Training in progress, step 900, checkpoint

checkpoint-900/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-900/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",
+    "down_proj",
+    "o_proj",
+    "qkv_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-900/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6677d701e30f847b5b34b6960e08511c9c0ead259a397e69d8d27fa8c501d2f9
+size 25200088

checkpoint-900/global_step900/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cf3f9aadd062b0c1e4728de13de2ae3adb1b0e7ac943ae1febb7a22b936ff07
+size 18881328

checkpoint-900/global_step900/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:884a0d446002c2f481f65e18739014e5faa9b1c07434237918b062a8443957e8
+size 18881328

checkpoint-900/global_step900/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:770ca732f50f9ebbcadba66ae95e844200ec47bc0374d239752781ddb2c5c8a3
+size 18881328

checkpoint-900/global_step900/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d8ca2ac733bc5829e86edc9b8c496a878345cf8612b76d6af374f89eb335d81
+size 18881392

checkpoint-900/global_step900/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:754f37108ed87551b7cb34451997f8ced0a32040c63a98cd49ce5863e77843fc
+size 18881392

checkpoint-900/global_step900/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:292654b47f7bfe1e68da7f1ddb0026b14da73705008ca44e63d7991cc7b0178d
+size 18881392

checkpoint-900/global_step900/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:b4e3fb025d006fbfa0fa113c85fc5b625e81a5841371f301a07636cdd2071370
+size 18881392

checkpoint-900/global_step900/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:27c0179947de2df4dd290bb0e7efb4173985e623c9cc83ef56d771ccec34a4eb
+size 18881392

checkpoint-900/global_step900/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f66b9100821a8e1394cc045ff6034f78a2f32edb1789c2d26766e9d8fd587bbe
+size 25379244

checkpoint-900/latest

@@ -0,0 +1 @@
+global_step900

checkpoint-900/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f8cd848ab2a07adddced81416824c30892938e6339a9526ba20bc04895acd38
+size 15984

checkpoint-900/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06c469d4c6989043e1aa3023f3a47e9927df05c9bff1b2001fd3479658ef8b67
+size 15984

checkpoint-900/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d445ca0faed960041a4e73cd8bcbb5aa9f4402433d3a654c8e3598e2ed4d2e06
+size 15984

checkpoint-900/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37a357bdfd92ec9afc20cd0c0bac019e3b6045e94db9655241d00dbe470220c4
+size 15984

checkpoint-900/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0de3ed4f4f321bee3e9f7aea58c0e15341b95e471817619cdcbfa27cb00040c0
+size 15984

checkpoint-900/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4763576336586d5f2709a34aa03f826e9761ceb0b917373026439721114383df
+size 15984

checkpoint-900/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23f2af66459a393bc9f0c394cd21750603270c1a724424b3cbac0aa517b835eb
+size 15984

checkpoint-900/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a93fc93cde1d1828aa93a8b5819dd22405b9d37d2ae0566af6ac14eb7db3bcb4
+size 15984

checkpoint-900/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d98b2c342fd62ef0682ff0655032d74bd7370aef21fbad323e9f52688207f7a
+size 1064

checkpoint-900/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-900/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-900/trainer_state.json

@@ -0,0 +1,1671 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.7265388496468214,
+  "eval_steps": 50,
+  "global_step": 900,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008072653884964682,
+      "grad_norm": 0.04381619393825531,
+      "learning_rate": 4.999451708687114e-06,
+      "logits/chosen": 14.719314575195312,
+      "logits/rejected": 15.156938552856445,
+      "logps/chosen": -0.2856016755104065,
+      "logps/rejected": -0.31895095109939575,
+      "loss": 0.9242,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.42840251326560974,
+      "rewards/margins": 0.050023891031742096,
+      "rewards/rejected": -0.47842639684677124,
+      "step": 10
+    },
+    {
+      "epoch": 0.016145307769929364,
+      "grad_norm": 0.05155143886804581,
+      "learning_rate": 4.997807075247147e-06,
+      "logits/chosen": 14.559402465820312,
+      "logits/rejected": 15.32939338684082,
+      "logps/chosen": -0.2736968398094177,
+      "logps/rejected": -0.3458033502101898,
+      "loss": 0.9127,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.4105452597141266,
+      "rewards/margins": 0.10815979540348053,
+      "rewards/rejected": -0.5187050104141235,
+      "step": 20
+    },
+    {
+      "epoch": 0.024217961654894045,
+      "grad_norm": 0.05071854218840599,
+      "learning_rate": 4.9950668210706795e-06,
+      "logits/chosen": 14.653738021850586,
+      "logits/rejected": 15.168347358703613,
+      "logps/chosen": -0.2985997200012207,
+      "logps/rejected": -0.34624338150024414,
+      "loss": 0.9141,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": -0.44789963960647583,
+      "rewards/margins": 0.07146544009447098,
+      "rewards/rejected": -0.5193650722503662,
+      "step": 30
+    },
+    {
+      "epoch": 0.03229061553985873,
+      "grad_norm": 0.052318744361400604,
+      "learning_rate": 4.9912321481237616e-06,
+      "logits/chosen": 14.621539115905762,
+      "logits/rejected": 15.138806343078613,
+      "logps/chosen": -0.27971988916397095,
+      "logps/rejected": -0.360626757144928,
+      "loss": 0.9313,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.4195798337459564,
+      "rewards/margins": 0.12136033922433853,
+      "rewards/rejected": -0.5409401655197144,
+      "step": 40
+    },
+    {
+      "epoch": 0.04036326942482341,
+      "grad_norm": 0.06900553405284882,
+      "learning_rate": 4.986304738420684e-06,
+      "logits/chosen": 14.308789253234863,
+      "logits/rejected": 14.605737686157227,
+      "logps/chosen": -0.2685723304748535,
+      "logps/rejected": -0.323064386844635,
+      "loss": 0.9076,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.40285855531692505,
+      "rewards/margins": 0.08173803985118866,
+      "rewards/rejected": -0.4845965802669525,
+      "step": 50
+    },
+    {
+      "epoch": 0.04036326942482341,
+      "eval_logits/chosen": 14.528907775878906,
+      "eval_logits/rejected": 15.016877174377441,
+      "eval_logps/chosen": -0.2801212966442108,
+      "eval_logps/rejected": -0.34862396121025085,
+      "eval_loss": 0.9108895063400269,
+      "eval_rewards/accuracies": 0.5544554591178894,
+      "eval_rewards/chosen": -0.4201819598674774,
+      "eval_rewards/margins": 0.10275395959615707,
+      "eval_rewards/rejected": -0.5229359865188599,
+      "eval_runtime": 30.01,
+      "eval_samples_per_second": 26.691,
+      "eval_steps_per_second": 3.366,
+      "step": 50
+    },
+    {
+      "epoch": 0.04843592330978809,
+      "grad_norm": 0.32321593165397644,
+      "learning_rate": 4.980286753286196e-06,
+      "logits/chosen": 14.644981384277344,
+      "logits/rejected": 15.177103996276855,
+      "logps/chosen": -0.26382654905319214,
+      "logps/rejected": -0.33932510018348694,
+      "loss": 0.9204,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.3957397937774658,
+      "rewards/margins": 0.1132478266954422,
+      "rewards/rejected": -0.5089876055717468,
+      "step": 60
+    },
+    {
+      "epoch": 0.056508577194752774,
+      "grad_norm": 0.07268164306879044,
+      "learning_rate": 4.973180832407471e-06,
+      "logits/chosen": 14.562113761901855,
+      "logits/rejected": 15.092450141906738,
+      "logps/chosen": -0.2856511175632477,
+      "logps/rejected": -0.34295767545700073,
+      "loss": 0.915,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.42847663164138794,
+      "rewards/margins": 0.08595988899469376,
+      "rewards/rejected": -0.5144366025924683,
+      "step": 70
+    },
+    {
+      "epoch": 0.06458123107971746,
+      "grad_norm": 0.06727313250303268,
+      "learning_rate": 4.964990092676263e-06,
+      "logits/chosen": 13.979713439941406,
+      "logits/rejected": 14.924532890319824,
+      "logps/chosen": -0.27184560894966125,
+      "logps/rejected": -0.3679867386817932,
+      "loss": 0.9223,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.4077683985233307,
+      "rewards/margins": 0.14421164989471436,
+      "rewards/rejected": -0.5519800186157227,
+      "step": 80
+    },
+    {
+      "epoch": 0.07265388496468214,
+      "grad_norm": 0.06138753890991211,
+      "learning_rate": 4.9557181268217225e-06,
+      "logits/chosen": 14.366241455078125,
+      "logits/rejected": 14.924840927124023,
+      "logps/chosen": -0.2656143009662628,
+      "logps/rejected": -0.3583180606365204,
+      "loss": 0.9117,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.39842137694358826,
+      "rewards/margins": 0.13905569911003113,
+      "rewards/rejected": -0.5374771356582642,
+      "step": 90
+    },
+    {
+      "epoch": 0.08072653884964683,
+      "grad_norm": 0.14299456775188446,
+      "learning_rate": 4.9453690018345144e-06,
+      "logits/chosen": 14.401769638061523,
+      "logits/rejected": 14.532609939575195,
+      "logps/chosen": -0.2966740131378174,
+      "logps/rejected": -0.3347373604774475,
+      "loss": 0.9162,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.4450109899044037,
+      "rewards/margins": 0.05709508806467056,
+      "rewards/rejected": -0.5021060705184937,
+      "step": 100
+    },
+    {
+      "epoch": 0.08072653884964683,
+      "eval_logits/chosen": 14.096770286560059,
+      "eval_logits/rejected": 14.638699531555176,
+      "eval_logps/chosen": -0.2713560461997986,
+      "eval_logps/rejected": -0.35128629207611084,
+      "eval_loss": 0.900999128818512,
+      "eval_rewards/accuracies": 0.5643564462661743,
+      "eval_rewards/chosen": -0.4070340394973755,
+      "eval_rewards/margins": 0.11989541351795197,
+      "eval_rewards/rejected": -0.5269294381141663,
+      "eval_runtime": 29.986,
+      "eval_samples_per_second": 26.712,
+      "eval_steps_per_second": 3.368,
+      "step": 100
+    },
+    {
+      "epoch": 0.08879919273461151,
+      "grad_norm": 0.0759090781211853,
+      "learning_rate": 4.933947257182901e-06,
+      "logits/chosen": 13.392621994018555,
+      "logits/rejected": 14.395462036132812,
+      "logps/chosen": -0.22954440116882324,
+      "logps/rejected": -0.36977845430374146,
+      "loss": 0.8951,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.3443165421485901,
+      "rewards/margins": 0.21035107970237732,
+      "rewards/rejected": -0.5546677112579346,
+      "step": 110
+    },
+    {
+      "epoch": 0.09687184661957618,
+      "grad_norm": 0.155408576130867,
+      "learning_rate": 4.921457902821578e-06,
+      "logits/chosen": 13.7767915725708,
+      "logits/rejected": 14.654029846191406,
+      "logps/chosen": -0.2693817615509033,
+      "logps/rejected": -0.38339418172836304,
+      "loss": 0.9012,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.40407267212867737,
+      "rewards/margins": 0.17101867496967316,
+      "rewards/rejected": -0.5750913619995117,
+      "step": 120
+    },
+    {
+      "epoch": 0.10494450050454086,
+      "grad_norm": 0.31760165095329285,
+      "learning_rate": 4.907906416994146e-06,
+      "logits/chosen": 13.619878768920898,
+      "logits/rejected": 14.147298812866211,
+      "logps/chosen": -0.2787878215312958,
+      "logps/rejected": -0.35886240005493164,
+      "loss": 0.8911,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.41818171739578247,
+      "rewards/margins": 0.12011190503835678,
+      "rewards/rejected": -0.5382936000823975,
+      "step": 130
+    },
+    {
+      "epoch": 0.11301715438950555,
+      "grad_norm": 0.10735614597797394,
+      "learning_rate": 4.893298743830168e-06,
+      "logits/chosen": 13.27166748046875,
+      "logits/rejected": 13.826273918151855,
+      "logps/chosen": -0.25437131524086,
+      "logps/rejected": -0.3877837061882019,
+      "loss": 0.8915,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.3815569281578064,
+      "rewards/margins": 0.20011858642101288,
+      "rewards/rejected": -0.5816755890846252,
+      "step": 140
+    },
+    {
+      "epoch": 0.12108980827447023,
+      "grad_norm": 0.12984605133533478,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": 12.981040954589844,
+      "logits/rejected": 13.055997848510742,
+      "logps/chosen": -0.27253809571266174,
+      "logps/rejected": -0.32365134358406067,
+      "loss": 0.8954,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.4088071286678314,
+      "rewards/margins": 0.07666991651058197,
+      "rewards/rejected": -0.4854770302772522,
+      "step": 150
+    },
+    {
+      "epoch": 0.12108980827447023,
+      "eval_logits/chosen": 12.205692291259766,
+      "eval_logits/rejected": 12.830544471740723,
+      "eval_logps/chosen": -0.2716449201107025,
+      "eval_logps/rejected": -0.37988847494125366,
+      "eval_loss": 0.8780961036682129,
+      "eval_rewards/accuracies": 0.5841584205627441,
+      "eval_rewards/chosen": -0.40746742486953735,
+      "eval_rewards/margins": 0.16236530244350433,
+      "eval_rewards/rejected": -0.5698326826095581,
+      "eval_runtime": 30.0006,
+      "eval_samples_per_second": 26.699,
+      "eval_steps_per_second": 3.367,
+      "step": 150
+    },
+    {
+      "epoch": 0.12916246215943492,
+      "grad_norm": 0.1466989368200302,
+      "learning_rate": 4.860940925593703e-06,
+      "logits/chosen": 12.592086791992188,
+      "logits/rejected": 12.590131759643555,
+      "logps/chosen": -0.2901991307735443,
+      "logps/rejected": -0.37141314148902893,
+      "loss": 0.8814,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.4352986812591553,
+      "rewards/margins": 0.12182100117206573,
+      "rewards/rejected": -0.5571196675300598,
+      "step": 160
+    },
+    {
+      "epoch": 0.13723511604439959,
+      "grad_norm": 0.15794874727725983,
+      "learning_rate": 4.84320497372973e-06,
+      "logits/chosen": 10.324287414550781,
+      "logits/rejected": 11.248865127563477,
+      "logps/chosen": -0.24505829811096191,
+      "logps/rejected": -0.4437941610813141,
+      "loss": 0.8739,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.3675874173641205,
+      "rewards/margins": 0.29810377955436707,
+      "rewards/rejected": -0.6656912565231323,
+      "step": 170
+    },
+    {
+      "epoch": 0.14530776992936428,
+      "grad_norm": 0.34027722477912903,
+      "learning_rate": 4.824441214720629e-06,
+      "logits/chosen": 9.277830123901367,
+      "logits/rejected": 10.56584644317627,
+      "logps/chosen": -0.29055729508399963,
+      "logps/rejected": -0.4694874882698059,
+      "loss": 0.8732,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.43583592772483826,
+      "rewards/margins": 0.2683953046798706,
+      "rewards/rejected": -0.7042312026023865,
+      "step": 180
+    },
+    {
+      "epoch": 0.15338042381432895,
+      "grad_norm": 0.21653155982494354,
+      "learning_rate": 4.804657878971252e-06,
+      "logits/chosen": 6.692442417144775,
+      "logits/rejected": 8.371492385864258,
+      "logps/chosen": -0.2739722728729248,
+      "logps/rejected": -0.5168331265449524,
+      "loss": 0.8425,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.4109583795070648,
+      "rewards/margins": 0.36429136991500854,
+      "rewards/rejected": -0.7752498388290405,
+      "step": 190
+    },
+    {
+      "epoch": 0.16145307769929365,
+      "grad_norm": 0.27401649951934814,
+      "learning_rate": 4.783863644106502e-06,
+      "logits/chosen": 7.028637886047363,
+      "logits/rejected": 7.22598123550415,
+      "logps/chosen": -0.32309776544570923,
+      "logps/rejected": -0.5094671249389648,
+      "loss": 0.8327,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.48464661836624146,
+      "rewards/margins": 0.2795540988445282,
+      "rewards/rejected": -0.7642006874084473,
+      "step": 200
+    },
+    {
+      "epoch": 0.16145307769929365,
+      "eval_logits/chosen": 6.369185924530029,
+      "eval_logits/rejected": 6.641132831573486,
+      "eval_logps/chosen": -0.32840561866760254,
+      "eval_logps/rejected": -0.5301258563995361,
+      "eval_loss": 0.8103437423706055,
+      "eval_rewards/accuracies": 0.6237623691558838,
+      "eval_rewards/chosen": -0.4926084876060486,
+      "eval_rewards/margins": 0.30258041620254517,
+      "eval_rewards/rejected": -0.795188844203949,
+      "eval_runtime": 29.9886,
+      "eval_samples_per_second": 26.71,
+      "eval_steps_per_second": 3.368,
+      "step": 200
+    },
+    {
+      "epoch": 0.16952573158425832,
+      "grad_norm": 0.3183073103427887,
+      "learning_rate": 4.762067631165049e-06,
+      "logits/chosen": 5.25254487991333,
+      "logits/rejected": 5.84013032913208,
+      "logps/chosen": -0.3624248802661896,
+      "logps/rejected": -0.6147049069404602,
+      "loss": 0.7877,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.5436373949050903,
+      "rewards/margins": 0.3784201443195343,
+      "rewards/rejected": -0.9220573306083679,
+      "step": 210
+    },
+    {
+      "epoch": 0.17759838546922302,
+      "grad_norm": 0.3535729646682739,
+      "learning_rate": 4.7392794005985324e-06,
+      "logits/chosen": 4.473980903625488,
+      "logits/rejected": 3.9927191734313965,
+      "logps/chosen": -0.3647093176841736,
+      "logps/rejected": -0.6410630345344543,
+      "loss": 0.7816,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.547063946723938,
+      "rewards/margins": 0.4145306646823883,
+      "rewards/rejected": -0.9615945816040039,
+      "step": 220
+    },
+    {
+      "epoch": 0.1856710393541877,
+      "grad_norm": 0.4819677174091339,
+      "learning_rate": 4.715508948078037e-06,
+      "logits/chosen": 2.7333035469055176,
+      "logits/rejected": 2.521853446960449,
+      "logps/chosen": -0.40259629487991333,
+      "logps/rejected": -0.7537732720375061,
+      "loss": 0.7306,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.6038944721221924,
+      "rewards/margins": 0.5267654657363892,
+      "rewards/rejected": -1.1306599378585815,
+      "step": 230
+    },
+    {
+      "epoch": 0.19374369323915236,
+      "grad_norm": 0.4125296175479889,
+      "learning_rate": 4.690766700109659e-06,
+      "logits/chosen": 2.212467908859253,
+      "logits/rejected": 1.1434030532836914,
+      "logps/chosen": -0.4652811884880066,
+      "logps/rejected": -0.8928227424621582,
+      "loss": 0.7214,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.6979218125343323,
+      "rewards/margins": 0.6413123607635498,
+      "rewards/rejected": -1.3392341136932373,
+      "step": 240
+    },
+    {
+      "epoch": 0.20181634712411706,
+      "grad_norm": 0.4265546202659607,
+      "learning_rate": 4.665063509461098e-06,
+      "logits/chosen": 0.4756811559200287,
+      "logits/rejected": 0.07218921184539795,
+      "logps/chosen": -0.4880926012992859,
+      "logps/rejected": -1.0095646381378174,
+      "loss": 0.6811,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.7321388721466064,
+      "rewards/margins": 0.7822080850601196,
+      "rewards/rejected": -1.5143468379974365,
+      "step": 250
+    },
+    {
+      "epoch": 0.20181634712411706,
+      "eval_logits/chosen": 1.6732138395309448,
+      "eval_logits/rejected": 0.5167235732078552,
+      "eval_logps/chosen": -0.5383209586143494,
+      "eval_logps/rejected": -1.0026048421859741,
+      "eval_loss": 0.6842760443687439,
+      "eval_rewards/accuracies": 0.6336633563041687,
+      "eval_rewards/chosen": -0.8074814677238464,
+      "eval_rewards/margins": 0.6964258551597595,
+      "eval_rewards/rejected": -1.5039072036743164,
+      "eval_runtime": 29.9884,
+      "eval_samples_per_second": 26.71,
+      "eval_steps_per_second": 3.368,
+      "step": 250
+    },
+    {
+      "epoch": 0.20988900100908173,
+      "grad_norm": 0.5196985006332397,
+      "learning_rate": 4.638410650401267e-06,
+      "logits/chosen": 1.7947940826416016,
+      "logits/rejected": 0.9839111566543579,
+      "logps/chosen": -0.6005308032035828,
+      "logps/rejected": -0.9484688639640808,
+      "loss": 0.7141,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.9007962942123413,
+      "rewards/margins": 0.5219069719314575,
+      "rewards/rejected": -1.4227031469345093,
+      "step": 260
+    },
+    {
+      "epoch": 0.21796165489404642,
+      "grad_norm": 1.302403450012207,
+      "learning_rate": 4.610819813755038e-06,
+      "logits/chosen": 2.2894890308380127,
+      "logits/rejected": 1.2887728214263916,
+      "logps/chosen": -0.5904151797294617,
+      "logps/rejected": -1.1889005899429321,
+      "loss": 0.6647,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.8856227993965149,
+      "rewards/margins": 0.8977279663085938,
+      "rewards/rejected": -1.783350944519043,
+      "step": 270
+    },
+    {
+      "epoch": 0.2260343087790111,
+      "grad_norm": 0.7729688286781311,
+      "learning_rate": 4.582303101775249e-06,
+      "logits/chosen": 0.4874440133571625,
+      "logits/rejected": -0.3855375349521637,
+      "logps/chosen": -0.624158501625061,
+      "logps/rejected": -1.4413455724716187,
+      "loss": 0.5629,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.9362378120422363,
+      "rewards/margins": 1.2257804870605469,
+      "rewards/rejected": -2.162018299102783,
+      "step": 280
+    },
+    {
+      "epoch": 0.2341069626639758,
+      "grad_norm": 0.41621893644332886,
+      "learning_rate": 4.55287302283426e-06,
+      "logits/chosen": 1.3461151123046875,
+      "logits/rejected": 0.733107328414917,
+      "logps/chosen": -0.7516278624534607,
+      "logps/rejected": -1.6215450763702393,
+      "loss": 0.5544,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -1.1274420022964478,
+      "rewards/margins": 1.3048756122589111,
+      "rewards/rejected": -2.4323174953460693,
+      "step": 290
+    },
+    {
+      "epoch": 0.24217961654894046,
+      "grad_norm": 0.45633476972579956,
+      "learning_rate": 4.522542485937369e-06,
+      "logits/chosen": 0.6991375684738159,
+      "logits/rejected": 0.1667344868183136,
+      "logps/chosen": -0.7924041152000427,
+      "logps/rejected": -2.521883249282837,
+      "loss": 0.4968,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -1.1886063814163208,
+      "rewards/margins": 2.5942187309265137,
+      "rewards/rejected": -3.782824754714966,
+      "step": 300
+    },
+    {
+      "epoch": 0.24217961654894046,
+      "eval_logits/chosen": 1.107033610343933,
+      "eval_logits/rejected": 0.10493909567594528,
+      "eval_logps/chosen": -0.8693537712097168,
+      "eval_logps/rejected": -2.1045310497283936,
+      "eval_loss": 0.4899609684944153,
+      "eval_rewards/accuracies": 0.6633663177490234,
+      "eval_rewards/chosen": -1.3040307760238647,
+      "eval_rewards/margins": 1.852765679359436,
+      "eval_rewards/rejected": -3.15679669380188,
+      "eval_runtime": 30.0102,
+      "eval_samples_per_second": 26.691,
+      "eval_steps_per_second": 3.366,
+      "step": 300
+    },
+    {
+      "epoch": 0.25025227043390513,
+      "grad_norm": 2.2053284645080566,
+      "learning_rate": 4.491324795060491e-06,
+      "logits/chosen": 0.6556342244148254,
+      "logits/rejected": -0.14924369752407074,
+      "logps/chosen": -0.8949082493782043,
+      "logps/rejected": -2.1485395431518555,
+      "loss": 0.5372,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -1.342362403869629,
+      "rewards/margins": 1.8804467916488647,
+      "rewards/rejected": -3.222809314727783,
+      "step": 310
+    },
+    {
+      "epoch": 0.25832492431886983,
+      "grad_norm": 0.8357079029083252,
+      "learning_rate": 4.4592336433146e-06,
+      "logits/chosen": 1.4236268997192383,
+      "logits/rejected": 0.44775086641311646,
+      "logps/chosen": -0.8566747903823853,
+      "logps/rejected": -2.4265379905700684,
+      "loss": 0.4974,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -1.285012125968933,
+      "rewards/margins": 2.35479474067688,
+      "rewards/rejected": -3.6398072242736816,
+      "step": 320
+    },
+    {
+      "epoch": 0.26639757820383453,
+      "grad_norm": 1.5514745712280273,
+      "learning_rate": 4.426283106939474e-06,
+      "logits/chosen": 1.0238934755325317,
+      "logits/rejected": 0.31885427236557007,
+      "logps/chosen": -0.9286335110664368,
+      "logps/rejected": -2.957723379135132,
+      "loss": 0.4777,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.3929500579833984,
+      "rewards/margins": 3.043635129928589,
+      "rewards/rejected": -4.436585426330566,
+      "step": 330
+    },
+    {
+      "epoch": 0.27447023208879917,
+      "grad_norm": 0.7523798942565918,
+      "learning_rate": 4.3924876391293915e-06,
+      "logits/chosen": 1.0386161804199219,
+      "logits/rejected": 0.1279783844947815,
+      "logps/chosen": -1.0053694248199463,
+      "logps/rejected": -2.8961727619171143,
+      "loss": 0.4718,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -1.5080541372299194,
+      "rewards/margins": 2.8362045288085938,
+      "rewards/rejected": -4.344258785247803,
+      "step": 340
+    },
+    {
+      "epoch": 0.28254288597376387,
+      "grad_norm": 0.6102933287620544,
+      "learning_rate": 4.357862063693486e-06,
+      "logits/chosen": 0.5982325077056885,
+      "logits/rejected": 0.07386422157287598,
+      "logps/chosen": -1.0740950107574463,
+      "logps/rejected": -2.4773449897766113,
+      "loss": 0.4909,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.6111425161361694,
+      "rewards/margins": 2.104874849319458,
+      "rewards/rejected": -3.716017246246338,
+      "step": 350
+    },
+    {
+      "epoch": 0.28254288597376387,
+      "eval_logits/chosen": 1.3659090995788574,
+      "eval_logits/rejected": 0.5649093985557556,
+      "eval_logps/chosen": -1.113811731338501,
+      "eval_logps/rejected": -2.65985107421875,
+      "eval_loss": 0.44899094104766846,
+      "eval_rewards/accuracies": 0.6633663177490234,
+      "eval_rewards/chosen": -1.6707175970077515,
+      "eval_rewards/margins": 2.319058895111084,
+      "eval_rewards/rejected": -3.989776849746704,
+      "eval_runtime": 29.9887,
+      "eval_samples_per_second": 26.71,
+      "eval_steps_per_second": 3.368,
+      "step": 350
+    },
+    {
+      "epoch": 0.29061553985872857,
+      "grad_norm": 0.4950815737247467,
+      "learning_rate": 4.322421568553529e-06,
+      "logits/chosen": 1.4646549224853516,
+      "logits/rejected": 1.0656757354736328,
+      "logps/chosen": -1.0681949853897095,
+      "logps/rejected": -2.9191997051239014,
+      "loss": 0.4561,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -1.6022924184799194,
+      "rewards/margins": 2.7765071392059326,
+      "rewards/rejected": -4.3787994384765625,
+      "step": 360
+    },
+    {
+      "epoch": 0.29868819374369326,
+      "grad_norm": 1.830091118812561,
+      "learning_rate": 4.286181699082008e-06,
+      "logits/chosen": 2.0835390090942383,
+      "logits/rejected": 1.3285930156707764,
+      "logps/chosen": -1.1288923025131226,
+      "logps/rejected": -3.2559380531311035,
+      "loss": 0.4305,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -1.693338394165039,
+      "rewards/margins": 3.190568208694458,
+      "rewards/rejected": -4.883906364440918,
+      "step": 370
+    },
+    {
+      "epoch": 0.3067608476286579,
+      "grad_norm": 2.1292569637298584,
+      "learning_rate": 4.249158351283414e-06,
+      "logits/chosen": 1.5609261989593506,
+      "logits/rejected": 1.0038378238677979,
+      "logps/chosen": -1.2937371730804443,
+      "logps/rejected": -3.3292288780212402,
+      "loss": 0.4358,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -1.9406057596206665,
+      "rewards/margins": 3.0532374382019043,
+      "rewards/rejected": -4.993843078613281,
+      "step": 380
+    },
+    {
+      "epoch": 0.3148335015136226,
+      "grad_norm": 2.124483346939087,
+      "learning_rate": 4.211367764821722e-06,
+      "logits/chosen": 1.9905683994293213,
+      "logits/rejected": 1.498375415802002,
+      "logps/chosen": -1.4837720394134521,
+      "logps/rejected": -3.7814183235168457,
+      "loss": 0.4565,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -2.225658416748047,
+      "rewards/margins": 3.44646954536438,
+      "rewards/rejected": -5.672127723693848,
+      "step": 390
+    },
+    {
+      "epoch": 0.3229061553985873,
+      "grad_norm": 1.8990832567214966,
+      "learning_rate": 4.172826515897146e-06,
+      "logits/chosen": 2.05663800239563,
+      "logits/rejected": 1.7521194219589233,
+      "logps/chosen": -1.662553071975708,
+      "logps/rejected": -4.254827976226807,
+      "loss": 0.408,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -2.4938297271728516,
+      "rewards/margins": 3.8884124755859375,
+      "rewards/rejected": -6.382241725921631,
+      "step": 400
+    },
+    {
+      "epoch": 0.3229061553985873,
+      "eval_logits/chosen": 1.7070311307907104,
+      "eval_logits/rejected": 1.3909664154052734,
+      "eval_logps/chosen": -2.0459556579589844,
+      "eval_logps/rejected": -4.069729804992676,
+      "eval_loss": 0.38578492403030396,
+      "eval_rewards/accuracies": 0.8613861203193665,
+      "eval_rewards/chosen": -3.0689334869384766,
+      "eval_rewards/margins": 3.035661458969116,
+      "eval_rewards/rejected": -6.104594707489014,
+      "eval_runtime": 30.015,
+      "eval_samples_per_second": 26.687,
+      "eval_steps_per_second": 3.365,
+      "step": 400
+    },
+    {
+      "epoch": 0.33097880928355194,
+      "grad_norm": 2.627016067504883,
+      "learning_rate": 4.133551509975264e-06,
+      "logits/chosen": 1.8818864822387695,
+      "logits/rejected": 1.6801849603652954,
+      "logps/chosen": -2.145986318588257,
+      "logps/rejected": -4.5329155921936035,
+      "loss": 0.3894,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -3.218979597091675,
+      "rewards/margins": 3.5803935527801514,
+      "rewards/rejected": -6.799372673034668,
+      "step": 410
+    },
+    {
+      "epoch": 0.33905146316851664,
+      "grad_norm": 2.1047608852386475,
+      "learning_rate": 4.093559974371725e-06,
+      "logits/chosen": 2.082624912261963,
+      "logits/rejected": 1.7804310321807861,
+      "logps/chosen": -2.498039722442627,
+      "logps/rejected": -4.982306957244873,
+      "loss": 0.3853,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -3.7470593452453613,
+      "rewards/margins": 3.7264015674591064,
+      "rewards/rejected": -7.4734601974487305,
+      "step": 420
+    },
+    {
+      "epoch": 0.34712411705348134,
+      "grad_norm": 1.7552043199539185,
+      "learning_rate": 4.052869450695776e-06,
+      "logits/chosen": 2.693586826324463,
+      "logits/rejected": 2.2467360496520996,
+      "logps/chosen": -2.717463731765747,
+      "logps/rejected": -5.170213222503662,
+      "loss": 0.3669,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -4.07619571685791,
+      "rewards/margins": 3.6791248321533203,
+      "rewards/rejected": -7.755320072174072,
+      "step": 430
+    },
+    {
+      "epoch": 0.35519677093844604,
+      "grad_norm": 1.5780164003372192,
+      "learning_rate": 4.011497787155938e-06,
+      "logits/chosen": 1.6175695657730103,
+      "logits/rejected": 1.4590227603912354,
+      "logps/chosen": -2.6292812824249268,
+      "logps/rejected": -5.3910956382751465,
+      "loss": 0.3333,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -3.9439215660095215,
+      "rewards/margins": 4.142721176147461,
+      "rewards/rejected": -8.086642265319824,
+      "step": 440
+    },
+    {
+      "epoch": 0.3632694248234107,
+      "grad_norm": 4.690535545349121,
+      "learning_rate": 3.969463130731183e-06,
+      "logits/chosen": 1.8307468891143799,
+      "logits/rejected": 1.500794768333435,
+      "logps/chosen": -2.872072696685791,
+      "logps/rejected": -5.305605411529541,
+      "loss": 0.356,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.308108329772949,
+      "rewards/margins": 3.6502983570098877,
+      "rewards/rejected": -7.958407402038574,
+      "step": 450
+    },
+    {
+      "epoch": 0.3632694248234107,
+      "eval_logits/chosen": 1.702280044555664,
+      "eval_logits/rejected": 1.3865309953689575,
+      "eval_logps/chosen": -2.8212814331054688,
+      "eval_logps/rejected": -5.271735668182373,
+      "eval_loss": 0.33659350872039795,
+      "eval_rewards/accuracies": 0.9009901285171509,
+      "eval_rewards/chosen": -4.231922149658203,
+      "eval_rewards/margins": 3.6756813526153564,
+      "eval_rewards/rejected": -7.9076032638549805,
+      "eval_runtime": 30.0131,
+      "eval_samples_per_second": 26.688,
+      "eval_steps_per_second": 3.365,
+      "step": 450
+    },
+    {
+      "epoch": 0.3713420787083754,
+      "grad_norm": 2.641991138458252,
+      "learning_rate": 3.92678391921108e-06,
+      "logits/chosen": 2.2038960456848145,
+      "logits/rejected": 1.8302087783813477,
+      "logps/chosen": -3.0162768363952637,
+      "logps/rejected": -5.90580940246582,
+      "loss": 0.3141,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -4.524415493011475,
+      "rewards/margins": 4.334298610687256,
+      "rewards/rejected": -8.85871410369873,
+      "step": 460
+    },
+    {
+      "epoch": 0.3794147325933401,
+      "grad_norm": 1.9930158853530884,
+      "learning_rate": 3.88347887310836e-06,
+      "logits/chosen": 1.655861258506775,
+      "logits/rejected": 1.562140703201294,
+      "logps/chosen": -2.6926944255828857,
+      "logps/rejected": -5.58877420425415,
+      "loss": 0.3298,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.039041519165039,
+      "rewards/margins": 4.344119548797607,
+      "rewards/rejected": -8.383161544799805,
+      "step": 470
+    },
+    {
+      "epoch": 0.3874873864783047,
+      "grad_norm": 2.796666383743286,
+      "learning_rate": 3.839566987447492e-06,
+      "logits/chosen": 1.465962290763855,
+      "logits/rejected": 1.0425068140029907,
+      "logps/chosen": -2.9049878120422363,
+      "logps/rejected": -5.620814323425293,
+      "loss": 0.3071,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -4.357481956481934,
+      "rewards/margins": 4.073739051818848,
+      "rewards/rejected": -8.431221961975098,
+      "step": 480
+    },
+    {
+      "epoch": 0.3955600403632694,
+      "grad_norm": 3.4124133586883545,
+      "learning_rate": 3.795067523432826e-06,
+      "logits/chosen": 1.545910120010376,
+      "logits/rejected": 1.2031619548797607,
+      "logps/chosen": -3.081360340118408,
+      "logps/rejected": -6.2000017166137695,
+      "loss": 0.2638,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.622040748596191,
+      "rewards/margins": 4.677962779998779,
+      "rewards/rejected": -9.300004959106445,
+      "step": 490
+    },
+    {
+      "epoch": 0.4036326942482341,
+      "grad_norm": 2.4986650943756104,
+      "learning_rate": 3.7500000000000005e-06,
+      "logits/chosen": 2.3995444774627686,
+      "logits/rejected": 2.111048698425293,
+      "logps/chosen": -3.1852879524230957,
+      "logps/rejected": -5.824184417724609,
+      "loss": 0.2592,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.777932167053223,
+      "rewards/margins": 3.9583442211151123,
+      "rewards/rejected": -8.736276626586914,
+      "step": 500
+    },
+    {
+      "epoch": 0.4036326942482341,
+      "eval_logits/chosen": 2.3460052013397217,
+      "eval_logits/rejected": 1.8444068431854248,
+      "eval_logps/chosen": -3.00115704536438,
+      "eval_logps/rejected": -5.741134166717529,
+      "eval_loss": 0.3161226809024811,
+      "eval_rewards/accuracies": 0.9009901285171509,
+      "eval_rewards/chosen": -4.501734733581543,
+      "eval_rewards/margins": 4.1099653244018555,
+      "eval_rewards/rejected": -8.611700057983398,
+      "eval_runtime": 30.0035,
+      "eval_samples_per_second": 26.697,
+      "eval_steps_per_second": 3.366,
+      "step": 500
+    },
+    {
+      "epoch": 0.4117053481331988,
+      "grad_norm": 2.1488800048828125,
+      "learning_rate": 3.7043841852542884e-06,
+      "logits/chosen": 2.289475440979004,
+      "logits/rejected": 2.1274220943450928,
+      "logps/chosen": -3.0586507320404053,
+      "logps/rejected": -6.597572326660156,
+      "loss": 0.2772,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.587975978851318,
+      "rewards/margins": 5.308382034301758,
+      "rewards/rejected": -9.89635944366455,
+      "step": 510
+    },
+    {
+      "epoch": 0.41977800201816345,
+      "grad_norm": 4.558802604675293,
+      "learning_rate": 3.658240087799655e-06,
+      "logits/chosen": 2.7750000953674316,
+      "logits/rejected": 2.718843936920166,
+      "logps/chosen": -3.142796039581299,
+      "logps/rejected": -6.112117767333984,
+      "loss": 0.2724,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.714193820953369,
+      "rewards/margins": 4.453982830047607,
+      "rewards/rejected": -9.168176651000977,
+      "step": 520
+    },
+    {
+      "epoch": 0.42785065590312815,
+      "grad_norm": 4.052336692810059,
+      "learning_rate": 3.611587947962319e-06,
+      "logits/chosen": 2.6630032062530518,
+      "logits/rejected": 2.3567159175872803,
+      "logps/chosen": -2.735839605331421,
+      "logps/rejected": -5.979222774505615,
+      "loss": 0.2616,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.103759288787842,
+      "rewards/margins": 4.865074634552002,
+      "rewards/rejected": -8.968833923339844,
+      "step": 530
+    },
+    {
+      "epoch": 0.43592330978809285,
+      "grad_norm": 4.516399383544922,
+      "learning_rate": 3.564448228912682e-06,
+      "logits/chosen": 2.0932493209838867,
+      "logits/rejected": 2.0154340267181396,
+      "logps/chosen": -3.2643024921417236,
+      "logps/rejected": -6.601607322692871,
+      "loss": 0.2954,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -4.896453857421875,
+      "rewards/margins": 5.00595760345459,
+      "rewards/rejected": -9.902410507202148,
+      "step": 540
+    },
+    {
+      "epoch": 0.4439959636730575,
+      "grad_norm": 5.320217132568359,
+      "learning_rate": 3.516841607689501e-06,
+      "logits/chosen": 2.524789333343506,
+      "logits/rejected": 2.2494006156921387,
+      "logps/chosen": -3.190767526626587,
+      "logps/rejected": -6.47235107421875,
+      "loss": 0.26,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.786150932312012,
+      "rewards/margins": 4.922375679016113,
+      "rewards/rejected": -9.708526611328125,
+      "step": 550
+    },
+    {
+      "epoch": 0.4439959636730575,
+      "eval_logits/chosen": 2.328671932220459,
+      "eval_logits/rejected": 1.9636019468307495,
+      "eval_logps/chosen": -3.0293216705322266,
+      "eval_logps/rejected": -6.1273980140686035,
+      "eval_loss": 0.28268176317214966,
+      "eval_rewards/accuracies": 0.9405940771102905,
+      "eval_rewards/chosen": -4.54398250579834,
+      "eval_rewards/margins": 4.64711332321167,
+      "eval_rewards/rejected": -9.191096305847168,
+      "eval_runtime": 30.0135,
+      "eval_samples_per_second": 26.688,
+      "eval_steps_per_second": 3.365,
+      "step": 550
+    },
+    {
+      "epoch": 0.4520686175580222,
+      "grad_norm": 2.362356424331665,
+      "learning_rate": 3.4687889661302577e-06,
+      "logits/chosen": 2.4379239082336426,
+      "logits/rejected": 2.2238965034484863,
+      "logps/chosen": -3.121654510498047,
+      "logps/rejected": -6.944204807281494,
+      "loss": 0.2752,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.68248176574707,
+      "rewards/margins": 5.733824729919434,
+      "rewards/rejected": -10.41630744934082,
+      "step": 560
+    },
+    {
+      "epoch": 0.4601412714429869,
+      "grad_norm": 6.080317497253418,
+      "learning_rate": 3.4203113817116955e-06,
+      "logits/chosen": 1.9741709232330322,
+      "logits/rejected": 1.7863889932632446,
+      "logps/chosen": -3.5348823070526123,
+      "logps/rejected": -7.0638427734375,
+      "loss": 0.2666,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.302323818206787,
+      "rewards/margins": 5.293439865112305,
+      "rewards/rejected": -10.595763206481934,
+      "step": 570
+    },
+    {
+      "epoch": 0.4682139253279516,
+      "grad_norm": 2.5477423667907715,
+      "learning_rate": 3.3714301183045382e-06,
+      "logits/chosen": 2.9160943031311035,
+      "logits/rejected": 2.747884750366211,
+      "logps/chosen": -3.287876844406128,
+      "logps/rejected": -6.996206760406494,
+      "loss": 0.2362,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -4.9318156242370605,
+      "rewards/margins": 5.562494277954102,
+      "rewards/rejected": -10.494308471679688,
+      "step": 580
+    },
+    {
+      "epoch": 0.47628657921291623,
+      "grad_norm": 2.0826616287231445,
+      "learning_rate": 3.3221666168464584e-06,
+      "logits/chosen": 2.3059380054473877,
+      "logits/rejected": 2.1479601860046387,
+      "logps/chosen": -3.25227689743042,
+      "logps/rejected": -6.894504547119141,
+      "loss": 0.2774,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.878415584564209,
+      "rewards/margins": 5.46334171295166,
+      "rewards/rejected": -10.341755867004395,
+      "step": 590
+    },
+    {
+      "epoch": 0.4843592330978809,
+      "grad_norm": 2.689568519592285,
+      "learning_rate": 3.272542485937369e-06,
+      "logits/chosen": 2.776691436767578,
+      "logits/rejected": 2.309044361114502,
+      "logps/chosen": -3.4026119709014893,
+      "logps/rejected": -6.8883819580078125,
+      "loss": 0.3082,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.103917121887207,
+      "rewards/margins": 5.228655815124512,
+      "rewards/rejected": -10.332572937011719,
+      "step": 600
+    },
+    {
+      "epoch": 0.4843592330978809,
+      "eval_logits/chosen": 2.1815295219421387,
+      "eval_logits/rejected": 1.8988540172576904,
+      "eval_logps/chosen": -3.1169536113739014,
+      "eval_logps/rejected": -6.34333610534668,
+      "eval_loss": 0.2698034942150116,
+      "eval_rewards/accuracies": 0.9504950642585754,
+      "eval_rewards/chosen": -4.6754302978515625,
+      "eval_rewards/margins": 4.839573860168457,
+      "eval_rewards/rejected": -9.51500415802002,
+      "eval_runtime": 30.0083,
+      "eval_samples_per_second": 26.693,
+      "eval_steps_per_second": 3.366,
+      "step": 600
+    },
+    {
+      "epoch": 0.4924318869828456,
+      "grad_norm": 3.95121169090271,
+      "learning_rate": 3.222579492361179e-06,
+      "logits/chosen": 2.4650866985321045,
+      "logits/rejected": 2.0424320697784424,
+      "logps/chosen": -3.28851580619812,
+      "logps/rejected": -6.309131622314453,
+      "loss": 0.267,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -4.932773590087891,
+      "rewards/margins": 4.530923366546631,
+      "rewards/rejected": -9.46369743347168,
+      "step": 610
+    },
+    {
+      "epoch": 0.5005045408678103,
+      "grad_norm": 3.0987000465393066,
+      "learning_rate": 3.1722995515381644e-06,
+      "logits/chosen": 2.4938290119171143,
+      "logits/rejected": 2.324687957763672,
+      "logps/chosen": -3.5282440185546875,
+      "logps/rejected": -6.8709845542907715,
+      "loss": 0.2849,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.292366027832031,
+      "rewards/margins": 5.014110565185547,
+      "rewards/rejected": -10.306477546691895,
+      "step": 620
+    },
+    {
+      "epoch": 0.508577194752775,
+      "grad_norm": 2.19297456741333,
+      "learning_rate": 3.121724717912138e-06,
+      "logits/chosen": 3.0522961616516113,
+      "logits/rejected": 2.7846601009368896,
+      "logps/chosen": -3.524689197540283,
+      "logps/rejected": -6.6990180015563965,
+      "loss": 0.2448,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.2870330810546875,
+      "rewards/margins": 4.76149320602417,
+      "rewards/rejected": -10.048526763916016,
+      "step": 630
+    },
+    {
+      "epoch": 0.5166498486377397,
+      "grad_norm": 2.607801914215088,
+      "learning_rate": 3.0708771752766397e-06,
+      "logits/chosen": 2.4197397232055664,
+      "logits/rejected": 2.1711297035217285,
+      "logps/chosen": -3.635500431060791,
+      "logps/rejected": -7.1246538162231445,
+      "loss": 0.2505,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.453250885009766,
+      "rewards/margins": 5.233729362487793,
+      "rewards/rejected": -10.686980247497559,
+      "step": 640
+    },
+    {
+      "epoch": 0.5247225025227044,
+      "grad_norm": 2.6676313877105713,
+      "learning_rate": 3.019779227044398e-06,
+      "logits/chosen": 3.1394009590148926,
+      "logits/rejected": 3.1322834491729736,
+      "logps/chosen": -3.8464252948760986,
+      "logps/rejected": -6.729602813720703,
+      "loss": 0.2889,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.7696380615234375,
+      "rewards/margins": 4.324765682220459,
+      "rewards/rejected": -10.094404220581055,
+      "step": 650
+    },
+    {
+      "epoch": 0.5247225025227044,
+      "eval_logits/chosen": 2.311185359954834,
+      "eval_logits/rejected": 2.0363073348999023,
+      "eval_logps/chosen": -3.5034563541412354,
+      "eval_logps/rejected": -6.952638149261475,
+      "eval_loss": 0.2533933222293854,
+      "eval_rewards/accuracies": 0.9405940771102905,
+      "eval_rewards/chosen": -5.255184173583984,
+      "eval_rewards/margins": 5.173771858215332,
+      "eval_rewards/rejected": -10.428956031799316,
+      "eval_runtime": 30.005,
+      "eval_samples_per_second": 26.696,
+      "eval_steps_per_second": 3.366,
+      "step": 650
+    },
+    {
+      "epoch": 0.5327951564076691,
+      "grad_norm": 4.812018871307373,
+      "learning_rate": 2.9684532864643123e-06,
+      "logits/chosen": 2.8078818321228027,
+      "logits/rejected": 2.887716770172119,
+      "logps/chosen": -3.8643996715545654,
+      "logps/rejected": -7.7119011878967285,
+      "loss": 0.2228,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.796599388122559,
+      "rewards/margins": 5.771251678466797,
+      "rewards/rejected": -11.567851066589355,
+      "step": 660
+    },
+    {
+      "epoch": 0.5408678102926338,
+      "grad_norm": 4.40342378616333,
+      "learning_rate": 2.9169218667902562e-06,
+      "logits/chosen": 1.7638460397720337,
+      "logits/rejected": 1.8495973348617554,
+      "logps/chosen": -3.6541030406951904,
+      "logps/rejected": -7.6542510986328125,
+      "loss": 0.2443,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.481154441833496,
+      "rewards/margins": 6.000222682952881,
+      "rewards/rejected": -11.481378555297852,
+      "step": 670
+    },
+    {
+      "epoch": 0.5489404641775983,
+      "grad_norm": 3.875150680541992,
+      "learning_rate": 2.8652075714060296e-06,
+      "logits/chosen": 2.295011043548584,
+      "logits/rejected": 2.3379902839660645,
+      "logps/chosen": -3.6108944416046143,
+      "logps/rejected": -7.423912048339844,
+      "loss": 0.2413,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.4163408279418945,
+      "rewards/margins": 5.7195258140563965,
+      "rewards/rejected": -11.13586711883545,
+      "step": 680
+    },
+    {
+      "epoch": 0.557013118062563,
+      "grad_norm": 2.429936408996582,
+      "learning_rate": 2.813333083910761e-06,
+      "logits/chosen": 2.482734441757202,
+      "logits/rejected": 2.283761978149414,
+      "logps/chosen": -4.071568965911865,
+      "logps/rejected": -8.294123649597168,
+      "loss": 0.2356,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.107354164123535,
+      "rewards/margins": 6.333830833435059,
+      "rewards/rejected": -12.44118595123291,
+      "step": 690
+    },
+    {
+      "epoch": 0.5650857719475277,
+      "grad_norm": 6.751905918121338,
+      "learning_rate": 2.761321158169134e-06,
+      "logits/chosen": 2.7583765983581543,
+      "logits/rejected": 2.3130671977996826,
+      "logps/chosen": -3.6415677070617676,
+      "logps/rejected": -8.02357006072998,
+      "loss": 0.212,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.4623517990112305,
+      "rewards/margins": 6.573002815246582,
+      "rewards/rejected": -12.035354614257812,
+      "step": 700
+    },
+    {
+      "epoch": 0.5650857719475277,
+      "eval_logits/chosen": 2.289931058883667,
+      "eval_logits/rejected": 2.0797195434570312,
+      "eval_logps/chosen": -3.6428418159484863,
+      "eval_logps/rejected": -7.301985263824463,
+      "eval_loss": 0.24302434921264648,
+      "eval_rewards/accuracies": 0.9504950642585754,
+      "eval_rewards/chosen": -5.464262962341309,
+      "eval_rewards/margins": 5.48871374130249,
+      "eval_rewards/rejected": -10.952977180480957,
+      "eval_runtime": 30.0073,
+      "eval_samples_per_second": 26.693,
+      "eval_steps_per_second": 3.366,
+      "step": 700
+    },
+    {
+      "epoch": 0.5731584258324924,
+      "grad_norm": 3.459409475326538,
+      "learning_rate": 2.70919460833079e-06,
+      "logits/chosen": 2.6869912147521973,
+      "logits/rejected": 2.5928313732147217,
+      "logps/chosen": -3.7079944610595703,
+      "logps/rejected": -8.225436210632324,
+      "loss": 0.192,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.5619916915893555,
+      "rewards/margins": 6.776162624359131,
+      "rewards/rejected": -12.338154792785645,
+      "step": 710
+    },
+    {
+      "epoch": 0.5812310797174571,
+      "grad_norm": 3.1674091815948486,
+      "learning_rate": 2.6569762988232838e-06,
+      "logits/chosen": 2.8864941596984863,
+      "logits/rejected": 2.7974016666412354,
+      "logps/chosen": -3.6625328063964844,
+      "logps/rejected": -7.5159406661987305,
+      "loss": 0.2578,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.493799686431885,
+      "rewards/margins": 5.780111789703369,
+      "rewards/rejected": -11.273911476135254,
+      "step": 720
+    },
+    {
+      "epoch": 0.5893037336024218,
+      "grad_norm": 6.737880229949951,
+      "learning_rate": 2.604689134322999e-06,
+      "logits/chosen": 2.790544033050537,
+      "logits/rejected": 2.9442079067230225,
+      "logps/chosen": -3.4699268341064453,
+      "logps/rejected": -7.267726898193359,
+      "loss": 0.2569,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.204890727996826,
+      "rewards/margins": 5.6967010498046875,
+      "rewards/rejected": -10.901591300964355,
+      "step": 730
+    },
+    {
+      "epoch": 0.5973763874873865,
+      "grad_norm": 5.664917945861816,
+      "learning_rate": 2.5523560497083927e-06,
+      "logits/chosen": 2.466938018798828,
+      "logits/rejected": 2.281191349029541,
+      "logps/chosen": -3.6285691261291504,
+      "logps/rejected": -7.843980312347412,
+      "loss": 0.205,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.442853927612305,
+      "rewards/margins": 6.323116302490234,
+      "rewards/rejected": -11.765970230102539,
+      "step": 740
+    },
+    {
+      "epoch": 0.6054490413723511,
+      "grad_norm": 5.38569450378418,
+      "learning_rate": 2.5e-06,
+      "logits/chosen": 2.642348527908325,
+      "logits/rejected": 2.497035503387451,
+      "logps/chosen": -3.4735655784606934,
+      "logps/rejected": -8.545342445373535,
+      "loss": 0.2279,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.210348129272461,
+      "rewards/margins": 7.607665061950684,
+      "rewards/rejected": -12.818013191223145,
+      "step": 750
+    },
+    {
+      "epoch": 0.6054490413723511,
+      "eval_logits/chosen": 2.3449151515960693,
+      "eval_logits/rejected": 2.140186309814453,
+      "eval_logps/chosen": -3.696176290512085,
+      "eval_logps/rejected": -7.5047287940979,
+      "eval_loss": 0.23516784608364105,
+      "eval_rewards/accuracies": 0.9603960514068604,
+      "eval_rewards/chosen": -5.544264793395996,
+      "eval_rewards/margins": 5.712828159332275,
+      "eval_rewards/rejected": -11.25709342956543,
+      "eval_runtime": 30.0103,
+      "eval_samples_per_second": 26.691,
+      "eval_steps_per_second": 3.366,
+      "step": 750
+    },
+    {
+      "epoch": 0.6135216952573158,
+      "grad_norm": 4.593304634094238,
+      "learning_rate": 2.447643950291608e-06,
+      "logits/chosen": 2.9201483726501465,
+      "logits/rejected": 2.6473968029022217,
+      "logps/chosen": -3.9514071941375732,
+      "logps/rejected": -8.410246849060059,
+      "loss": 0.2413,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.927111625671387,
+      "rewards/margins": 6.688260555267334,
+      "rewards/rejected": -12.615371704101562,
+      "step": 760
+    },
+    {
+      "epoch": 0.6215943491422805,
+      "grad_norm": 8.970620155334473,
+      "learning_rate": 2.3953108656770018e-06,
+      "logits/chosen": 2.294482707977295,
+      "logits/rejected": 2.2701497077941895,
+      "logps/chosen": -4.16445255279541,
+      "logps/rejected": -8.862281799316406,
+      "loss": 0.2045,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.246678352355957,
+      "rewards/margins": 7.0467424392700195,
+      "rewards/rejected": -13.293420791625977,
+      "step": 770
+    },
+    {
+      "epoch": 0.6296670030272452,
+      "grad_norm": 4.757198333740234,
+      "learning_rate": 2.3430237011767166e-06,
+      "logits/chosen": 2.465088367462158,
+      "logits/rejected": 2.3629677295684814,
+      "logps/chosen": -3.9091827869415283,
+      "logps/rejected": -7.637685298919678,
+      "loss": 0.2382,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.863774299621582,
+      "rewards/margins": 5.5927534103393555,
+      "rewards/rejected": -11.456528663635254,
+      "step": 780
+    },
+    {
+      "epoch": 0.6377396569122099,
+      "grad_norm": 3.500101089477539,
+      "learning_rate": 2.290805391669212e-06,
+      "logits/chosen": 2.6884655952453613,
+      "logits/rejected": 2.5679004192352295,
+      "logps/chosen": -3.4037024974823,
+      "logps/rejected": -8.056086540222168,
+      "loss": 0.2333,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.105554103851318,
+      "rewards/margins": 6.978575706481934,
+      "rewards/rejected": -12.08413028717041,
+      "step": 790
+    },
+    {
+      "epoch": 0.6458123107971746,
+      "grad_norm": 3.8024752140045166,
+      "learning_rate": 2.238678841830867e-06,
+      "logits/chosen": 2.357815742492676,
+      "logits/rejected": 2.6706268787384033,
+      "logps/chosen": -3.713919162750244,
+      "logps/rejected": -7.829472541809082,
+      "loss": 0.2555,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.5708794593811035,
+      "rewards/margins": 6.173328876495361,
+      "rewards/rejected": -11.744208335876465,
+      "step": 800
+    },
+    {
+      "epoch": 0.6458123107971746,
+      "eval_logits/chosen": 2.3652234077453613,
+      "eval_logits/rejected": 2.1688482761383057,
+      "eval_logps/chosen": -3.6815528869628906,
+      "eval_logps/rejected": -7.657222270965576,
+      "eval_loss": 0.22401823103427887,
+      "eval_rewards/accuracies": 0.9702970385551453,
+      "eval_rewards/chosen": -5.522329330444336,
+      "eval_rewards/margins": 5.963504791259766,
+      "eval_rewards/rejected": -11.485835075378418,
+      "eval_runtime": 30.0067,
+      "eval_samples_per_second": 26.694,
+      "eval_steps_per_second": 3.366,
+      "step": 800
+    },
+    {
+      "epoch": 0.6538849646821393,
+      "grad_norm": 3.282121419906616,
+      "learning_rate": 2.186666916089239e-06,
+      "logits/chosen": 2.4973549842834473,
+      "logits/rejected": 2.5276618003845215,
+      "logps/chosen": -4.102963447570801,
+      "logps/rejected": -8.789884567260742,
+      "loss": 0.2228,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.154445648193359,
+      "rewards/margins": 7.030381679534912,
+      "rewards/rejected": -13.184826850891113,
+      "step": 810
+    },
+    {
+      "epoch": 0.6619576185671039,
+      "grad_norm": 3.714979410171509,
+      "learning_rate": 2.134792428593971e-06,
+      "logits/chosen": 2.866018772125244,
+      "logits/rejected": 2.7473669052124023,
+      "logps/chosen": -3.602524518966675,
+      "logps/rejected": -7.64529275894165,
+      "loss": 0.2156,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.403786659240723,
+      "rewards/margins": 6.064152717590332,
+      "rewards/rejected": -11.467939376831055,
+      "step": 820
+    },
+    {
+      "epoch": 0.6700302724520686,
+      "grad_norm": 4.395923614501953,
+      "learning_rate": 2.0830781332097446e-06,
+      "logits/chosen": 3.2006161212921143,
+      "logits/rejected": 3.0712599754333496,
+      "logps/chosen": -3.731093168258667,
+      "logps/rejected": -7.530323028564453,
+      "loss": 0.2229,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.596639633178711,
+      "rewards/margins": 5.6988444328308105,
+      "rewards/rejected": -11.295483589172363,
+      "step": 830
+    },
+    {
+      "epoch": 0.6781029263370333,
+      "grad_norm": 4.669759273529053,
+      "learning_rate": 2.031546713535688e-06,
+      "logits/chosen": 2.637826919555664,
+      "logits/rejected": 2.753279209136963,
+      "logps/chosen": -4.231513500213623,
+      "logps/rejected": -8.334096908569336,
+      "loss": 0.2255,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.3472700119018555,
+      "rewards/margins": 6.15387487411499,
+      "rewards/rejected": -12.501144409179688,
+      "step": 840
+    },
+    {
+      "epoch": 0.686175580221998,
+      "grad_norm": 3.5380821228027344,
+      "learning_rate": 1.9802207729556023e-06,
+      "logits/chosen": 2.7282066345214844,
+      "logits/rejected": 2.3621890544891357,
+      "logps/chosen": -3.6814780235290527,
+      "logps/rejected": -8.813811302185059,
+      "loss": 0.2116,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.522217273712158,
+      "rewards/margins": 7.698500633239746,
+      "rewards/rejected": -13.220718383789062,
+      "step": 850
+    },
+    {
+      "epoch": 0.686175580221998,
+      "eval_logits/chosen": 2.483670473098755,
+      "eval_logits/rejected": 2.245927333831787,
+      "eval_logps/chosen": -3.632673501968384,
+      "eval_logps/rejected": -7.6703104972839355,
+      "eval_loss": 0.21706117689609528,
+      "eval_rewards/accuracies": 0.9603960514068604,
+      "eval_rewards/chosen": -5.449010848999023,
+      "eval_rewards/margins": 6.056455612182617,
+      "eval_rewards/rejected": -11.505465507507324,
+      "eval_runtime": 29.9959,
+      "eval_samples_per_second": 26.704,
+      "eval_steps_per_second": 3.367,
+      "step": 850
+    },
+    {
+      "epoch": 0.6942482341069627,
+      "grad_norm": 3.1895673274993896,
+      "learning_rate": 1.9291228247233607e-06,
+      "logits/chosen": 2.936154842376709,
+      "logits/rejected": 2.8326685428619385,
+      "logps/chosen": -3.70434308052063,
+      "logps/rejected": -7.762692451477051,
+      "loss": 0.237,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.556514739990234,
+      "rewards/margins": 6.087524890899658,
+      "rewards/rejected": -11.644038200378418,
+      "step": 860
+    },
+    {
+      "epoch": 0.7023208879919274,
+      "grad_norm": 3.543269395828247,
+      "learning_rate": 1.8782752820878636e-06,
+      "logits/chosen": 3.0214743614196777,
+      "logits/rejected": 2.7598631381988525,
+      "logps/chosen": -4.115224361419678,
+      "logps/rejected": -8.607885360717773,
+      "loss": 0.2437,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.172836780548096,
+      "rewards/margins": 6.738990783691406,
+      "rewards/rejected": -12.911827087402344,
+      "step": 870
+    },
+    {
+      "epoch": 0.7103935418768921,
+      "grad_norm": 4.019063472747803,
+      "learning_rate": 1.827700448461836e-06,
+      "logits/chosen": 2.520110607147217,
+      "logits/rejected": 2.5731875896453857,
+      "logps/chosen": -3.931302547454834,
+      "logps/rejected": -8.647977828979492,
+      "loss": 0.2125,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.896953582763672,
+      "rewards/margins": 7.075013160705566,
+      "rewards/rejected": -12.971966743469238,
+      "step": 880
+    },
+    {
+      "epoch": 0.7184661957618567,
+      "grad_norm": 2.3453805446624756,
+      "learning_rate": 1.7774205076388207e-06,
+      "logits/chosen": 3.297231674194336,
+      "logits/rejected": 2.8459277153015137,
+      "logps/chosen": -4.021622657775879,
+      "logps/rejected": -9.263435363769531,
+      "loss": 0.1604,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.032434463500977,
+      "rewards/margins": 7.862718105316162,
+      "rewards/rejected": -13.89515209197998,
+      "step": 890
+    },
+    {
+      "epoch": 0.7265388496468214,
+      "grad_norm": 5.522162914276123,
+      "learning_rate": 1.7274575140626318e-06,
+      "logits/chosen": 2.0340800285339355,
+      "logits/rejected": 2.0777273178100586,
+      "logps/chosen": -3.8453431129455566,
+      "logps/rejected": -9.153608322143555,
+      "loss": 0.2305,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -5.768013954162598,
+      "rewards/margins": 7.962396144866943,
+      "rewards/rejected": -13.7304105758667,
+      "step": 900
+    },
+    {
+      "epoch": 0.7265388496468214,
+      "eval_logits/chosen": 2.456989049911499,
+      "eval_logits/rejected": 2.3152406215667725,
+      "eval_logps/chosen": -3.8842220306396484,
+      "eval_logps/rejected": -8.038114547729492,
+      "eval_loss": 0.21270446479320526,
+      "eval_rewards/accuracies": 0.9801980257034302,
+      "eval_rewards/chosen": -5.826333999633789,
+      "eval_rewards/margins": 6.230839729309082,
+      "eval_rewards/rejected": -12.057172775268555,
+      "eval_runtime": 30.0018,
+      "eval_samples_per_second": 26.698,
+      "eval_steps_per_second": 3.366,
+      "step": 900
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 1500,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 50,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.1917477978484244e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-900/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b4baf06be36eabcb39d3d1f4e89871aeacdc0329940b46389df0d3da443c59c
+size 7224

checkpoint-900/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)