "loss": 0.2723,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.523770332336426,
+      "rewards/margins": 6.9876580238342285,
+      "rewards/rejected": -12.511428833007812,
+      "step": 740
+    },
+    {
+      "epoch": 0.5707762557077626,
+      "grad_norm": 2.30711030960083,
+      "learning_rate": 2.5e-06,
+      "logits/chosen": 1.8446356058120728,
+      "logits/rejected": 1.9468234777450562,
+      "logps/chosen": -3.808454990386963,
+      "logps/rejected": -8.047523498535156,
+      "loss": 0.1936,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.712681770324707,
+      "rewards/margins": 6.3586015701293945,
+      "rewards/rejected": -12.071284294128418,
+      "step": 750
+    },
+    {
+      "epoch": 0.5707762557077626,
+      "eval_logits/chosen": 2.7250914573669434,
+      "eval_logits/rejected": 2.7465131282806396,
+      "eval_logps/chosen": -3.5594334602355957,
+      "eval_logps/rejected": -7.899675369262695,
+      "eval_loss": 0.2329329252243042,
+      "eval_rewards/accuracies": 0.9345794320106506,
+      "eval_rewards/chosen": -5.339150905609131,
+      "eval_rewards/margins": 6.510361671447754,
+      "eval_rewards/rejected": -11.84951114654541,
+      "eval_runtime": 30.7783,
+      "eval_samples_per_second": 27.617,
+      "eval_steps_per_second": 3.476,
+      "step": 750
+    },
+    {
+      "epoch": 0.578386605783866,
+      "grad_norm": 3.0719997882843018,
+      "learning_rate": 2.447643950291608e-06,
+      "logits/chosen": 2.3534064292907715,
+      "logits/rejected": 2.401563882827759,
+      "logps/chosen": -3.610807418823242,
+      "logps/rejected": -7.985041618347168,
+      "loss": 0.1847,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.416211128234863,
+      "rewards/margins": 6.561351776123047,
+      "rewards/rejected": -11.97756290435791,
+      "step": 760
+    },
+    {
+      "epoch": 0.5859969558599696,
+      "grad_norm": 3.6076536178588867,
+      "learning_rate": 2.3953108656770018e-06,
+      "logits/chosen": 1.969351053237915,
+      "logits/rejected": 2.2207655906677246,
+      "logps/chosen": -2.9666709899902344,
+      "logps/rejected": -8.2462797164917,
+      "loss": 0.2211,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.450006484985352,
+      "rewards/margins": 7.919413089752197,
+      "rewards/rejected": -12.369420051574707,
+      "step": 770
+    },
+    {
+      "epoch": 0.593607305936073,
+      "grad_norm": 3.9795055389404297,
+      "learning_rate": 2.3430237011767166e-06,
+      "logits/chosen": 2.6336114406585693,
+      "logits/rejected": 2.438262462615967,
+      "logps/chosen": -3.4108245372772217,
+      "logps/rejected": -8.591341018676758,
+      "loss": 0.2306,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.116236209869385,
+      "rewards/margins": 7.770774841308594,
+      "rewards/rejected": -12.887011528015137,
+      "step": 780
+    },
+    {
+      "epoch": 0.6012176560121766,
+      "grad_norm": 3.288729667663574,
+      "learning_rate": 2.290805391669212e-06,
+      "logits/chosen": 1.6185804605484009,
+      "logits/rejected": 1.6822645664215088,
+      "logps/chosen": -3.2728798389434814,
+      "logps/rejected": -7.791805267333984,
+      "loss": 0.2329,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.909319877624512,
+      "rewards/margins": 6.778387546539307,
+      "rewards/rejected": -11.687707901000977,
+      "step": 790
+    },
+    {
+      "epoch": 0.60882800608828,
+      "grad_norm": 2.315206289291382,
+      "learning_rate": 2.238678841830867e-06,
+      "logits/chosen": 2.018578052520752,
+      "logits/rejected": 1.8465204238891602,
+      "logps/chosen": -3.2794997692108154,
+      "logps/rejected": -8.164708137512207,
+      "loss": 0.2082,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.919250011444092,
+      "rewards/margins": 7.327812194824219,
+      "rewards/rejected": -12.247061729431152,
+      "step": 800
+    },
+    {
+      "epoch": 0.60882800608828,
+      "eval_logits/chosen": 3.08111572265625,
+      "eval_logits/rejected": 3.0503101348876953,
+      "eval_logps/chosen": -3.4727284908294678,
+      "eval_logps/rejected": -7.895880699157715,
+      "eval_loss": 0.22325536608695984,
+      "eval_rewards/accuracies": 0.9252336621284485,
+      "eval_rewards/chosen": -5.209092617034912,
+      "eval_rewards/margins": 6.634730339050293,
+      "eval_rewards/rejected": -11.84382152557373,
+      "eval_runtime": 30.7753,
+      "eval_samples_per_second": 27.62,
+      "eval_steps_per_second": 3.477,
+      "step": 800
+    }
+  ],
+  "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": 1.862174564721623e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-800/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41f4cd2f804ea1e74c05047097e9244ba2f0dfdda24bfccd07e006f56a3fae84
+size 7224

checkpoint-800/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)