Training in progress, step 950, checkpoint

checkpoint-950/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-950/adapter_config.json

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

checkpoint-950/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b8efbac71665abd10ce94a37cb1f7cc15dda9ee73479e57f21c8feb9643d4e7
+size 25200088

checkpoint-950/global_step950/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:faa01c6b04e4aa65a52f2147c2b59ff8b4d62416db615a7e6d4fcd5e60ca8da3
+size 18881328

checkpoint-950/global_step950/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ad015ee8586f22096b57ad415f676068903697232e3b0630751b1e13b3a0d99
+size 18881328

checkpoint-950/global_step950/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:ba5ca8dbcc086e72040b2cab036c8437281068aa63ba5e2820159249e545b102
+size 18881328

checkpoint-950/global_step950/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c83510eac5a64e611143a93f33a2c543d2d39d503fd2455e096b517833b9357
+size 18881392

checkpoint-950/global_step950/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:c1d5484104dc36626d6867fb43ccd83b85618810f725ba4afca10d2d14bfd05f
+size 18881392

checkpoint-950/global_step950/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7beacbed1728ca2161c9ef79e656df77edb3f84b905a975af46e858a8cce7efc
+size 18881392

checkpoint-950/global_step950/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:ace43982ea8767681a64e0f6ea21a6cd7a0250d3042f0006002f089a72085d21
+size 18881392

checkpoint-950/global_step950/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d913dd7567c0d660288763a27b965aa34e4d1494ce7aefdea42e45fe9c07b5a
+size 18881392

checkpoint-950/global_step950/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a93cbe08b7fc32d1ea6bb53524f85532917f7ce11c5df6e69a485e027c889ca
+size 25379244

checkpoint-950/latest

@@ -0,0 +1 @@
+global_step950

checkpoint-950/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc0b4819898a49cb50f6fb7bab4d801df3c0173ece6ef2ff38f36207597856d0
+size 15984

checkpoint-950/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae28471a0a04a53d3ba94b6bbeb3d304fb092f7bb69e61fbcd60ade743970b87
+size 15984

checkpoint-950/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f12ddc933576ec84d344c12bddeb7de08d5c44d45bf1bc029a542a7a06ff2adc
+size 15984

checkpoint-950/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19328fcb03008873543059492b92b3800cb0049a38d31ca04fe302a8bf5456ad
+size 15984

checkpoint-950/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7655a205ef4a9a18c7a399dbda37bb18fc5c674f87796d6d28102e102e54a9de
+size 15984

checkpoint-950/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e69388d4203ea379b10a270c3adb691e4f32323504669b7636bf09f3404a904
+size 15984

checkpoint-950/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46badbd50f32686ee554c7eb87f933665e9fed831942f21e950e87da676c9596
+size 15984

checkpoint-950/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:662673c4fd4eeb350cb25f2d55ea54a71a5aa0fd48aff7b60e3a32e64b494e8e
+size 15984

checkpoint-950/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:512e1e5e95d661cfadb7039d790a1233b423c22e8fec0c0e5f038c40be21d11e
+size 1064

checkpoint-950/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-950/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-950/trainer_state.json

@@ -0,0 +1,1762 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.8165019338203696,
+  "eval_steps": 50,
+  "global_step": 950,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008594757198109154,
+      "grad_norm": 0.05934199318289757,
+      "learning_rate": 4.999451708687114e-06,
+      "logits/chosen": 14.762972831726074,
+      "logits/rejected": 15.199728012084961,
+      "logps/chosen": -0.3259914815425873,
+      "logps/rejected": -0.34297481179237366,
+      "loss": 0.9377,
+      "rewards/accuracies": 0.4000000059604645,
+      "rewards/chosen": -0.4889872074127197,
+      "rewards/margins": 0.02547495998442173,
+      "rewards/rejected": -0.5144621729850769,
+      "step": 10
+    },
+    {
+      "epoch": 0.017189514396218308,
+      "grad_norm": 0.06342790275812149,
+      "learning_rate": 4.997807075247147e-06,
+      "logits/chosen": 14.351249694824219,
+      "logits/rejected": 15.068448066711426,
+      "logps/chosen": -0.2809392511844635,
+      "logps/rejected": -0.3711296617984772,
+      "loss": 0.9352,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.42140883207321167,
+      "rewards/margins": 0.1352856159210205,
+      "rewards/rejected": -0.5566944479942322,
+      "step": 20
+    },
+    {
+      "epoch": 0.02578427159432746,
+      "grad_norm": 0.053961098194122314,
+      "learning_rate": 4.9950668210706795e-06,
+      "logits/chosen": 14.636960983276367,
+      "logits/rejected": 15.265243530273438,
+      "logps/chosen": -0.2820780873298645,
+      "logps/rejected": -0.34024301171302795,
+      "loss": 0.9351,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.42311716079711914,
+      "rewards/margins": 0.08724743127822876,
+      "rewards/rejected": -0.5103646516799927,
+      "step": 30
+    },
+    {
+      "epoch": 0.034379028792436615,
+      "grad_norm": 0.13506193459033966,
+      "learning_rate": 4.9912321481237616e-06,
+      "logits/chosen": 14.4556884765625,
+      "logits/rejected": 15.048967361450195,
+      "logps/chosen": -0.2897028625011444,
+      "logps/rejected": -0.34129124879837036,
+      "loss": 0.922,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.43455424904823303,
+      "rewards/margins": 0.07738252729177475,
+      "rewards/rejected": -0.5119368433952332,
+      "step": 40
+    },
+    {
+      "epoch": 0.042973785990545764,
+      "grad_norm": 0.05230574309825897,
+      "learning_rate": 4.986304738420684e-06,
+      "logits/chosen": 14.628789901733398,
+      "logits/rejected": 15.307828903198242,
+      "logps/chosen": -0.28786614537239075,
+      "logps/rejected": -0.3513876795768738,
+      "loss": 0.9201,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.4317992329597473,
+      "rewards/margins": 0.09528233855962753,
+      "rewards/rejected": -0.5270815491676331,
+      "step": 50
+    },
+    {
+      "epoch": 0.042973785990545764,
+      "eval_logits/chosen": 14.234943389892578,
+      "eval_logits/rejected": 15.258601188659668,
+      "eval_logps/chosen": -0.2844341993331909,
+      "eval_logps/rejected": -0.3695394694805145,
+      "eval_loss": 0.9226060509681702,
+      "eval_rewards/accuracies": 0.5157894492149353,
+      "eval_rewards/chosen": -0.42665132880210876,
+      "eval_rewards/margins": 0.1276579648256302,
+      "eval_rewards/rejected": -0.5543092489242554,
+      "eval_runtime": 25.2466,
+      "eval_samples_per_second": 29.826,
+      "eval_steps_per_second": 3.763,
+      "step": 50
+    },
+    {
+      "epoch": 0.05156854318865492,
+      "grad_norm": 0.09328428655862808,
+      "learning_rate": 4.980286753286196e-06,
+      "logits/chosen": 14.35963249206543,
+      "logits/rejected": 15.055354118347168,
+      "logps/chosen": -0.27534741163253784,
+      "logps/rejected": -0.33098170161247253,
+      "loss": 0.9356,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.4130210876464844,
+      "rewards/margins": 0.08345144242048264,
+      "rewards/rejected": -0.4964725375175476,
+      "step": 60
+    },
+    {
+      "epoch": 0.060163300386764075,
+      "grad_norm": 0.06518550217151642,
+      "learning_rate": 4.973180832407471e-06,
+      "logits/chosen": 14.599525451660156,
+      "logits/rejected": 14.825297355651855,
+      "logps/chosen": -0.2708163857460022,
+      "logps/rejected": -0.3305850923061371,
+      "loss": 0.9257,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.4062245786190033,
+      "rewards/margins": 0.08965305984020233,
+      "rewards/rejected": -0.4958776533603668,
+      "step": 70
+    },
+    {
+      "epoch": 0.06875805758487323,
+      "grad_norm": 0.07543154805898666,
+      "learning_rate": 4.964990092676263e-06,
+      "logits/chosen": 14.947430610656738,
+      "logits/rejected": 15.093690872192383,
+      "logps/chosen": -0.2602943778038025,
+      "logps/rejected": -0.31820863485336304,
+      "loss": 0.9168,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.39044153690338135,
+      "rewards/margins": 0.08687138557434082,
+      "rewards/rejected": -0.47731298208236694,
+      "step": 80
+    },
+    {
+      "epoch": 0.07735281478298238,
+      "grad_norm": 0.06628195196390152,
+      "learning_rate": 4.9557181268217225e-06,
+      "logits/chosen": 14.43529987335205,
+      "logits/rejected": 14.750699043273926,
+      "logps/chosen": -0.2884291708469391,
+      "logps/rejected": -0.34193652868270874,
+      "loss": 0.9273,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.43264374136924744,
+      "rewards/margins": 0.08026103675365448,
+      "rewards/rejected": -0.5129047632217407,
+      "step": 90
+    },
+    {
+      "epoch": 0.08594757198109153,
+      "grad_norm": 0.08684897422790527,
+      "learning_rate": 4.9453690018345144e-06,
+      "logits/chosen": 13.573002815246582,
+      "logits/rejected": 14.441877365112305,
+      "logps/chosen": -0.2569890320301056,
+      "logps/rejected": -0.37049269676208496,
+      "loss": 0.9009,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.3854835629463196,
+      "rewards/margins": 0.17025551199913025,
+      "rewards/rejected": -0.5557390451431274,
+      "step": 100
+    },
+    {
+      "epoch": 0.08594757198109153,
+      "eval_logits/chosen": 14.026633262634277,
+      "eval_logits/rejected": 15.08835220336914,
+      "eval_logps/chosen": -0.2761566936969757,
+      "eval_logps/rejected": -0.3717801570892334,
+      "eval_loss": 0.9138591885566711,
+      "eval_rewards/accuracies": 0.5368421077728271,
+      "eval_rewards/chosen": -0.41423505544662476,
+      "eval_rewards/margins": 0.1434352546930313,
+      "eval_rewards/rejected": -0.5576702952384949,
+      "eval_runtime": 25.2451,
+      "eval_samples_per_second": 29.828,
+      "eval_steps_per_second": 3.763,
+      "step": 100
+    },
+    {
+      "epoch": 0.09454232917920069,
+      "grad_norm": 0.08046824485063553,
+      "learning_rate": 4.933947257182901e-06,
+      "logits/chosen": 14.500630378723145,
+      "logits/rejected": 14.831761360168457,
+      "logps/chosen": -0.30049553513526917,
+      "logps/rejected": -0.3315966725349426,
+      "loss": 0.916,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": -0.45074325799942017,
+      "rewards/margins": 0.04665176197886467,
+      "rewards/rejected": -0.49739497900009155,
+      "step": 110
+    },
+    {
+      "epoch": 0.10313708637730984,
+      "grad_norm": 0.12244562804698944,
+      "learning_rate": 4.921457902821578e-06,
+      "logits/chosen": 14.26713752746582,
+      "logits/rejected": 14.495455741882324,
+      "logps/chosen": -0.2670941650867462,
+      "logps/rejected": -0.32481229305267334,
+      "loss": 0.9167,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.4006412625312805,
+      "rewards/margins": 0.08657723665237427,
+      "rewards/rejected": -0.4872184693813324,
+      "step": 120
+    },
+    {
+      "epoch": 0.11173184357541899,
+      "grad_norm": 0.1828213334083557,
+      "learning_rate": 4.907906416994146e-06,
+      "logits/chosen": 14.009546279907227,
+      "logits/rejected": 14.297094345092773,
+      "logps/chosen": -0.27995598316192627,
+      "logps/rejected": -0.3530685007572174,
+      "loss": 0.9087,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.419933944940567,
+      "rewards/margins": 0.10966875404119492,
+      "rewards/rejected": -0.5296027660369873,
+      "step": 130
+    },
+    {
+      "epoch": 0.12032660077352815,
+      "grad_norm": 0.10407563298940659,
+      "learning_rate": 4.893298743830168e-06,
+      "logits/chosen": 13.689155578613281,
+      "logits/rejected": 14.1933012008667,
+      "logps/chosen": -0.25955715775489807,
+      "logps/rejected": -0.3815004229545593,
+      "loss": 0.9053,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.3893357217311859,
+      "rewards/margins": 0.18291489779949188,
+      "rewards/rejected": -0.5722506046295166,
+      "step": 140
+    },
+    {
+      "epoch": 0.1289213579716373,
+      "grad_norm": 0.10028588026762009,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": 12.851397514343262,
+      "logits/rejected": 13.509778022766113,
+      "logps/chosen": -0.23652991652488708,
+      "logps/rejected": -0.3720462918281555,
+      "loss": 0.8999,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.3547949194908142,
+      "rewards/margins": 0.2032744586467743,
+      "rewards/rejected": -0.5580693483352661,
+      "step": 150
+    },
+    {
+      "epoch": 0.1289213579716373,
+      "eval_logits/chosen": 12.384929656982422,
+      "eval_logits/rejected": 13.672826766967773,
+      "eval_logps/chosen": -0.27857670187950134,
+      "eval_logps/rejected": -0.4014737904071808,
+      "eval_loss": 0.8956203460693359,
+      "eval_rewards/accuracies": 0.5684210658073425,
+      "eval_rewards/chosen": -0.4178650677204132,
+      "eval_rewards/margins": 0.18434564769268036,
+      "eval_rewards/rejected": -0.6022107601165771,
+      "eval_runtime": 25.2468,
+      "eval_samples_per_second": 29.826,
+      "eval_steps_per_second": 3.763,
+      "step": 150
+    },
+    {
+      "epoch": 0.13751611516974646,
+      "grad_norm": 0.12453093379735947,
+      "learning_rate": 4.860940925593703e-06,
+      "logits/chosen": 12.110003471374512,
+      "logits/rejected": 13.076980590820312,
+      "logps/chosen": -0.27192068099975586,
+      "logps/rejected": -0.3863692879676819,
+      "loss": 0.8907,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.4078810214996338,
+      "rewards/margins": 0.1716729700565338,
+      "rewards/rejected": -0.5795539617538452,
+      "step": 160
+    },
+    {
+      "epoch": 0.1461108723678556,
+      "grad_norm": 0.17137788236141205,
+      "learning_rate": 4.84320497372973e-06,
+      "logits/chosen": 11.92918586730957,
+      "logits/rejected": 12.573629379272461,
+      "logps/chosen": -0.27472984790802,
+      "logps/rejected": -0.41249385476112366,
+      "loss": 0.8831,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.41209474205970764,
+      "rewards/margins": 0.20664596557617188,
+      "rewards/rejected": -0.6187406778335571,
+      "step": 170
+    },
+    {
+      "epoch": 0.15470562956596476,
+      "grad_norm": 0.3904883861541748,
+      "learning_rate": 4.824441214720629e-06,
+      "logits/chosen": 11.182531356811523,
+      "logits/rejected": 12.176573753356934,
+      "logps/chosen": -0.2953718304634094,
+      "logps/rejected": -0.4208717942237854,
+      "loss": 0.8736,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.4430577754974365,
+      "rewards/margins": 0.18824996054172516,
+      "rewards/rejected": -0.6313077211380005,
+      "step": 180
+    },
+    {
+      "epoch": 0.1633003867640739,
+      "grad_norm": 0.17574089765548706,
+      "learning_rate": 4.804657878971252e-06,
+      "logits/chosen": 10.119890213012695,
+      "logits/rejected": 11.05900764465332,
+      "logps/chosen": -0.29340866208076477,
+      "logps/rejected": -0.4555762708187103,
+      "loss": 0.884,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.44011297821998596,
+      "rewards/margins": 0.24325144290924072,
+      "rewards/rejected": -0.6833644509315491,
+      "step": 190
+    },
+    {
+      "epoch": 0.17189514396218306,
+      "grad_norm": 0.2242884337902069,
+      "learning_rate": 4.783863644106502e-06,
+      "logits/chosen": 9.674784660339355,
+      "logits/rejected": 10.418611526489258,
+      "logps/chosen": -0.3504490852355957,
+      "logps/rejected": -0.5431731939315796,
+      "loss": 0.8419,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.5256736278533936,
+      "rewards/margins": 0.2890861928462982,
+      "rewards/rejected": -0.8147598505020142,
+      "step": 200
+    },
+    {
+      "epoch": 0.17189514396218306,
+      "eval_logits/chosen": 7.944870471954346,
+      "eval_logits/rejected": 8.979729652404785,
+      "eval_logps/chosen": -0.33341673016548157,
+      "eval_logps/rejected": -0.5431775450706482,
+      "eval_loss": 0.8462886810302734,
+      "eval_rewards/accuracies": 0.6000000238418579,
+      "eval_rewards/chosen": -0.5001251101493835,
+      "eval_rewards/margins": 0.3146411180496216,
+      "eval_rewards/rejected": -0.8147663474082947,
+      "eval_runtime": 25.2453,
+      "eval_samples_per_second": 29.827,
+      "eval_steps_per_second": 3.763,
+      "step": 200
+    },
+    {
+      "epoch": 0.18048990116029223,
+      "grad_norm": 0.32119837403297424,
+      "learning_rate": 4.762067631165049e-06,
+      "logits/chosen": 7.16138219833374,
+      "logits/rejected": 8.43680477142334,
+      "logps/chosen": -0.36649250984191895,
+      "logps/rejected": -0.5420924425125122,
+      "loss": 0.8187,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.5497387647628784,
+      "rewards/margins": 0.2633998692035675,
+      "rewards/rejected": -0.8131386041641235,
+      "step": 210
+    },
+    {
+      "epoch": 0.18908465835840138,
+      "grad_norm": 0.48516562581062317,
+      "learning_rate": 4.7392794005985324e-06,
+      "logits/chosen": 4.770083427429199,
+      "logits/rejected": 5.710458278656006,
+      "logps/chosen": -0.34041497111320496,
+      "logps/rejected": -0.6309320330619812,
+      "loss": 0.8448,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.510622501373291,
+      "rewards/margins": 0.4357755780220032,
+      "rewards/rejected": -0.9463980793952942,
+      "step": 220
+    },
+    {
+      "epoch": 0.19767941555651053,
+      "grad_norm": 0.29154208302497864,
+      "learning_rate": 4.715508948078037e-06,
+      "logits/chosen": 5.168765068054199,
+      "logits/rejected": 5.421420574188232,
+      "logps/chosen": -0.3792352080345154,
+      "logps/rejected": -0.65748131275177,
+      "loss": 0.8066,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.5688528418540955,
+      "rewards/margins": 0.41736921668052673,
+      "rewards/rejected": -0.986221969127655,
+      "step": 230
+    },
+    {
+      "epoch": 0.20627417275461968,
+      "grad_norm": 0.42973750829696655,
+      "learning_rate": 4.690766700109659e-06,
+      "logits/chosen": 4.204717636108398,
+      "logits/rejected": 3.706291913986206,
+      "logps/chosen": -0.39414530992507935,
+      "logps/rejected": -0.7194588780403137,
+      "loss": 0.7787,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.5912179350852966,
+      "rewards/margins": 0.4879704415798187,
+      "rewards/rejected": -1.079188346862793,
+      "step": 240
+    },
+    {
+      "epoch": 0.21486892995272883,
+      "grad_norm": 0.5244571566581726,
+      "learning_rate": 4.665063509461098e-06,
+      "logits/chosen": 3.335484743118286,
+      "logits/rejected": 3.3176345825195312,
+      "logps/chosen": -0.4493131637573242,
+      "logps/rejected": -0.8293434381484985,
+      "loss": 0.7776,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.6739697456359863,
+      "rewards/margins": 0.5700454115867615,
+      "rewards/rejected": -1.244015097618103,
+      "step": 250
+    },
+    {
+      "epoch": 0.21486892995272883,
+      "eval_logits/chosen": 2.590949058532715,
+      "eval_logits/rejected": 2.2929749488830566,
+      "eval_logps/chosen": -0.48714593052864075,
+      "eval_logps/rejected": -0.9267774224281311,
+      "eval_loss": 0.7469337582588196,
+      "eval_rewards/accuracies": 0.6526315808296204,
+      "eval_rewards/chosen": -0.7307189106941223,
+      "eval_rewards/margins": 0.659447193145752,
+      "eval_rewards/rejected": -1.390166163444519,
+      "eval_runtime": 25.2479,
+      "eval_samples_per_second": 29.824,
+      "eval_steps_per_second": 3.763,
+      "step": 250
+    },
+    {
+      "epoch": 0.22346368715083798,
+      "grad_norm": 0.39347293972969055,
+      "learning_rate": 4.638410650401267e-06,
+      "logits/chosen": 2.2975668907165527,
+      "logits/rejected": 1.2855035066604614,
+      "logps/chosen": -0.5228341817855835,
+      "logps/rejected": -1.00227952003479,
+      "loss": 0.6981,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.78425133228302,
+      "rewards/margins": 0.7191681265830994,
+      "rewards/rejected": -1.5034195184707642,
+      "step": 260
+    },
+    {
+      "epoch": 0.23205844434894715,
+      "grad_norm": 0.69575434923172,
+      "learning_rate": 4.610819813755038e-06,
+      "logits/chosen": 2.8782780170440674,
+      "logits/rejected": 1.9394336938858032,
+      "logps/chosen": -0.4982885718345642,
+      "logps/rejected": -1.035541296005249,
+      "loss": 0.7174,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.7474328875541687,
+      "rewards/margins": 0.8058789372444153,
+      "rewards/rejected": -1.5533119440078735,
+      "step": 270
+    },
+    {
+      "epoch": 0.2406532015470563,
+      "grad_norm": 0.7858326435089111,
+      "learning_rate": 4.582303101775249e-06,
+      "logits/chosen": 2.710908889770508,
+      "logits/rejected": 1.6444288492202759,
+      "logps/chosen": -0.600068211555481,
+      "logps/rejected": -1.1271780729293823,
+      "loss": 0.6972,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.9001023173332214,
+      "rewards/margins": 0.7906648516654968,
+      "rewards/rejected": -1.6907672882080078,
+      "step": 280
+    },
+    {
+      "epoch": 0.24924795874516545,
+      "grad_norm": 0.7384620904922485,
+      "learning_rate": 4.55287302283426e-06,
+      "logits/chosen": 1.5841500759124756,
+      "logits/rejected": 0.640514612197876,
+      "logps/chosen": -0.6465060710906982,
+      "logps/rejected": -1.4245095252990723,
+      "loss": 0.6192,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.9697591066360474,
+      "rewards/margins": 1.1670053005218506,
+      "rewards/rejected": -2.1367642879486084,
+      "step": 290
+    },
+    {
+      "epoch": 0.2578427159432746,
+      "grad_norm": 0.8262321352958679,
+      "learning_rate": 4.522542485937369e-06,
+      "logits/chosen": 1.7300422191619873,
+      "logits/rejected": 0.7782856225967407,
+      "logps/chosen": -0.7083590626716614,
+      "logps/rejected": -1.6742557287216187,
+      "loss": 0.5721,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -1.062538504600525,
+      "rewards/margins": 1.4488452672958374,
+      "rewards/rejected": -2.511383533477783,
+      "step": 300
+    },
+    {
+      "epoch": 0.2578427159432746,
+      "eval_logits/chosen": 1.3559931516647339,
+      "eval_logits/rejected": 0.6592276096343994,
+      "eval_logps/chosen": -0.7815767526626587,
+      "eval_logps/rejected": -2.1176154613494873,
+      "eval_loss": 0.5730626583099365,
+      "eval_rewards/accuracies": 0.7052631378173828,
+      "eval_rewards/chosen": -1.1723653078079224,
+      "eval_rewards/margins": 2.0040581226348877,
+      "eval_rewards/rejected": -3.1764233112335205,
+      "eval_runtime": 25.2488,
+      "eval_samples_per_second": 29.823,
+      "eval_steps_per_second": 3.763,
+      "step": 300
+    },
+    {
+      "epoch": 0.2664374731413838,
+      "grad_norm": 0.8472572565078735,
+      "learning_rate": 4.491324795060491e-06,
+      "logits/chosen": 1.4461088180541992,
+      "logits/rejected": 0.49669915437698364,
+      "logps/chosen": -0.7694377899169922,
+      "logps/rejected": -2.362783432006836,
+      "loss": 0.5091,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -1.1541565656661987,
+      "rewards/margins": 2.390018939971924,
+      "rewards/rejected": -3.544174909591675,
+      "step": 310
+    },
+    {
+      "epoch": 0.2750322303394929,
+      "grad_norm": 0.41847530007362366,
+      "learning_rate": 4.4592336433146e-06,
+      "logits/chosen": 2.172646999359131,
+      "logits/rejected": 1.0526962280273438,
+      "logps/chosen": -0.7410945296287537,
+      "logps/rejected": -1.9158353805541992,
+      "loss": 0.5352,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.1116416454315186,
+      "rewards/margins": 1.7621114253997803,
+      "rewards/rejected": -2.873753070831299,
+      "step": 320
+    },
+    {
+      "epoch": 0.28362698753760207,
+      "grad_norm": 1.7422096729278564,
+      "learning_rate": 4.426283106939474e-06,
+      "logits/chosen": 2.611234188079834,
+      "logits/rejected": 1.7068111896514893,
+      "logps/chosen": -0.8319486379623413,
+      "logps/rejected": -2.32024884223938,
+      "loss": 0.5397,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -1.2479230165481567,
+      "rewards/margins": 2.232450008392334,
+      "rewards/rejected": -3.480372905731201,
+      "step": 330
+    },
+    {
+      "epoch": 0.2922217447357112,
+      "grad_norm": 0.8699240684509277,
+      "learning_rate": 4.3924876391293915e-06,
+      "logits/chosen": 1.996747612953186,
+      "logits/rejected": 1.1473515033721924,
+      "logps/chosen": -0.8445833921432495,
+      "logps/rejected": -2.675687551498413,
+      "loss": 0.4817,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -1.2668750286102295,
+      "rewards/margins": 2.7466559410095215,
+      "rewards/rejected": -4.01353120803833,
+      "step": 340
+    },
+    {
+      "epoch": 0.30081650193382037,
+      "grad_norm": 2.089289426803589,
+      "learning_rate": 4.357862063693486e-06,
+      "logits/chosen": 1.7134803533554077,
+      "logits/rejected": 1.3000510931015015,
+      "logps/chosen": -0.8976927995681763,
+      "logps/rejected": -2.1593873500823975,
+      "loss": 0.5098,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -1.3465392589569092,
+      "rewards/margins": 1.8925418853759766,
+      "rewards/rejected": -3.2390809059143066,
+      "step": 350
+    },
+    {
+      "epoch": 0.30081650193382037,
+      "eval_logits/chosen": 1.6772903203964233,
+      "eval_logits/rejected": 1.2370609045028687,
+      "eval_logps/chosen": -0.9737761616706848,
+      "eval_logps/rejected": -3.1528680324554443,
+      "eval_loss": 0.5162621736526489,
+      "eval_rewards/accuracies": 0.7263157963752747,
+      "eval_rewards/chosen": -1.46066415309906,
+      "eval_rewards/margins": 3.2686376571655273,
+      "eval_rewards/rejected": -4.729301929473877,
+      "eval_runtime": 25.251,
+      "eval_samples_per_second": 29.821,
+      "eval_steps_per_second": 3.762,
+      "step": 350
+    },
+    {
+      "epoch": 0.3094112591319295,
+      "grad_norm": 0.47079572081565857,
+      "learning_rate": 4.322421568553529e-06,
+      "logits/chosen": 1.9561872482299805,
+      "logits/rejected": 0.8960329294204712,
+      "logps/chosen": -0.9378088712692261,
+      "logps/rejected": -2.8065876960754395,
+      "loss": 0.5046,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.4067132472991943,
+      "rewards/margins": 2.8031680583953857,
+      "rewards/rejected": -4.209881782531738,
+      "step": 360
+    },
+    {
+      "epoch": 0.31800601633003867,
+      "grad_norm": 0.6202365159988403,
+      "learning_rate": 4.286181699082008e-06,
+      "logits/chosen": 2.152726411819458,
+      "logits/rejected": 1.4309433698654175,
+      "logps/chosen": -1.007157564163208,
+      "logps/rejected": -3.3813462257385254,
+      "loss": 0.4526,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -1.5107364654541016,
+      "rewards/margins": 3.561283588409424,
+      "rewards/rejected": -5.072019577026367,
+      "step": 370
+    },
+    {
+      "epoch": 0.3266007735281478,
+      "grad_norm": 1.080393671989441,
+      "learning_rate": 4.249158351283414e-06,
+      "logits/chosen": 1.7528371810913086,
+      "logits/rejected": 1.3293968439102173,
+      "logps/chosen": -1.0258004665374756,
+      "logps/rejected": -2.984057903289795,
+      "loss": 0.4879,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.5387006998062134,
+      "rewards/margins": 2.9373860359191895,
+      "rewards/rejected": -4.476086616516113,
+      "step": 380
+    },
+    {
+      "epoch": 0.33519553072625696,
+      "grad_norm": 1.4520032405853271,
+      "learning_rate": 4.211367764821722e-06,
+      "logits/chosen": 3.061373233795166,
+      "logits/rejected": 2.0103466510772705,
+      "logps/chosen": -1.0191391706466675,
+      "logps/rejected": -2.9054081439971924,
+      "loss": 0.4776,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -1.5287089347839355,
+      "rewards/margins": 2.8294031620025635,
+      "rewards/rejected": -4.358112335205078,
+      "step": 390
+    },
+    {
+      "epoch": 0.3437902879243661,
+      "grad_norm": 0.5479139089584351,
+      "learning_rate": 4.172826515897146e-06,
+      "logits/chosen": 2.8395092487335205,
+      "logits/rejected": 2.0935282707214355,
+      "logps/chosen": -1.0769506692886353,
+      "logps/rejected": -3.11635160446167,
+      "loss": 0.4686,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -1.6154258251190186,
+      "rewards/margins": 3.0591015815734863,
+      "rewards/rejected": -4.674527168273926,
+      "step": 400
+    },
+    {
+      "epoch": 0.3437902879243661,
+      "eval_logits/chosen": 2.5064592361450195,
+      "eval_logits/rejected": 2.108433485031128,
+      "eval_logps/chosen": -1.1957285404205322,
+      "eval_logps/rejected": -3.7678382396698,
+      "eval_loss": 0.46578800678253174,
+      "eval_rewards/accuracies": 0.7368420958518982,
+      "eval_rewards/chosen": -1.793592929840088,
+      "eval_rewards/margins": 3.8581647872924805,
+      "eval_rewards/rejected": -5.651757717132568,
+      "eval_runtime": 25.2478,
+      "eval_samples_per_second": 29.824,
+      "eval_steps_per_second": 3.763,
+      "step": 400
+    },
+    {
+      "epoch": 0.3523850451224753,
+      "grad_norm": 0.9966821670532227,
+      "learning_rate": 4.133551509975264e-06,
+      "logits/chosen": 2.6411917209625244,
+      "logits/rejected": 1.8634885549545288,
+      "logps/chosen": -1.0934125185012817,
+      "logps/rejected": -3.2207794189453125,
+      "loss": 0.4335,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -1.6401188373565674,
+      "rewards/margins": 3.1910502910614014,
+      "rewards/rejected": -4.831169128417969,
+      "step": 410
+    },
+    {
+      "epoch": 0.36097980232058446,
+      "grad_norm": 0.6384722590446472,
+      "learning_rate": 4.093559974371725e-06,
+      "logits/chosen": 3.1368844509124756,
+      "logits/rejected": 2.3800251483917236,
+      "logps/chosen": -1.2108217477798462,
+      "logps/rejected": -3.484806537628174,
+      "loss": 0.4543,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.816232681274414,
+      "rewards/margins": 3.4109771251678467,
+      "rewards/rejected": -5.227209568023682,
+      "step": 420
+    },
+    {
+      "epoch": 0.3695745595186936,
+      "grad_norm": 0.856741726398468,
+      "learning_rate": 4.052869450695776e-06,
+      "logits/chosen": 3.155728816986084,
+      "logits/rejected": 2.257838726043701,
+      "logps/chosen": -1.4214586019515991,
+      "logps/rejected": -4.186622619628906,
+      "loss": 0.4091,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -2.132187604904175,
+      "rewards/margins": 4.1477460861206055,
+      "rewards/rejected": -6.279933929443359,
+      "step": 430
+    },
+    {
+      "epoch": 0.37816931671680276,
+      "grad_norm": 1.3310774564743042,
+      "learning_rate": 4.011497787155938e-06,
+      "logits/chosen": 1.9942185878753662,
+      "logits/rejected": 1.6246827840805054,
+      "logps/chosen": -1.8575637340545654,
+      "logps/rejected": -4.5355329513549805,
+      "loss": 0.3995,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -2.7863457202911377,
+      "rewards/margins": 4.016953945159912,
+      "rewards/rejected": -6.8032989501953125,
+      "step": 440
+    },
+    {
+      "epoch": 0.3867640739149119,
+      "grad_norm": 2.0849101543426514,
+      "learning_rate": 3.969463130731183e-06,
+      "logits/chosen": 2.406555652618408,
+      "logits/rejected": 2.0490009784698486,
+      "logps/chosen": -2.392570972442627,
+      "logps/rejected": -5.055584907531738,
+      "loss": 0.3671,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -3.588855743408203,
+      "rewards/margins": 3.994520902633667,
+      "rewards/rejected": -7.583376884460449,
+      "step": 450
+    },
+    {
+      "epoch": 0.3867640739149119,
+      "eval_logits/chosen": 2.2324020862579346,
+      "eval_logits/rejected": 2.365755319595337,
+      "eval_logps/chosen": -2.736898422241211,
+      "eval_logps/rejected": -5.73967170715332,
+      "eval_loss": 0.3965117633342743,
+      "eval_rewards/accuracies": 0.8736842274665833,
+      "eval_rewards/chosen": -4.105347633361816,
+      "eval_rewards/margins": 4.504159927368164,
+      "eval_rewards/rejected": -8.60950756072998,
+      "eval_runtime": 25.2455,
+      "eval_samples_per_second": 29.827,
+      "eval_steps_per_second": 3.763,
+      "step": 450
+    },
+    {
+      "epoch": 0.39535883111302106,
+      "grad_norm": 2.223949432373047,
+      "learning_rate": 3.92678391921108e-06,
+      "logits/chosen": 2.651564598083496,
+      "logits/rejected": 2.383842945098877,
+      "logps/chosen": -2.591308355331421,
+      "logps/rejected": -5.308972358703613,
+      "loss": 0.3412,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -3.886962413787842,
+      "rewards/margins": 4.07649564743042,
+      "rewards/rejected": -7.963458061218262,
+      "step": 460
+    },
+    {
+      "epoch": 0.4039535883111302,
+      "grad_norm": 3.110624074935913,
+      "learning_rate": 3.88347887310836e-06,
+      "logits/chosen": 2.5435309410095215,
+      "logits/rejected": 2.46763277053833,
+      "logps/chosen": -2.413583993911743,
+      "logps/rejected": -5.543262481689453,
+      "loss": 0.3832,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -3.620375871658325,
+      "rewards/margins": 4.694517135620117,
+      "rewards/rejected": -8.314892768859863,
+      "step": 470
+    },
+    {
+      "epoch": 0.41254834550923936,
+      "grad_norm": 1.6255794763565063,
+      "learning_rate": 3.839566987447492e-06,
+      "logits/chosen": 3.842928409576416,
+      "logits/rejected": 3.5797982215881348,
+      "logps/chosen": -2.6448044776916504,
+      "logps/rejected": -4.98160982131958,
+      "loss": 0.3547,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -3.9672069549560547,
+      "rewards/margins": 3.5052082538604736,
+      "rewards/rejected": -7.472414493560791,
+      "step": 480
+    },
+    {
+      "epoch": 0.4211431027073485,
+      "grad_norm": 2.9274284839630127,
+      "learning_rate": 3.795067523432826e-06,
+      "logits/chosen": 3.3297150135040283,
+      "logits/rejected": 3.0205535888671875,
+      "logps/chosen": -2.811923027038574,
+      "logps/rejected": -6.040881156921387,
+      "loss": 0.3097,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.217884063720703,
+      "rewards/margins": 4.843437194824219,
+      "rewards/rejected": -9.061322212219238,
+      "step": 490
+    },
+    {
+      "epoch": 0.42973785990545765,
+      "grad_norm": 2.9143636226654053,
+      "learning_rate": 3.7500000000000005e-06,
+      "logits/chosen": 2.760014772415161,
+      "logits/rejected": 2.535520315170288,
+      "logps/chosen": -3.068406820297241,
+      "logps/rejected": -5.877435684204102,
+      "loss": 0.3031,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.602609634399414,
+      "rewards/margins": 4.21354341506958,
+      "rewards/rejected": -8.816153526306152,
+      "step": 500
+    },
+    {
+      "epoch": 0.42973785990545765,
+      "eval_logits/chosen": 2.0952131748199463,
+      "eval_logits/rejected": 2.1864659786224365,
+      "eval_logps/chosen": -3.392296075820923,
+      "eval_logps/rejected": -6.948195457458496,
+      "eval_loss": 0.33660775423049927,
+      "eval_rewards/accuracies": 0.9263157844543457,
+      "eval_rewards/chosen": -5.088444232940674,
+      "eval_rewards/margins": 5.3338494300842285,
+      "eval_rewards/rejected": -10.422293663024902,
+      "eval_runtime": 25.2433,
+      "eval_samples_per_second": 29.83,
+      "eval_steps_per_second": 3.763,
+      "step": 500
+    },
+    {
+      "epoch": 0.4383326171035668,
+      "grad_norm": 2.563810348510742,
+      "learning_rate": 3.7043841852542884e-06,
+      "logits/chosen": 2.950286388397217,
+      "logits/rejected": 2.619025945663452,
+      "logps/chosen": -3.237391710281372,
+      "logps/rejected": -5.953216552734375,
+      "loss": 0.318,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.856087684631348,
+      "rewards/margins": 4.073737144470215,
+      "rewards/rejected": -8.929824829101562,
+      "step": 510
+    },
+    {
+      "epoch": 0.44692737430167595,
+      "grad_norm": 2.0339434146881104,
+      "learning_rate": 3.658240087799655e-06,
+      "logits/chosen": 2.987595558166504,
+      "logits/rejected": 2.6243975162506104,
+      "logps/chosen": -3.5633530616760254,
+      "logps/rejected": -7.0458879470825195,
+      "loss": 0.3053,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.345029354095459,
+      "rewards/margins": 5.223802089691162,
+      "rewards/rejected": -10.568831443786621,
+      "step": 520
+    },
+    {
+      "epoch": 0.45552213149978515,
+      "grad_norm": 4.091029644012451,
+      "learning_rate": 3.611587947962319e-06,
+      "logits/chosen": 2.297576904296875,
+      "logits/rejected": 2.0218777656555176,
+      "logps/chosen": -3.297245502471924,
+      "logps/rejected": -6.101919651031494,
+      "loss": 0.3255,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.945868015289307,
+      "rewards/margins": 4.207010746002197,
+      "rewards/rejected": -9.152878761291504,
+      "step": 530
+    },
+    {
+      "epoch": 0.4641168886978943,
+      "grad_norm": 2.7896900177001953,
+      "learning_rate": 3.564448228912682e-06,
+      "logits/chosen": 2.103950023651123,
+      "logits/rejected": 1.9478647708892822,
+      "logps/chosen": -2.9360263347625732,
+      "logps/rejected": -6.406435489654541,
+      "loss": 0.3361,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.40403938293457,
+      "rewards/margins": 5.20561408996582,
+      "rewards/rejected": -9.60965347290039,
+      "step": 540
+    },
+    {
+      "epoch": 0.47271164589600345,
+      "grad_norm": 2.657970905303955,
+      "learning_rate": 3.516841607689501e-06,
+      "logits/chosen": 2.1658639907836914,
+      "logits/rejected": 2.214900493621826,
+      "logps/chosen": -3.084073066711426,
+      "logps/rejected": -6.935500144958496,
+      "loss": 0.2928,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.626110076904297,
+      "rewards/margins": 5.7771406173706055,
+      "rewards/rejected": -10.403249740600586,
+      "step": 550
+    },
+    {
+      "epoch": 0.47271164589600345,
+      "eval_logits/chosen": 2.285294771194458,
+      "eval_logits/rejected": 2.3312103748321533,
+      "eval_logps/chosen": -3.35794997215271,
+      "eval_logps/rejected": -7.37537145614624,
+      "eval_loss": 0.3121817409992218,
+      "eval_rewards/accuracies": 0.9263157844543457,
+      "eval_rewards/chosen": -5.036925792694092,
+      "eval_rewards/margins": 6.026132106781006,
+      "eval_rewards/rejected": -11.063057899475098,
+      "eval_runtime": 25.2442,
+      "eval_samples_per_second": 29.829,
+      "eval_steps_per_second": 3.763,
+      "step": 550
+    },
+    {
+      "epoch": 0.4813064030941126,
+      "grad_norm": 2.940019369125366,
+      "learning_rate": 3.4687889661302577e-06,
+      "logits/chosen": 1.9122416973114014,
+      "logits/rejected": 1.9943454265594482,
+      "logps/chosen": -3.27177095413208,
+      "logps/rejected": -7.023342132568359,
+      "loss": 0.3105,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.907656669616699,
+      "rewards/margins": 5.6273579597473145,
+      "rewards/rejected": -10.535014152526855,
+      "step": 560
+    },
+    {
+      "epoch": 0.48990116029222175,
+      "grad_norm": 1.8887412548065186,
+      "learning_rate": 3.4203113817116955e-06,
+      "logits/chosen": 2.274843692779541,
+      "logits/rejected": 2.392199993133545,
+      "logps/chosen": -3.383749008178711,
+      "logps/rejected": -7.265415191650391,
+      "loss": 0.3003,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.075623512268066,
+      "rewards/margins": 5.8224992752075195,
+      "rewards/rejected": -10.898123741149902,
+      "step": 570
+    },
+    {
+      "epoch": 0.4984959174903309,
+      "grad_norm": 1.6364414691925049,
+      "learning_rate": 3.3714301183045382e-06,
+      "logits/chosen": 2.423910617828369,
+      "logits/rejected": 2.244985818862915,
+      "logps/chosen": -3.0959205627441406,
+      "logps/rejected": -6.822405815124512,
+      "loss": 0.2471,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -4.643880844116211,
+      "rewards/margins": 5.58972692489624,
+      "rewards/rejected": -10.233609199523926,
+      "step": 580
+    },
+    {
+      "epoch": 0.50709067468844,
+      "grad_norm": 2.6540188789367676,
+      "learning_rate": 3.3221666168464584e-06,
+      "logits/chosen": 2.8146812915802,
+      "logits/rejected": 2.5971922874450684,
+      "logps/chosen": -4.139407157897949,
+      "logps/rejected": -7.71649694442749,
+      "loss": 0.2809,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.209111213684082,
+      "rewards/margins": 5.365634441375732,
+      "rewards/rejected": -11.574746131896973,
+      "step": 590
+    },
+    {
+      "epoch": 0.5156854318865493,
+      "grad_norm": 4.229885578155518,
+      "learning_rate": 3.272542485937369e-06,
+      "logits/chosen": 2.2735249996185303,
+      "logits/rejected": 1.8577899932861328,
+      "logps/chosen": -3.731342315673828,
+      "logps/rejected": -7.2900390625,
+      "loss": 0.2956,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.5970139503479,
+      "rewards/margins": 5.338044166564941,
+      "rewards/rejected": -10.93505859375,
+      "step": 600
+    },
+    {
+      "epoch": 0.5156854318865493,
+      "eval_logits/chosen": 2.3333992958068848,
+      "eval_logits/rejected": 2.529745578765869,
+      "eval_logps/chosen": -3.679597854614258,
+      "eval_logps/rejected": -7.917842864990234,
+      "eval_loss": 0.3030374050140381,
+      "eval_rewards/accuracies": 0.9263157844543457,
+      "eval_rewards/chosen": -5.519396781921387,
+      "eval_rewards/margins": 6.357367992401123,
+      "eval_rewards/rejected": -11.876765251159668,
+      "eval_runtime": 25.2464,
+      "eval_samples_per_second": 29.826,
+      "eval_steps_per_second": 3.763,
+      "step": 600
+    },
+    {
+      "epoch": 0.5242801890846583,
+      "grad_norm": 2.657008647918701,
+      "learning_rate": 3.222579492361179e-06,
+      "logits/chosen": 2.699007034301758,
+      "logits/rejected": 2.731860876083374,
+      "logps/chosen": -3.3311946392059326,
+      "logps/rejected": -7.005735874176025,
+      "loss": 0.2898,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.996791839599609,
+      "rewards/margins": 5.511812686920166,
+      "rewards/rejected": -10.508604049682617,
+      "step": 610
+    },
+    {
+      "epoch": 0.5328749462827675,
+      "grad_norm": 3.046638250350952,
+      "learning_rate": 3.1722995515381644e-06,
+      "logits/chosen": 2.7617671489715576,
+      "logits/rejected": 2.7338194847106934,
+      "logps/chosen": -3.336381435394287,
+      "logps/rejected": -7.058961391448975,
+      "loss": 0.2895,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.004572868347168,
+      "rewards/margins": 5.583868980407715,
+      "rewards/rejected": -10.588441848754883,
+      "step": 620
+    },
+    {
+      "epoch": 0.5414697034808766,
+      "grad_norm": 2.342069387435913,
+      "learning_rate": 3.121724717912138e-06,
+      "logits/chosen": 2.5818216800689697,
+      "logits/rejected": 1.987378716468811,
+      "logps/chosen": -3.0970518589019775,
+      "logps/rejected": -6.240235805511475,
+      "loss": 0.2634,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.645577430725098,
+      "rewards/margins": 4.714776039123535,
+      "rewards/rejected": -9.36035442352295,
+      "step": 630
+    },
+    {
+      "epoch": 0.5500644606789858,
+      "grad_norm": 1.9333513975143433,
+      "learning_rate": 3.0708771752766397e-06,
+      "logits/chosen": 2.911674737930298,
+      "logits/rejected": 2.7606472969055176,
+      "logps/chosen": -3.2809441089630127,
+      "logps/rejected": -7.210829257965088,
+      "loss": 0.2594,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.921416282653809,
+      "rewards/margins": 5.894827365875244,
+      "rewards/rejected": -10.816244125366211,
+      "step": 640
+    },
+    {
+      "epoch": 0.5586592178770949,
+      "grad_norm": 5.659445285797119,
+      "learning_rate": 3.019779227044398e-06,
+      "logits/chosen": 2.4733409881591797,
+      "logits/rejected": 2.102668285369873,
+      "logps/chosen": -3.4448726177215576,
+      "logps/rejected": -7.304962158203125,
+      "loss": 0.2399,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.167309284210205,
+      "rewards/margins": 5.790134429931641,
+      "rewards/rejected": -10.957443237304688,
+      "step": 650
+    },
+    {
+      "epoch": 0.5586592178770949,
+      "eval_logits/chosen": 2.482032537460327,
+      "eval_logits/rejected": 2.66147780418396,
+      "eval_logps/chosen": -3.728013515472412,
+      "eval_logps/rejected": -8.231985092163086,
+      "eval_loss": 0.2814938426017761,
+      "eval_rewards/accuracies": 0.9263157844543457,
+      "eval_rewards/chosen": -5.592020511627197,
+      "eval_rewards/margins": 6.75595760345459,
+      "eval_rewards/rejected": -12.347977638244629,
+      "eval_runtime": 25.2423,
+      "eval_samples_per_second": 29.831,
+      "eval_steps_per_second": 3.764,
+      "step": 650
+    },
+    {
+      "epoch": 0.5672539750752041,
+      "grad_norm": 2.189638137817383,
+      "learning_rate": 2.9684532864643123e-06,
+      "logits/chosen": 2.875077962875366,
+      "logits/rejected": 2.712646484375,
+      "logps/chosen": -3.757338762283325,
+      "logps/rejected": -6.6974897384643555,
+      "loss": 0.2759,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.636007785797119,
+      "rewards/margins": 4.410226821899414,
+      "rewards/rejected": -10.046236038208008,
+      "step": 660
+    },
+    {
+      "epoch": 0.5758487322733132,
+      "grad_norm": 3.5755774974823,
+      "learning_rate": 2.9169218667902562e-06,
+      "logits/chosen": 2.9562981128692627,
+      "logits/rejected": 2.7660539150238037,
+      "logps/chosen": -3.2358715534210205,
+      "logps/rejected": -6.90399169921875,
+      "loss": 0.2586,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.853806972503662,
+      "rewards/margins": 5.502181053161621,
+      "rewards/rejected": -10.355987548828125,
+      "step": 670
+    },
+    {
+      "epoch": 0.5844434894714224,
+      "grad_norm": 2.5616958141326904,
+      "learning_rate": 2.8652075714060296e-06,
+      "logits/chosen": 2.5067126750946045,
+      "logits/rejected": 2.3888354301452637,
+      "logps/chosen": -3.462563991546631,
+      "logps/rejected": -6.964964866638184,
+      "loss": 0.251,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.193846225738525,
+      "rewards/margins": 5.253602027893066,
+      "rewards/rejected": -10.447446823120117,
+      "step": 680
+    },
+    {
+      "epoch": 0.5930382466695315,
+      "grad_norm": 2.964050531387329,
+      "learning_rate": 2.813333083910761e-06,
+      "logits/chosen": 2.659935474395752,
+      "logits/rejected": 2.6573758125305176,
+      "logps/chosen": -3.9107768535614014,
+      "logps/rejected": -7.865903377532959,
+      "loss": 0.2294,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.866166114807129,
+      "rewards/margins": 5.9326887130737305,
+      "rewards/rejected": -11.79885482788086,
+      "step": 690
+    },
+    {
+      "epoch": 0.6016330038676407,
+      "grad_norm": 4.389697551727295,
+      "learning_rate": 2.761321158169134e-06,
+      "logits/chosen": 2.217245578765869,
+      "logits/rejected": 2.421597957611084,
+      "logps/chosen": -4.029661655426025,
+      "logps/rejected": -8.073125839233398,
+      "loss": 0.2469,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.044493675231934,
+      "rewards/margins": 6.065195083618164,
+      "rewards/rejected": -12.109688758850098,
+      "step": 700
+    },
+    {
+      "epoch": 0.6016330038676407,
+      "eval_logits/chosen": 2.0770955085754395,
+      "eval_logits/rejected": 2.3815462589263916,
+      "eval_logps/chosen": -3.924149751663208,
+      "eval_logps/rejected": -8.844257354736328,
+      "eval_loss": 0.2584603726863861,
+      "eval_rewards/accuracies": 0.9263157844543457,
+      "eval_rewards/chosen": -5.886224746704102,
+      "eval_rewards/margins": 7.380159854888916,
+      "eval_rewards/rejected": -13.26638412475586,
+      "eval_runtime": 25.243,
+      "eval_samples_per_second": 29.83,
+      "eval_steps_per_second": 3.763,
+      "step": 700
+    },
+    {
+      "epoch": 0.6102277610657499,
+      "grad_norm": 3.290154457092285,
+      "learning_rate": 2.70919460833079e-06,
+      "logits/chosen": 2.458578586578369,
+      "logits/rejected": 2.275515079498291,
+      "logps/chosen": -3.2734694480895996,
+      "logps/rejected": -7.873226165771484,
+      "loss": 0.2732,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.91020393371582,
+      "rewards/margins": 6.899635314941406,
+      "rewards/rejected": -11.809839248657227,
+      "step": 710
+    },
+    {
+      "epoch": 0.618822518263859,
+      "grad_norm": 2.2760908603668213,
+      "learning_rate": 2.6569762988232838e-06,
+      "logits/chosen": 2.6856372356414795,
+      "logits/rejected": 2.722838878631592,
+      "logps/chosen": -3.589418411254883,
+      "logps/rejected": -7.638446807861328,
+      "loss": 0.2583,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.384127140045166,
+      "rewards/margins": 6.073542594909668,
+      "rewards/rejected": -11.457670211791992,
+      "step": 720
+    },
+    {
+      "epoch": 0.6274172754619682,
+      "grad_norm": 6.937672138214111,
+      "learning_rate": 2.604689134322999e-06,
+      "logits/chosen": 2.928969383239746,
+      "logits/rejected": 2.5493836402893066,
+      "logps/chosen": -3.3862743377685547,
+      "logps/rejected": -7.568005561828613,
+      "loss": 0.2889,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.07941198348999,
+      "rewards/margins": 6.27259635925293,
+      "rewards/rejected": -11.352007865905762,
+      "step": 730
+    },
+    {
+      "epoch": 0.6360120326600773,
+      "grad_norm": 2.1878838539123535,
+      "learning_rate": 2.5523560497083927e-06,
+      "logits/chosen": 2.3824827671051025,
+      "logits/rejected": 2.257145404815674,
+      "logps/chosen": -3.5448341369628906,
+      "logps/rejected": -7.594444274902344,
+      "loss": 0.1972,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.3172502517700195,
+      "rewards/margins": 6.074415683746338,
+      "rewards/rejected": -11.3916654586792,
+      "step": 740
+    },
+    {
+      "epoch": 0.6446067898581865,
+      "grad_norm": 4.405832767486572,
+      "learning_rate": 2.5e-06,
+      "logits/chosen": 3.204157590866089,
+      "logits/rejected": 3.0262837409973145,
+      "logps/chosen": -3.67409086227417,
+      "logps/rejected": -8.078901290893555,
+      "loss": 0.2282,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.511136531829834,
+      "rewards/margins": 6.607214450836182,
+      "rewards/rejected": -12.118351936340332,
+      "step": 750
+    },
+    {
+      "epoch": 0.6446067898581865,
+      "eval_logits/chosen": 2.1246254444122314,
+      "eval_logits/rejected": 2.4088852405548096,
+      "eval_logps/chosen": -4.221064567565918,
+      "eval_logps/rejected": -9.4141206741333,
+      "eval_loss": 0.2537557780742645,
+      "eval_rewards/accuracies": 0.9368420839309692,
+      "eval_rewards/chosen": -6.331596374511719,
+      "eval_rewards/margins": 7.789584159851074,
+      "eval_rewards/rejected": -14.121179580688477,
+      "eval_runtime": 25.2436,
+      "eval_samples_per_second": 29.829,
+      "eval_steps_per_second": 3.763,
+      "step": 750
+    },
+    {
+      "epoch": 0.6532015470562956,
+      "grad_norm": 2.8693907260894775,
+      "learning_rate": 2.447643950291608e-06,
+      "logits/chosen": 2.5033986568450928,
+      "logits/rejected": 2.2746779918670654,
+      "logps/chosen": -4.256644248962402,
+      "logps/rejected": -8.564817428588867,
+      "loss": 0.2337,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.3849663734436035,
+      "rewards/margins": 6.462259769439697,
+      "rewards/rejected": -12.8472261428833,
+      "step": 760
+    },
+    {
+      "epoch": 0.6617963042544048,
+      "grad_norm": 4.912906646728516,
+      "learning_rate": 2.3953108656770018e-06,
+      "logits/chosen": 2.861431837081909,
+      "logits/rejected": 2.974611759185791,
+      "logps/chosen": -3.9564735889434814,
+      "logps/rejected": -7.863286018371582,
+      "loss": 0.2585,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.934710502624512,
+      "rewards/margins": 5.860217571258545,
+      "rewards/rejected": -11.794927597045898,
+      "step": 770
+    },
+    {
+      "epoch": 0.6703910614525139,
+      "grad_norm": 3.215716600418091,
+      "learning_rate": 2.3430237011767166e-06,
+      "logits/chosen": 1.9008615016937256,
+      "logits/rejected": 1.9049352407455444,
+      "logps/chosen": -4.304060935974121,
+      "logps/rejected": -8.806629180908203,
+      "loss": 0.2279,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.456091403961182,
+      "rewards/margins": 6.753852844238281,
+      "rewards/rejected": -13.209943771362305,
+      "step": 780
+    },
+    {
+      "epoch": 0.6789858186506231,
+      "grad_norm": 3.8724021911621094,
+      "learning_rate": 2.290805391669212e-06,
+      "logits/chosen": 2.2521636486053467,
+      "logits/rejected": 2.2159788608551025,
+      "logps/chosen": -4.012774467468262,
+      "logps/rejected": -8.53366470336914,
+      "loss": 0.2437,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.019161701202393,
+      "rewards/margins": 6.78133487701416,
+      "rewards/rejected": -12.800497055053711,
+      "step": 790
+    },
+    {
+      "epoch": 0.6875805758487322,
+      "grad_norm": 3.56345796585083,
+      "learning_rate": 2.238678841830867e-06,
+      "logits/chosen": 2.0579304695129395,
+      "logits/rejected": 2.304316997528076,
+      "logps/chosen": -3.590430736541748,
+      "logps/rejected": -8.182169914245605,
+      "loss": 0.213,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.385646820068359,
+      "rewards/margins": 6.887608528137207,
+      "rewards/rejected": -12.27325439453125,
+      "step": 800
+    },
+    {
+      "epoch": 0.6875805758487322,
+      "eval_logits/chosen": 2.228646755218506,
+      "eval_logits/rejected": 2.444817543029785,
+      "eval_logps/chosen": -3.8403449058532715,
+      "eval_logps/rejected": -9.179658889770508,
+      "eval_loss": 0.23895224928855896,
+      "eval_rewards/accuracies": 0.9368420839309692,
+      "eval_rewards/chosen": -5.76051664352417,
+      "eval_rewards/margins": 8.00897216796875,
+      "eval_rewards/rejected": -13.769490242004395,
+      "eval_runtime": 25.2458,
+      "eval_samples_per_second": 29.827,
+      "eval_steps_per_second": 3.763,
+      "step": 800
+    },
+    {
+      "epoch": 0.6961753330468414,
+      "grad_norm": 3.4880526065826416,
+      "learning_rate": 2.186666916089239e-06,
+      "logits/chosen": 1.7993383407592773,
+      "logits/rejected": 1.754417061805725,
+      "logps/chosen": -4.045234680175781,
+      "logps/rejected": -8.927519798278809,
+      "loss": 0.2391,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.067852020263672,
+      "rewards/margins": 7.323427677154541,
+      "rewards/rejected": -13.391279220581055,
+      "step": 810
+    },
+    {
+      "epoch": 0.7047700902449506,
+      "grad_norm": 3.56809139251709,
+      "learning_rate": 2.134792428593971e-06,
+      "logits/chosen": 2.9591994285583496,
+      "logits/rejected": 2.960444211959839,
+      "logps/chosen": -4.150156497955322,
+      "logps/rejected": -8.512441635131836,
+      "loss": 0.1972,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -6.225234031677246,
+      "rewards/margins": 6.54342794418335,
+      "rewards/rejected": -12.768662452697754,
+      "step": 820
+    },
+    {
+      "epoch": 0.7133648474430597,
+      "grad_norm": 4.127833843231201,
+      "learning_rate": 2.0830781332097446e-06,
+      "logits/chosen": 3.008269786834717,
+      "logits/rejected": 2.63409686088562,
+      "logps/chosen": -3.8291163444519043,
+      "logps/rejected": -8.657347679138184,
+      "loss": 0.2161,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.743674278259277,
+      "rewards/margins": 7.24234676361084,
+      "rewards/rejected": -12.986021041870117,
+      "step": 830
+    },
+    {
+      "epoch": 0.7219596046411689,
+      "grad_norm": 4.475767612457275,
+      "learning_rate": 2.031546713535688e-06,
+      "logits/chosen": 2.7164976596832275,
+      "logits/rejected": 2.5976195335388184,
+      "logps/chosen": -4.153134346008301,
+      "logps/rejected": -8.893486022949219,
+      "loss": 0.1895,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.229701519012451,
+      "rewards/margins": 7.110527992248535,
+      "rewards/rejected": -13.340228080749512,
+      "step": 840
+    },
+    {
+      "epoch": 0.730554361839278,
+      "grad_norm": 4.190205097198486,
+      "learning_rate": 1.9802207729556023e-06,
+      "logits/chosen": 2.6235451698303223,
+      "logits/rejected": 2.5486202239990234,
+      "logps/chosen": -3.899543046951294,
+      "logps/rejected": -8.277327537536621,
+      "loss": 0.2239,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.8493146896362305,
+      "rewards/margins": 6.566677093505859,
+      "rewards/rejected": -12.415990829467773,
+      "step": 850
+    },
+    {
+      "epoch": 0.730554361839278,
+      "eval_logits/chosen": 2.173233985900879,
+      "eval_logits/rejected": 2.433162212371826,
+      "eval_logps/chosen": -4.13487434387207,
+      "eval_logps/rejected": -9.577596664428711,
+      "eval_loss": 0.23591776192188263,
+      "eval_rewards/accuracies": 0.9473684430122375,
+      "eval_rewards/chosen": -6.2023115158081055,
+      "eval_rewards/margins": 8.164085388183594,
+      "eval_rewards/rejected": -14.3663969039917,
+      "eval_runtime": 25.2441,
+      "eval_samples_per_second": 29.829,
+      "eval_steps_per_second": 3.763,
+      "step": 850
+    },
+    {
+      "epoch": 0.7391491190373872,
+      "grad_norm": 2.6548664569854736,
+      "learning_rate": 1.9291228247233607e-06,
+      "logits/chosen": 1.7737414836883545,
+      "logits/rejected": 2.080662965774536,
+      "logps/chosen": -3.9447720050811768,
+      "logps/rejected": -9.01865005493164,
+      "loss": 0.2268,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.9171576499938965,
+      "rewards/margins": 7.610815525054932,
+      "rewards/rejected": -13.527974128723145,
+      "step": 860
+    },
+    {
+      "epoch": 0.7477438762354963,
+      "grad_norm": 2.5912184715270996,
+      "learning_rate": 1.8782752820878636e-06,
+      "logits/chosen": 2.5428760051727295,
+      "logits/rejected": 2.3569278717041016,
+      "logps/chosen": -3.685049057006836,
+      "logps/rejected": -9.194517135620117,
+      "loss": 0.2001,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.527573585510254,
+      "rewards/margins": 8.264203071594238,
+      "rewards/rejected": -13.791775703430176,
+      "step": 870
+    },
+    {
+      "epoch": 0.7563386334336055,
+      "grad_norm": 3.789594888687134,
+      "learning_rate": 1.827700448461836e-06,
+      "logits/chosen": 3.139338970184326,
+      "logits/rejected": 3.003114700317383,
+      "logps/chosen": -4.347461700439453,
+      "logps/rejected": -8.560078620910645,
+      "loss": 0.2257,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.521193027496338,
+      "rewards/margins": 6.318924903869629,
+      "rewards/rejected": -12.840118408203125,
+      "step": 880
+    },
+    {
+      "epoch": 0.7649333906317146,
+      "grad_norm": 2.3799326419830322,
+      "learning_rate": 1.7774205076388207e-06,
+      "logits/chosen": 3.2622504234313965,
+      "logits/rejected": 2.922945261001587,
+      "logps/chosen": -4.306991100311279,
+      "logps/rejected": -8.622769355773926,
+      "loss": 0.2123,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.46048641204834,
+      "rewards/margins": 6.473666191101074,
+      "rewards/rejected": -12.93415355682373,
+      "step": 890
+    },
+    {
+      "epoch": 0.7735281478298238,
+      "grad_norm": 3.4133400917053223,
+      "learning_rate": 1.7274575140626318e-06,
+      "logits/chosen": 2.8558292388916016,
+      "logits/rejected": 2.919982433319092,
+      "logps/chosen": -3.791405200958252,
+      "logps/rejected": -9.348276138305664,
+      "loss": 0.2345,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.687107086181641,
+      "rewards/margins": 8.335307121276855,
+      "rewards/rejected": -14.022415161132812,
+      "step": 900
+    },
+    {
+      "epoch": 0.7735281478298238,
+      "eval_logits/chosen": 2.2851152420043945,
+      "eval_logits/rejected": 2.5511629581451416,
+      "eval_logps/chosen": -4.023584842681885,
+      "eval_logps/rejected": -9.625852584838867,
+      "eval_loss": 0.23031750321388245,
+      "eval_rewards/accuracies": 0.9473684430122375,
+      "eval_rewards/chosen": -6.035377025604248,
+      "eval_rewards/margins": 8.403401374816895,
+      "eval_rewards/rejected": -14.4387788772583,
+      "eval_runtime": 25.2442,
+      "eval_samples_per_second": 29.829,
+      "eval_steps_per_second": 3.763,
+      "step": 900
+    },
+    {
+      "epoch": 0.7821229050279329,
+      "grad_norm": 2.178900957107544,
+      "learning_rate": 1.677833383153542e-06,
+      "logits/chosen": 2.3001868724823,
+      "logits/rejected": 2.365304470062256,
+      "logps/chosen": -3.690169095993042,
+      "logps/rejected": -8.727324485778809,
+      "loss": 0.1988,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.535253047943115,
+      "rewards/margins": 7.555734157562256,
+      "rewards/rejected": -13.090988159179688,
+      "step": 910
+    },
+    {
+      "epoch": 0.7907176622260421,
+      "grad_norm": 4.60929536819458,
+      "learning_rate": 1.6285698816954626e-06,
+      "logits/chosen": 3.103785276412964,
+      "logits/rejected": 3.0096678733825684,
+      "logps/chosen": -4.184874534606934,
+      "logps/rejected": -8.704519271850586,
+      "loss": 0.2128,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.277312278747559,
+      "rewards/margins": 6.7794671058654785,
+      "rewards/rejected": -13.056779861450195,
+      "step": 920
+    },
+    {
+      "epoch": 0.7993124194241513,
+      "grad_norm": 1.1031241416931152,
+      "learning_rate": 1.5796886182883053e-06,
+      "logits/chosen": 3.2616991996765137,
+      "logits/rejected": 2.990100622177124,
+      "logps/chosen": -4.041825771331787,
+      "logps/rejected": -9.399754524230957,
+      "loss": 0.2131,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.062739372253418,
+      "rewards/margins": 8.03689193725586,
+      "rewards/rejected": -14.099630355834961,
+      "step": 930
+    },
+    {
+      "epoch": 0.8079071766222604,
+      "grad_norm": 1.8013640642166138,
+      "learning_rate": 1.5312110338697427e-06,
+      "logits/chosen": 2.2281856536865234,
+      "logits/rejected": 2.1705000400543213,
+      "logps/chosen": -3.802743434906006,
+      "logps/rejected": -8.745875358581543,
+      "loss": 0.2211,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.704115867614746,
+      "rewards/margins": 7.414697170257568,
+      "rewards/rejected": -13.118814468383789,
+      "step": 940
+    },
+    {
+      "epoch": 0.8165019338203696,
+      "grad_norm": 5.369480609893799,
+      "learning_rate": 1.4831583923105e-06,
+      "logits/chosen": 2.0146822929382324,
+      "logits/rejected": 2.0050222873687744,
+      "logps/chosen": -4.013974189758301,
+      "logps/rejected": -9.14311408996582,
+      "loss": 0.2359,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.020960807800293,
+      "rewards/margins": 7.693708896636963,
+      "rewards/rejected": -13.71467113494873,
+      "step": 950
+    },
+    {
+      "epoch": 0.8165019338203696,
+      "eval_logits/chosen": 2.4391298294067383,
+      "eval_logits/rejected": 2.693408250808716,
+      "eval_logps/chosen": -3.877185106277466,
+      "eval_logps/rejected": -9.608528137207031,
+      "eval_loss": 0.2290637195110321,
+      "eval_rewards/accuracies": 0.9473684430122375,
+      "eval_rewards/chosen": -5.81577730178833,
+      "eval_rewards/margins": 8.597016334533691,
+      "eval_rewards/rejected": -14.41279411315918,
+      "eval_runtime": 25.2429,
+      "eval_samples_per_second": 29.83,
+      "eval_steps_per_second": 3.763,
+      "step": 950
+    }
+  ],
+  "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": 2.184331746537898e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-950/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2724321929c8f23b3450898606159e8b53081fd4c6bdcde542d401d379befa1
+size 7224

checkpoint-950/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)