Training in progress, step 750, checkpoint

checkpoint-750/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-750/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": [
+    "down_proj",
+    "qkv_proj",
+    "gate_up_proj",
+    "o_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-750/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:633361ccdfdd139aa9f945143ab1df72e94551665506ce7bd77f575842b21b87
+size 25200088

checkpoint-750/global_step750/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bc74232bd3cd2131ce636b0359bb23a28f33fa9aaed9d217bf805d1722da061
+size 18881328

checkpoint-750/global_step750/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64496283c8af94a796b2f6cc618470fa5b90ddd0a7f7f566ba364b682a7fe044
+size 18881328

checkpoint-750/global_step750/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96827a5c46c8ecb8e87d9ba51b083dab6f0b0f5b24a2b4dc34e5738856d3e334
+size 18881328

checkpoint-750/global_step750/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:d98f07598e0c9847a1826dc3c7c63f8079ceb1072aa91882f71744f36aad484f
+size 18881392

checkpoint-750/global_step750/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:dd6acad15198c4c679af14819ddc21b14bb3c57c450899a1a3788258ec75ce26
+size 18881392

checkpoint-750/global_step750/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:d91c3a1036352d5212c22958272a8e2577a1fc06ad5ac370554bf3c256314194
+size 18881392

checkpoint-750/global_step750/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23534af77aff44e2bfdfcaeff1c4e51c7e78f8deae6e4aec096626e00744f00b
+size 18881392

checkpoint-750/global_step750/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:f8650496709fa9fa982538529dcc2c95a176414e5602d76086f60f6e82f8de43
+size 18881392

checkpoint-750/global_step750/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f90f23a5b27dfe645b16988ff0567af385ed05375db26793bbd21034e297fd9
+size 25379244

checkpoint-750/latest

@@ -0,0 +1 @@
+global_step750

checkpoint-750/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39d0a42a76c6856b42516358f397705cff8f5ae2210de23f6abc8fc7d370ce43
+size 15984

checkpoint-750/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c79d39762bf88c59ea58ab8c192f4d9721ab6eba78debc69a369654d4199af50
+size 15984

checkpoint-750/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1426a2458db12639f98377335a1109abade08a448981fb41c315ef1f9fd4191e
+size 15984

checkpoint-750/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e598407075fbe89ea2094160f92052415ad3b0b80d125438201859b9875d537b
+size 15984

checkpoint-750/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:437e8f540f343ec9a874078e652ba30d02c2d12e3039d8092e96942ade74967b
+size 15984

checkpoint-750/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23cfb7770b02ea62450b4853818ad587c09df566939a04e941091111cc9b7cf2
+size 15984

checkpoint-750/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54fe581e399def26af9ab0920fdea64c37d1eed5ad5a4b3fec55e45525aba99f
+size 15984

checkpoint-750/rng_state_7.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:491793f3baa6e0a6171458158bd6f4cce55a8696d0c0e279c19b74fbf532973f
+size 15984

checkpoint-750/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8befafa70cc542e0acab7097117df36aab3038e24f45897d4d12cf6f78e0085f
+size 1064

checkpoint-750/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-750/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-750/trainer_state.json

@@ -0,0 +1,1398 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.6660746003552398,
+  "eval_steps": 50,
+  "global_step": 750,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008880994671403197,
+      "grad_norm": 0.04571289196610451,
+      "learning_rate": 4.999451708687114e-06,
+      "logits/chosen": 14.56671142578125,
+      "logits/rejected": 15.112574577331543,
+      "logps/chosen": -0.26506316661834717,
+      "logps/rejected": -0.3439488410949707,
+      "loss": 0.9267,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.39759472012519836,
+      "rewards/margins": 0.11832849681377411,
+      "rewards/rejected": -0.5159232020378113,
+      "step": 10
+    },
+    {
+      "epoch": 0.017761989342806393,
+      "grad_norm": 0.0512714721262455,
+      "learning_rate": 4.997807075247147e-06,
+      "logits/chosen": 14.376543045043945,
+      "logits/rejected": 14.862703323364258,
+      "logps/chosen": -0.2708089351654053,
+      "logps/rejected": -0.32412824034690857,
+      "loss": 0.936,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.4062133729457855,
+      "rewards/margins": 0.07997899502515793,
+      "rewards/rejected": -0.4861923158168793,
+      "step": 20
+    },
+    {
+      "epoch": 0.02664298401420959,
+      "grad_norm": 0.058383647352457047,
+      "learning_rate": 4.9950668210706795e-06,
+      "logits/chosen": 14.208717346191406,
+      "logits/rejected": 15.370651245117188,
+      "logps/chosen": -0.28206294775009155,
+      "logps/rejected": -0.38387423753738403,
+      "loss": 0.9215,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.42309442162513733,
+      "rewards/margins": 0.15271687507629395,
+      "rewards/rejected": -0.5758112668991089,
+      "step": 30
+    },
+    {
+      "epoch": 0.035523978685612786,
+      "grad_norm": 0.06262075155973434,
+      "learning_rate": 4.9912321481237616e-06,
+      "logits/chosen": 14.768765449523926,
+      "logits/rejected": 15.169331550598145,
+      "logps/chosen": -0.27857059240341187,
+      "logps/rejected": -0.3388269543647766,
+      "loss": 0.9386,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.4178559184074402,
+      "rewards/margins": 0.09038447588682175,
+      "rewards/rejected": -0.5082404017448425,
+      "step": 40
+    },
+    {
+      "epoch": 0.04440497335701599,
+      "grad_norm": 0.06259036809206009,
+      "learning_rate": 4.986304738420684e-06,
+      "logits/chosen": 14.950456619262695,
+      "logits/rejected": 15.232122421264648,
+      "logps/chosen": -0.2961367070674896,
+      "logps/rejected": -0.3322262465953827,
+      "loss": 0.9317,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.44420504570007324,
+      "rewards/margins": 0.054134320467710495,
+      "rewards/rejected": -0.4983394145965576,
+      "step": 50
+    },
+    {
+      "epoch": 0.04440497335701599,
+      "eval_logits/chosen": 14.56529426574707,
+      "eval_logits/rejected": 14.895020484924316,
+      "eval_logps/chosen": -0.2806546986103058,
+      "eval_logps/rejected": -0.3486972451210022,
+      "eval_loss": 0.9381324052810669,
+      "eval_rewards/accuracies": 0.5274725556373596,
+      "eval_rewards/chosen": -0.4209820330142975,
+      "eval_rewards/margins": 0.10206379741430283,
+      "eval_rewards/rejected": -0.5230458974838257,
+      "eval_runtime": 25.2574,
+      "eval_samples_per_second": 28.823,
+      "eval_steps_per_second": 3.603,
+      "step": 50
+    },
+    {
+      "epoch": 0.05328596802841918,
+      "grad_norm": 0.07301533967256546,
+      "learning_rate": 4.980286753286196e-06,
+      "logits/chosen": 14.195574760437012,
+      "logits/rejected": 15.173194885253906,
+      "logps/chosen": -0.2693648636341095,
+      "logps/rejected": -0.33997970819473267,
+      "loss": 0.9319,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.40404725074768066,
+      "rewards/margins": 0.10592226684093475,
+      "rewards/rejected": -0.5099694728851318,
+      "step": 60
+    },
+    {
+      "epoch": 0.06216696269982238,
+      "grad_norm": 0.0659889206290245,
+      "learning_rate": 4.973180832407471e-06,
+      "logits/chosen": 14.910173416137695,
+      "logits/rejected": 15.361429214477539,
+      "logps/chosen": -0.28456225991249084,
+      "logps/rejected": -0.3702812194824219,
+      "loss": 0.9185,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.42684346437454224,
+      "rewards/margins": 0.12857840955257416,
+      "rewards/rejected": -0.5554218292236328,
+      "step": 70
+    },
+    {
+      "epoch": 0.07104795737122557,
+      "grad_norm": 0.05815625935792923,
+      "learning_rate": 4.964990092676263e-06,
+      "logits/chosen": 14.407182693481445,
+      "logits/rejected": 14.948204040527344,
+      "logps/chosen": -0.292889267206192,
+      "logps/rejected": -0.3381648063659668,
+      "loss": 0.9388,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.43933385610580444,
+      "rewards/margins": 0.06791339069604874,
+      "rewards/rejected": -0.5072472095489502,
+      "step": 80
+    },
+    {
+      "epoch": 0.07992895204262877,
+      "grad_norm": 0.06627190113067627,
+      "learning_rate": 4.9557181268217225e-06,
+      "logits/chosen": 14.622471809387207,
+      "logits/rejected": 15.167770385742188,
+      "logps/chosen": -0.28155821561813354,
+      "logps/rejected": -0.33633899688720703,
+      "loss": 0.9256,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.4223373532295227,
+      "rewards/margins": 0.08217118680477142,
+      "rewards/rejected": -0.5045084953308105,
+      "step": 90
+    },
+    {
+      "epoch": 0.08880994671403197,
+      "grad_norm": 0.0724545568227768,
+      "learning_rate": 4.9453690018345144e-06,
+      "logits/chosen": 14.289724349975586,
+      "logits/rejected": 14.882037162780762,
+      "logps/chosen": -0.2791440486907959,
+      "logps/rejected": -0.35329627990722656,
+      "loss": 0.9374,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.41871610283851624,
+      "rewards/margins": 0.11122839152812958,
+      "rewards/rejected": -0.5299445390701294,
+      "step": 100
+    },
+    {
+      "epoch": 0.08880994671403197,
+      "eval_logits/chosen": 14.337930679321289,
+      "eval_logits/rejected": 14.689269065856934,
+      "eval_logps/chosen": -0.2726942300796509,
+      "eval_logps/rejected": -0.34668418765068054,
+      "eval_loss": 0.9302808046340942,
+      "eval_rewards/accuracies": 0.5384615659713745,
+      "eval_rewards/chosen": -0.40904131531715393,
+      "eval_rewards/margins": 0.11098497360944748,
+      "eval_rewards/rejected": -0.5200263261795044,
+      "eval_runtime": 25.2585,
+      "eval_samples_per_second": 28.822,
+      "eval_steps_per_second": 3.603,
+      "step": 100
+    },
+    {
+      "epoch": 0.09769094138543517,
+      "grad_norm": 0.08156246691942215,
+      "learning_rate": 4.933947257182901e-06,
+      "logits/chosen": 14.499124526977539,
+      "logits/rejected": 14.916313171386719,
+      "logps/chosen": -0.2798352837562561,
+      "logps/rejected": -0.3477734327316284,
+      "loss": 0.9243,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.4197530150413513,
+      "rewards/margins": 0.10190720856189728,
+      "rewards/rejected": -0.5216602087020874,
+      "step": 110
+    },
+    {
+      "epoch": 0.10657193605683836,
+      "grad_norm": 0.08161844313144684,
+      "learning_rate": 4.921457902821578e-06,
+      "logits/chosen": 13.595013618469238,
+      "logits/rejected": 14.390353202819824,
+      "logps/chosen": -0.26682502031326294,
+      "logps/rejected": -0.3336995542049408,
+      "loss": 0.9123,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.400237500667572,
+      "rewards/margins": 0.10031183809041977,
+      "rewards/rejected": -0.5005493760108948,
+      "step": 120
+    },
+    {
+      "epoch": 0.11545293072824156,
+      "grad_norm": 0.28624778985977173,
+      "learning_rate": 4.907906416994146e-06,
+      "logits/chosen": 13.711044311523438,
+      "logits/rejected": 14.558542251586914,
+      "logps/chosen": -0.27874043583869934,
+      "logps/rejected": -0.3582325279712677,
+      "loss": 0.9163,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.41811060905456543,
+      "rewards/margins": 0.11923813819885254,
+      "rewards/rejected": -0.537348747253418,
+      "step": 130
+    },
+    {
+      "epoch": 0.12433392539964476,
+      "grad_norm": 0.10971464216709137,
+      "learning_rate": 4.893298743830168e-06,
+      "logits/chosen": 14.18798828125,
+      "logits/rejected": 14.993026733398438,
+      "logps/chosen": -0.2750400900840759,
+      "logps/rejected": -0.39451608061790466,
+      "loss": 0.9098,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.4125601351261139,
+      "rewards/margins": 0.17921395599842072,
+      "rewards/rejected": -0.5917741060256958,
+      "step": 140
+    },
+    {
+      "epoch": 0.13321492007104796,
+      "grad_norm": 0.09321591258049011,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": 12.775139808654785,
+      "logits/rejected": 13.751996994018555,
+      "logps/chosen": -0.28446996212005615,
+      "logps/rejected": -0.36404967308044434,
+      "loss": 0.9104,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.42670494318008423,
+      "rewards/margins": 0.11936960369348526,
+      "rewards/rejected": -0.5460745096206665,
+      "step": 150
+    },
+    {
+      "epoch": 0.13321492007104796,
+      "eval_logits/chosen": 12.97266960144043,
+      "eval_logits/rejected": 13.47339916229248,
+      "eval_logps/chosen": -0.27297571301460266,
+      "eval_logps/rejected": -0.36854612827301025,
+      "eval_loss": 0.9143257737159729,
+      "eval_rewards/accuracies": 0.5824176073074341,
+      "eval_rewards/chosen": -0.4094635546207428,
+      "eval_rewards/margins": 0.14335563778877258,
+      "eval_rewards/rejected": -0.5528191924095154,
+      "eval_runtime": 25.2406,
+      "eval_samples_per_second": 28.842,
+      "eval_steps_per_second": 3.605,
+      "step": 150
+    },
+    {
+      "epoch": 0.14209591474245115,
+      "grad_norm": 0.11029861867427826,
+      "learning_rate": 4.860940925593703e-06,
+      "logits/chosen": 12.677947998046875,
+      "logits/rejected": 13.396716117858887,
+      "logps/chosen": -0.2631794512271881,
+      "logps/rejected": -0.37102141976356506,
+      "loss": 0.9051,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.3947691321372986,
+      "rewards/margins": 0.161762997508049,
+      "rewards/rejected": -0.5565321445465088,
+      "step": 160
+    },
+    {
+      "epoch": 0.15097690941385436,
+      "grad_norm": 0.15728294849395752,
+      "learning_rate": 4.84320497372973e-06,
+      "logits/chosen": 12.620219230651855,
+      "logits/rejected": 13.189640998840332,
+      "logps/chosen": -0.2947639524936676,
+      "logps/rejected": -0.3843482732772827,
+      "loss": 0.8906,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.442145973443985,
+      "rewards/margins": 0.13437646627426147,
+      "rewards/rejected": -0.5765224099159241,
+      "step": 170
+    },
+    {
+      "epoch": 0.15985790408525755,
+      "grad_norm": 0.31504154205322266,
+      "learning_rate": 4.824441214720629e-06,
+      "logits/chosen": 11.487619400024414,
+      "logits/rejected": 12.33470344543457,
+      "logps/chosen": -0.271095871925354,
+      "logps/rejected": -0.4252637028694153,
+      "loss": 0.8766,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.406643807888031,
+      "rewards/margins": 0.2312517911195755,
+      "rewards/rejected": -0.6378955245018005,
+      "step": 180
+    },
+    {
+      "epoch": 0.16873889875666073,
+      "grad_norm": 0.19222252070903778,
+      "learning_rate": 4.804657878971252e-06,
+      "logits/chosen": 10.093737602233887,
+      "logits/rejected": 10.851752281188965,
+      "logps/chosen": -0.2679918110370636,
+      "logps/rejected": -0.437336266040802,
+      "loss": 0.884,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.4019877314567566,
+      "rewards/margins": 0.2540166974067688,
+      "rewards/rejected": -0.6560044288635254,
+      "step": 190
+    },
+    {
+      "epoch": 0.17761989342806395,
+      "grad_norm": 0.2275688648223877,
+      "learning_rate": 4.783863644106502e-06,
+      "logits/chosen": 9.483477592468262,
+      "logits/rejected": 10.106366157531738,
+      "logps/chosen": -0.2957404553890228,
+      "logps/rejected": -0.40739065408706665,
+      "loss": 0.8767,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.4436107575893402,
+      "rewards/margins": 0.16747523844242096,
+      "rewards/rejected": -0.6110859513282776,
+      "step": 200
+    },
+    {
+      "epoch": 0.17761989342806395,
+      "eval_logits/chosen": 8.491498947143555,
+      "eval_logits/rejected": 8.999146461486816,
+      "eval_logps/chosen": -0.3135836124420166,
+      "eval_logps/rejected": -0.4829566180706024,
+      "eval_loss": 0.8664290904998779,
+      "eval_rewards/accuracies": 0.6263736486434937,
+      "eval_rewards/chosen": -0.4703753888607025,
+      "eval_rewards/margins": 0.2540595233440399,
+      "eval_rewards/rejected": -0.7244349122047424,
+      "eval_runtime": 25.2553,
+      "eval_samples_per_second": 28.826,
+      "eval_steps_per_second": 3.603,
+      "step": 200
+    },
+    {
+      "epoch": 0.18650088809946713,
+      "grad_norm": 0.27885496616363525,
+      "learning_rate": 4.762067631165049e-06,
+      "logits/chosen": 7.234966278076172,
+      "logits/rejected": 8.313450813293457,
+      "logps/chosen": -0.29102542996406555,
+      "logps/rejected": -0.49241799116134644,
+      "loss": 0.8556,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.43653813004493713,
+      "rewards/margins": 0.3020888566970825,
+      "rewards/rejected": -0.7386269569396973,
+      "step": 210
+    },
+    {
+      "epoch": 0.19538188277087035,
+      "grad_norm": 0.29907363653182983,
+      "learning_rate": 4.7392794005985324e-06,
+      "logits/chosen": 7.907521724700928,
+      "logits/rejected": 8.253190994262695,
+      "logps/chosen": -0.33691853284835815,
+      "logps/rejected": -0.4829257130622864,
+      "loss": 0.8236,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.5053777694702148,
+      "rewards/margins": 0.21901080012321472,
+      "rewards/rejected": -0.724388599395752,
+      "step": 220
+    },
+    {
+      "epoch": 0.20426287744227353,
+      "grad_norm": 0.282474547624588,
+      "learning_rate": 4.715508948078037e-06,
+      "logits/chosen": 6.367492198944092,
+      "logits/rejected": 6.273728370666504,
+      "logps/chosen": -0.3519875705242157,
+      "logps/rejected": -0.5284813642501831,
+      "loss": 0.8027,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.5279813408851624,
+      "rewards/margins": 0.2647407650947571,
+      "rewards/rejected": -0.7927221059799194,
+      "step": 230
+    },
+    {
+      "epoch": 0.21314387211367672,
+      "grad_norm": 0.327765554189682,
+      "learning_rate": 4.690766700109659e-06,
+      "logits/chosen": 5.090893268585205,
+      "logits/rejected": 4.768380165100098,
+      "logps/chosen": -0.3851698040962219,
+      "logps/rejected": -0.6464222073554993,
+      "loss": 0.7898,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.5777546167373657,
+      "rewards/margins": 0.391878604888916,
+      "rewards/rejected": -0.9696332812309265,
+      "step": 240
+    },
+    {
+      "epoch": 0.22202486678507993,
+      "grad_norm": 0.4895865321159363,
+      "learning_rate": 4.665063509461098e-06,
+      "logits/chosen": 4.056812286376953,
+      "logits/rejected": 3.723601818084717,
+      "logps/chosen": -0.4400455951690674,
+      "logps/rejected": -0.7731422781944275,
+      "loss": 0.7626,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.6600683927536011,
+      "rewards/margins": 0.49964505434036255,
+      "rewards/rejected": -1.1597135066986084,
+      "step": 250
+    },
+    {
+      "epoch": 0.22202486678507993,
+      "eval_logits/chosen": 2.420060396194458,
+      "eval_logits/rejected": 2.1626052856445312,
+      "eval_logps/chosen": -0.4724067151546478,
+      "eval_logps/rejected": -0.8418064117431641,
+      "eval_loss": 0.7631083130836487,
+      "eval_rewards/accuracies": 0.6483516693115234,
+      "eval_rewards/chosen": -0.7086100578308105,
+      "eval_rewards/margins": 0.5540997385978699,
+      "eval_rewards/rejected": -1.2627097368240356,
+      "eval_runtime": 25.2418,
+      "eval_samples_per_second": 28.841,
+      "eval_steps_per_second": 3.605,
+      "step": 250
+    },
+    {
+      "epoch": 0.23090586145648312,
+      "grad_norm": 0.46291017532348633,
+      "learning_rate": 4.638410650401267e-06,
+      "logits/chosen": 1.5297390222549438,
+      "logits/rejected": 1.1381648778915405,
+      "logps/chosen": -0.4418027997016907,
+      "logps/rejected": -1.0542564392089844,
+      "loss": 0.7026,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -0.6627041697502136,
+      "rewards/margins": 0.9186803698539734,
+      "rewards/rejected": -1.5813844203948975,
+      "step": 260
+    },
+    {
+      "epoch": 0.23978685612788633,
+      "grad_norm": 0.9783313870429993,
+      "learning_rate": 4.610819813755038e-06,
+      "logits/chosen": 2.8311033248901367,
+      "logits/rejected": 1.9742711782455444,
+      "logps/chosen": -0.5430587530136108,
+      "logps/rejected": -0.9841039776802063,
+      "loss": 0.7317,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.814588189125061,
+      "rewards/margins": 0.6615679860115051,
+      "rewards/rejected": -1.4761559963226318,
+      "step": 270
+    },
+    {
+      "epoch": 0.24866785079928952,
+      "grad_norm": 2.102562189102173,
+      "learning_rate": 4.582303101775249e-06,
+      "logits/chosen": 1.8241952657699585,
+      "logits/rejected": 0.8777934312820435,
+      "logps/chosen": -0.5624039769172668,
+      "logps/rejected": -1.1460126638412476,
+      "loss": 0.6887,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.8436058163642883,
+      "rewards/margins": 0.8754131197929382,
+      "rewards/rejected": -1.7190189361572266,
+      "step": 280
+    },
+    {
+      "epoch": 0.25754884547069273,
+      "grad_norm": 0.9813026189804077,
+      "learning_rate": 4.55287302283426e-06,
+      "logits/chosen": 2.370732069015503,
+      "logits/rejected": 1.4697134494781494,
+      "logps/chosen": -0.6739786863327026,
+      "logps/rejected": -1.6581566333770752,
+      "loss": 0.5695,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -1.0109679698944092,
+      "rewards/margins": 1.476266622543335,
+      "rewards/rejected": -2.487234592437744,
+      "step": 290
+    },
+    {
+      "epoch": 0.2664298401420959,
+      "grad_norm": 2.187314510345459,
+      "learning_rate": 4.522542485937369e-06,
+      "logits/chosen": 1.6230781078338623,
+      "logits/rejected": 0.5460122227668762,
+      "logps/chosen": -0.6433733701705933,
+      "logps/rejected": -2.1001811027526855,
+      "loss": 0.5366,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -0.9650601148605347,
+      "rewards/margins": 2.185211658477783,
+      "rewards/rejected": -3.1502718925476074,
+      "step": 300
+    },
+    {
+      "epoch": 0.2664298401420959,
+      "eval_logits/chosen": 1.4087599515914917,
+      "eval_logits/rejected": 0.7888947129249573,
+      "eval_logps/chosen": -0.7579545974731445,
+      "eval_logps/rejected": -2.0049116611480713,
+      "eval_loss": 0.551510214805603,
+      "eval_rewards/accuracies": 0.6813187003135681,
+      "eval_rewards/chosen": -1.1369318962097168,
+      "eval_rewards/margins": 1.8704355955123901,
+      "eval_rewards/rejected": -3.0073673725128174,
+      "eval_runtime": 25.2647,
+      "eval_samples_per_second": 28.815,
+      "eval_steps_per_second": 3.602,
+      "step": 300
+    },
+    {
+      "epoch": 0.2753108348134991,
+      "grad_norm": 0.7035408616065979,
+      "learning_rate": 4.491324795060491e-06,
+      "logits/chosen": 1.5831315517425537,
+      "logits/rejected": 0.46730250120162964,
+      "logps/chosen": -0.7262418866157532,
+      "logps/rejected": -2.1209158897399902,
+      "loss": 0.5524,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.0893628597259521,
+      "rewards/margins": 2.092010974884033,
+      "rewards/rejected": -3.1813735961914062,
+      "step": 310
+    },
+    {
+      "epoch": 0.2841918294849023,
+      "grad_norm": 0.5678634643554688,
+      "learning_rate": 4.4592336433146e-06,
+      "logits/chosen": 1.265734076499939,
+      "logits/rejected": 0.7576489448547363,
+      "logps/chosen": -0.7938942313194275,
+      "logps/rejected": -2.3495612144470215,
+      "loss": 0.5233,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -1.1908413171768188,
+      "rewards/margins": 2.333500385284424,
+      "rewards/rejected": -3.5243420600891113,
+      "step": 320
+    },
+    {
+      "epoch": 0.29307282415630553,
+      "grad_norm": 1.1373224258422852,
+      "learning_rate": 4.426283106939474e-06,
+      "logits/chosen": 2.977414846420288,
+      "logits/rejected": 2.1573710441589355,
+      "logps/chosen": -0.8513160943984985,
+      "logps/rejected": -2.4125566482543945,
+      "loss": 0.556,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -1.2769742012023926,
+      "rewards/margins": 2.341860771179199,
+      "rewards/rejected": -3.6188347339630127,
+      "step": 330
+    },
+    {
+      "epoch": 0.3019538188277087,
+      "grad_norm": 4.7876176834106445,
+      "learning_rate": 4.3924876391293915e-06,
+      "logits/chosen": 2.4026589393615723,
+      "logits/rejected": 1.207395315170288,
+      "logps/chosen": -0.8529679179191589,
+      "logps/rejected": -2.456879138946533,
+      "loss": 0.5592,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -1.279451847076416,
+      "rewards/margins": 2.4058666229248047,
+      "rewards/rejected": -3.6853187084198,
+      "step": 340
+    },
+    {
+      "epoch": 0.3108348134991119,
+      "grad_norm": 0.5053763389587402,
+      "learning_rate": 4.357862063693486e-06,
+      "logits/chosen": 2.434265375137329,
+      "logits/rejected": 1.2504141330718994,
+      "logps/chosen": -0.9489291310310364,
+      "logps/rejected": -2.8521530628204346,
+      "loss": 0.4737,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -1.423393726348877,
+      "rewards/margins": 2.8548355102539062,
+      "rewards/rejected": -4.278229236602783,
+      "step": 350
+    },
+    {
+      "epoch": 0.3108348134991119,
+      "eval_logits/chosen": 1.6632592678070068,
+      "eval_logits/rejected": 1.235045075416565,
+      "eval_logps/chosen": -1.0692518949508667,
+      "eval_logps/rejected": -2.7428486347198486,
+      "eval_loss": 0.5021397471427917,
+      "eval_rewards/accuracies": 0.692307710647583,
+      "eval_rewards/chosen": -1.6038777828216553,
+      "eval_rewards/margins": 2.510395050048828,
+      "eval_rewards/rejected": -4.1142730712890625,
+      "eval_runtime": 25.2582,
+      "eval_samples_per_second": 28.822,
+      "eval_steps_per_second": 3.603,
+      "step": 350
+    },
+    {
+      "epoch": 0.3197158081705151,
+      "grad_norm": 0.8040274381637573,
+      "learning_rate": 4.322421568553529e-06,
+      "logits/chosen": 1.187036395072937,
+      "logits/rejected": 0.4290788769721985,
+      "logps/chosen": -1.1015206575393677,
+      "logps/rejected": -2.919748544692993,
+      "loss": 0.489,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -1.6522810459136963,
+      "rewards/margins": 2.727341651916504,
+      "rewards/rejected": -4.379622459411621,
+      "step": 360
+    },
+    {
+      "epoch": 0.3285968028419183,
+      "grad_norm": 0.9299562573432922,
+      "learning_rate": 4.286181699082008e-06,
+      "logits/chosen": 2.5852127075195312,
+      "logits/rejected": 2.0419259071350098,
+      "logps/chosen": -1.1498607397079468,
+      "logps/rejected": -3.0336194038391113,
+      "loss": 0.4812,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.724791169166565,
+      "rewards/margins": 2.8256375789642334,
+      "rewards/rejected": -4.55042839050293,
+      "step": 370
+    },
+    {
+      "epoch": 0.33747779751332146,
+      "grad_norm": 1.7739671468734741,
+      "learning_rate": 4.249158351283414e-06,
+      "logits/chosen": 2.246245861053467,
+      "logits/rejected": 1.5551975965499878,
+      "logps/chosen": -1.254900574684143,
+      "logps/rejected": -3.206178665161133,
+      "loss": 0.4651,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -1.8823509216308594,
+      "rewards/margins": 2.926917552947998,
+      "rewards/rejected": -4.809267997741699,
+      "step": 380
+    },
+    {
+      "epoch": 0.3463587921847247,
+      "grad_norm": 4.380665302276611,
+      "learning_rate": 4.211367764821722e-06,
+      "logits/chosen": 3.0754549503326416,
+      "logits/rejected": 2.622124433517456,
+      "logps/chosen": -1.9250037670135498,
+      "logps/rejected": -3.69482421875,
+      "loss": 0.4292,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -2.8875060081481934,
+      "rewards/margins": 2.6547305583953857,
+      "rewards/rejected": -5.542236328125,
+      "step": 390
+    },
+    {
+      "epoch": 0.3552397868561279,
+      "grad_norm": 1.5087212324142456,
+      "learning_rate": 4.172826515897146e-06,
+      "logits/chosen": 2.2718021869659424,
+      "logits/rejected": 1.8861210346221924,
+      "logps/chosen": -2.4473955631256104,
+      "logps/rejected": -4.387387752532959,
+      "loss": 0.3902,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -3.671093702316284,
+      "rewards/margins": 2.9099888801574707,
+      "rewards/rejected": -6.581082344055176,
+      "step": 400
+    },
+    {
+      "epoch": 0.3552397868561279,
+      "eval_logits/chosen": 1.759078860282898,
+      "eval_logits/rejected": 1.5246928930282593,
+      "eval_logps/chosen": -2.720665454864502,
+      "eval_logps/rejected": -4.613493919372559,
+      "eval_loss": 0.4054907560348511,
+      "eval_rewards/accuracies": 0.8791208863258362,
+      "eval_rewards/chosen": -4.080998420715332,
+      "eval_rewards/margins": 2.839242696762085,
+      "eval_rewards/rejected": -6.920241355895996,
+      "eval_runtime": 25.2363,
+      "eval_samples_per_second": 28.847,
+      "eval_steps_per_second": 3.606,
+      "step": 400
+    },
+    {
+      "epoch": 0.3641207815275311,
+      "grad_norm": 6.079421043395996,
+      "learning_rate": 4.133551509975264e-06,
+      "logits/chosen": 1.8841949701309204,
+      "logits/rejected": 1.3479797840118408,
+      "logps/chosen": -2.517265796661377,
+      "logps/rejected": -4.453648567199707,
+      "loss": 0.3977,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -3.7758986949920654,
+      "rewards/margins": 2.9045748710632324,
+      "rewards/rejected": -6.680473327636719,
+      "step": 410
+    },
+    {
+      "epoch": 0.37300177619893427,
+      "grad_norm": 3.0998194217681885,
+      "learning_rate": 4.093559974371725e-06,
+      "logits/chosen": 1.6409276723861694,
+      "logits/rejected": 1.2141990661621094,
+      "logps/chosen": -2.2561168670654297,
+      "logps/rejected": -4.470211029052734,
+      "loss": 0.3527,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -3.3841750621795654,
+      "rewards/margins": 3.321141004562378,
+      "rewards/rejected": -6.70531702041626,
+      "step": 420
+    },
+    {
+      "epoch": 0.38188277087033745,
+      "grad_norm": 6.982161045074463,
+      "learning_rate": 4.052869450695776e-06,
+      "logits/chosen": 2.835188388824463,
+      "logits/rejected": 2.3657329082489014,
+      "logps/chosen": -2.8557300567626953,
+      "logps/rejected": -5.075521469116211,
+      "loss": 0.387,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.283595085144043,
+      "rewards/margins": 3.3296875953674316,
+      "rewards/rejected": -7.613282680511475,
+      "step": 430
+    },
+    {
+      "epoch": 0.3907637655417407,
+      "grad_norm": 2.139338970184326,
+      "learning_rate": 4.011497787155938e-06,
+      "logits/chosen": 2.126509189605713,
+      "logits/rejected": 1.459567904472351,
+      "logps/chosen": -3.1412863731384277,
+      "logps/rejected": -5.423466682434082,
+      "loss": 0.3611,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.711928844451904,
+      "rewards/margins": 3.4232699871063232,
+      "rewards/rejected": -8.135198593139648,
+      "step": 440
+    },
+    {
+      "epoch": 0.3996447602131439,
+      "grad_norm": 1.7899377346038818,
+      "learning_rate": 3.969463130731183e-06,
+      "logits/chosen": 2.4551379680633545,
+      "logits/rejected": 2.0784289836883545,
+      "logps/chosen": -3.098043203353882,
+      "logps/rejected": -5.300747871398926,
+      "loss": 0.354,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -4.647065162658691,
+      "rewards/margins": 3.3040566444396973,
+      "rewards/rejected": -7.951122283935547,
+      "step": 450
+    },
+    {
+      "epoch": 0.3996447602131439,
+      "eval_logits/chosen": 1.9761625528335571,
+      "eval_logits/rejected": 1.6654667854309082,
+      "eval_logps/chosen": -2.8789772987365723,
+      "eval_logps/rejected": -5.1105055809021,
+      "eval_loss": 0.36211252212524414,
+      "eval_rewards/accuracies": 0.8791208863258362,
+      "eval_rewards/chosen": -4.3184661865234375,
+      "eval_rewards/margins": 3.347292423248291,
+      "eval_rewards/rejected": -7.6657586097717285,
+      "eval_runtime": 25.2549,
+      "eval_samples_per_second": 28.826,
+      "eval_steps_per_second": 3.603,
+      "step": 450
+    },
+    {
+      "epoch": 0.40852575488454707,
+      "grad_norm": 1.9171936511993408,
+      "learning_rate": 3.92678391921108e-06,
+      "logits/chosen": 2.1672446727752686,
+      "logits/rejected": 1.6228408813476562,
+      "logps/chosen": -2.688931703567505,
+      "logps/rejected": -5.2408246994018555,
+      "loss": 0.3266,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.033397197723389,
+      "rewards/margins": 3.8278393745422363,
+      "rewards/rejected": -7.861237525939941,
+      "step": 460
+    },
+    {
+      "epoch": 0.41740674955595025,
+      "grad_norm": 1.702635407447815,
+      "learning_rate": 3.88347887310836e-06,
+      "logits/chosen": 2.3164448738098145,
+      "logits/rejected": 2.047529697418213,
+      "logps/chosen": -2.6861701011657715,
+      "logps/rejected": -5.629918098449707,
+      "loss": 0.3339,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.02925443649292,
+      "rewards/margins": 4.415622711181641,
+      "rewards/rejected": -8.444877624511719,
+      "step": 470
+    },
+    {
+      "epoch": 0.42628774422735344,
+      "grad_norm": 2.48634934425354,
+      "learning_rate": 3.839566987447492e-06,
+      "logits/chosen": 2.5225472450256348,
+      "logits/rejected": 2.0870003700256348,
+      "logps/chosen": -3.041111946105957,
+      "logps/rejected": -5.3499016761779785,
+      "loss": 0.3226,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.561667442321777,
+      "rewards/margins": 3.4631850719451904,
+      "rewards/rejected": -8.024852752685547,
+      "step": 480
+    },
+    {
+      "epoch": 0.4351687388987567,
+      "grad_norm": 4.728499412536621,
+      "learning_rate": 3.795067523432826e-06,
+      "logits/chosen": 2.33893084526062,
+      "logits/rejected": 1.7909936904907227,
+      "logps/chosen": -2.7356209754943848,
+      "logps/rejected": -5.33417272567749,
+      "loss": 0.322,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.103431224822998,
+      "rewards/margins": 3.8978283405303955,
+      "rewards/rejected": -8.001258850097656,
+      "step": 490
+    },
+    {
+      "epoch": 0.44404973357015987,
+      "grad_norm": 8.412679672241211,
+      "learning_rate": 3.7500000000000005e-06,
+      "logits/chosen": 2.788668632507324,
+      "logits/rejected": 2.439873695373535,
+      "logps/chosen": -3.3219153881073,
+      "logps/rejected": -5.992051124572754,
+      "loss": 0.3075,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.98287296295166,
+      "rewards/margins": 4.005204200744629,
+      "rewards/rejected": -8.988077163696289,
+      "step": 500
+    },
+    {
+      "epoch": 0.44404973357015987,
+      "eval_logits/chosen": 2.165436029434204,
+      "eval_logits/rejected": 1.8186790943145752,
+      "eval_logps/chosen": -3.4299349784851074,
+      "eval_logps/rejected": -6.0660552978515625,
+      "eval_loss": 0.3319137990474701,
+      "eval_rewards/accuracies": 0.8901098966598511,
+      "eval_rewards/chosen": -5.14490270614624,
+      "eval_rewards/margins": 3.954181671142578,
+      "eval_rewards/rejected": -9.09908390045166,
+      "eval_runtime": 25.2602,
+      "eval_samples_per_second": 28.82,
+      "eval_steps_per_second": 3.603,
+      "step": 500
+    },
+    {
+      "epoch": 0.45293072824156305,
+      "grad_norm": 2.8339133262634277,
+      "learning_rate": 3.7043841852542884e-06,
+      "logits/chosen": 2.109018325805664,
+      "logits/rejected": 1.6996265649795532,
+      "logps/chosen": -3.288560390472412,
+      "logps/rejected": -5.986764430999756,
+      "loss": 0.3116,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.932840824127197,
+      "rewards/margins": 4.047306060791016,
+      "rewards/rejected": -8.980146408081055,
+      "step": 510
+    },
+    {
+      "epoch": 0.46181172291296624,
+      "grad_norm": 3.8578269481658936,
+      "learning_rate": 3.658240087799655e-06,
+      "logits/chosen": 1.7659969329833984,
+      "logits/rejected": 1.4596515893936157,
+      "logps/chosen": -3.0301425457000732,
+      "logps/rejected": -6.252682209014893,
+      "loss": 0.3015,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.5452141761779785,
+      "rewards/margins": 4.833809852600098,
+      "rewards/rejected": -9.379022598266602,
+      "step": 520
+    },
+    {
+      "epoch": 0.4706927175843694,
+      "grad_norm": 2.7795143127441406,
+      "learning_rate": 3.611587947962319e-06,
+      "logits/chosen": 2.472006320953369,
+      "logits/rejected": 1.9400993585586548,
+      "logps/chosen": -3.2479186058044434,
+      "logps/rejected": -6.475512504577637,
+      "loss": 0.3006,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.871877193450928,
+      "rewards/margins": 4.841391086578369,
+      "rewards/rejected": -9.713269233703613,
+      "step": 530
+    },
+    {
+      "epoch": 0.47957371225577267,
+      "grad_norm": 3.5200746059417725,
+      "learning_rate": 3.564448228912682e-06,
+      "logits/chosen": 2.911531925201416,
+      "logits/rejected": 2.2947440147399902,
+      "logps/chosen": -3.387556791305542,
+      "logps/rejected": -6.559035301208496,
+      "loss": 0.2684,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.081335544586182,
+      "rewards/margins": 4.757218360900879,
+      "rewards/rejected": -9.838552474975586,
+      "step": 540
+    },
+    {
+      "epoch": 0.48845470692717585,
+      "grad_norm": 2.368495225906372,
+      "learning_rate": 3.516841607689501e-06,
+      "logits/chosen": 1.276886224746704,
+      "logits/rejected": 1.3811718225479126,
+      "logps/chosen": -2.9689180850982666,
+      "logps/rejected": -6.619606971740723,
+      "loss": 0.3159,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.4533772468566895,
+      "rewards/margins": 5.4760332107543945,
+      "rewards/rejected": -9.929410934448242,
+      "step": 550
+    },
+    {
+      "epoch": 0.48845470692717585,
+      "eval_logits/chosen": 2.2735824584960938,
+      "eval_logits/rejected": 1.9788992404937744,
+      "eval_logps/chosen": -3.698131799697876,
+      "eval_logps/rejected": -6.64966344833374,
+      "eval_loss": 0.30747923254966736,
+      "eval_rewards/accuracies": 0.9230769276618958,
+      "eval_rewards/chosen": -5.5471978187561035,
+      "eval_rewards/margins": 4.427298545837402,
+      "eval_rewards/rejected": -9.974496841430664,
+      "eval_runtime": 25.2322,
+      "eval_samples_per_second": 28.852,
+      "eval_steps_per_second": 3.606,
+      "step": 550
+    },
+    {
+      "epoch": 0.49733570159857904,
+      "grad_norm": 1.901207685470581,
+      "learning_rate": 3.4687889661302577e-06,
+      "logits/chosen": 1.9734981060028076,
+      "logits/rejected": 1.8617655038833618,
+      "logps/chosen": -3.464953899383545,
+      "logps/rejected": -6.746106147766113,
+      "loss": 0.2912,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.1974310874938965,
+      "rewards/margins": 4.921727180480957,
+      "rewards/rejected": -10.119158744812012,
+      "step": 560
+    },
+    {
+      "epoch": 0.5062166962699822,
+      "grad_norm": 3.526299238204956,
+      "learning_rate": 3.4203113817116955e-06,
+      "logits/chosen": 3.0836069583892822,
+      "logits/rejected": 2.75875186920166,
+      "logps/chosen": -3.5981743335723877,
+      "logps/rejected": -6.405775547027588,
+      "loss": 0.3284,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.397261142730713,
+      "rewards/margins": 4.211403846740723,
+      "rewards/rejected": -9.608665466308594,
+      "step": 570
+    },
+    {
+      "epoch": 0.5150976909413855,
+      "grad_norm": 2.816272497177124,
+      "learning_rate": 3.3714301183045382e-06,
+      "logits/chosen": 3.0068678855895996,
+      "logits/rejected": 2.466287136077881,
+      "logps/chosen": -3.7227580547332764,
+      "logps/rejected": -6.9311113357543945,
+      "loss": 0.2728,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.584136962890625,
+      "rewards/margins": 4.812530040740967,
+      "rewards/rejected": -10.39666748046875,
+      "step": 580
+    },
+    {
+      "epoch": 0.5239786856127886,
+      "grad_norm": 2.433389902114868,
+      "learning_rate": 3.3221666168464584e-06,
+      "logits/chosen": 2.9992904663085938,
+      "logits/rejected": 2.678699254989624,
+      "logps/chosen": -3.540968418121338,
+      "logps/rejected": -7.228091239929199,
+      "loss": 0.2568,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.3114519119262695,
+      "rewards/margins": 5.530684947967529,
+      "rewards/rejected": -10.842137336730957,
+      "step": 590
+    },
+    {
+      "epoch": 0.5328596802841918,
+      "grad_norm": 2.7557125091552734,
+      "learning_rate": 3.272542485937369e-06,
+      "logits/chosen": 2.557410478591919,
+      "logits/rejected": 2.331958770751953,
+      "logps/chosen": -3.9404635429382324,
+      "logps/rejected": -7.266766548156738,
+      "loss": 0.2639,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.9106950759887695,
+      "rewards/margins": 4.9894537925720215,
+      "rewards/rejected": -10.900148391723633,
+      "step": 600
+    },
+    {
+      "epoch": 0.5328596802841918,
+      "eval_logits/chosen": 2.561415910720825,
+      "eval_logits/rejected": 2.2971484661102295,
+      "eval_logps/chosen": -4.015191555023193,
+      "eval_logps/rejected": -7.222255229949951,
+      "eval_loss": 0.28853774070739746,
+      "eval_rewards/accuracies": 0.9340659379959106,
+      "eval_rewards/chosen": -6.022787570953369,
+      "eval_rewards/margins": 4.810595989227295,
+      "eval_rewards/rejected": -10.833383560180664,
+      "eval_runtime": 25.2577,
+      "eval_samples_per_second": 28.823,
+      "eval_steps_per_second": 3.603,
+      "step": 600
+    },
+    {
+      "epoch": 0.5417406749555951,
+      "grad_norm": 2.582770824432373,
+      "learning_rate": 3.222579492361179e-06,
+      "logits/chosen": 2.7404181957244873,
+      "logits/rejected": 2.540113687515259,
+      "logps/chosen": -3.9154396057128906,
+      "logps/rejected": -7.631985664367676,
+      "loss": 0.2816,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.873159408569336,
+      "rewards/margins": 5.574820041656494,
+      "rewards/rejected": -11.447979927062988,
+      "step": 610
+    },
+    {
+      "epoch": 0.5506216696269982,
+      "grad_norm": 3.8167436122894287,
+      "learning_rate": 3.1722995515381644e-06,
+      "logits/chosen": 2.445218563079834,
+      "logits/rejected": 2.288620710372925,
+      "logps/chosen": -3.7501556873321533,
+      "logps/rejected": -7.918539524078369,
+      "loss": 0.2891,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.625233173370361,
+      "rewards/margins": 6.252577304840088,
+      "rewards/rejected": -11.877809524536133,
+      "step": 620
+    },
+    {
+      "epoch": 0.5595026642984015,
+      "grad_norm": 3.57536244392395,
+      "learning_rate": 3.121724717912138e-06,
+      "logits/chosen": 2.8337388038635254,
+      "logits/rejected": 2.1241557598114014,
+      "logps/chosen": -3.7040035724639893,
+      "logps/rejected": -7.53197717666626,
+      "loss": 0.2702,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.556005477905273,
+      "rewards/margins": 5.741961479187012,
+      "rewards/rejected": -11.297966003417969,
+      "step": 630
+    },
+    {
+      "epoch": 0.5683836589698046,
+      "grad_norm": 3.0520713329315186,
+      "learning_rate": 3.0708771752766397e-06,
+      "logits/chosen": 2.5255160331726074,
+      "logits/rejected": 2.0742428302764893,
+      "logps/chosen": -3.908573865890503,
+      "logps/rejected": -7.60653829574585,
+      "loss": 0.2598,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.862860679626465,
+      "rewards/margins": 5.546946048736572,
+      "rewards/rejected": -11.409807205200195,
+      "step": 640
+    },
+    {
+      "epoch": 0.5772646536412078,
+      "grad_norm": 8.067182540893555,
+      "learning_rate": 3.019779227044398e-06,
+      "logits/chosen": 1.6965067386627197,
+      "logits/rejected": 1.644667625427246,
+      "logps/chosen": -3.6588027477264404,
+      "logps/rejected": -7.457572937011719,
+      "loss": 0.2453,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.488204002380371,
+      "rewards/margins": 5.698155403137207,
+      "rewards/rejected": -11.186359405517578,
+      "step": 650
+    },
+    {
+      "epoch": 0.5772646536412078,
+      "eval_logits/chosen": 2.577754020690918,
+      "eval_logits/rejected": 2.265626907348633,
+      "eval_logps/chosen": -3.906606435775757,
+      "eval_logps/rejected": -7.3099446296691895,
+      "eval_loss": 0.27302286028862,
+      "eval_rewards/accuracies": 0.9560439586639404,
+      "eval_rewards/chosen": -5.859910011291504,
+      "eval_rewards/margins": 5.105007171630859,
+      "eval_rewards/rejected": -10.964917182922363,
+      "eval_runtime": 25.2446,
+      "eval_samples_per_second": 28.838,
+      "eval_steps_per_second": 3.605,
+      "step": 650
+    },
+    {
+      "epoch": 0.5861456483126111,
+      "grad_norm": 2.9298255443573,
+      "learning_rate": 2.9684532864643123e-06,
+      "logits/chosen": 2.901702404022217,
+      "logits/rejected": 2.595918655395508,
+      "logps/chosen": -3.9949543476104736,
+      "logps/rejected": -7.430356502532959,
+      "loss": 0.2726,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -5.992431163787842,
+      "rewards/margins": 5.153104782104492,
+      "rewards/rejected": -11.145535469055176,
+      "step": 660
+    },
+    {
+      "epoch": 0.5950266429840142,
+      "grad_norm": 3.085571050643921,
+      "learning_rate": 2.9169218667902562e-06,
+      "logits/chosen": 2.8471388816833496,
+      "logits/rejected": 2.5500330924987793,
+      "logps/chosen": -4.014147758483887,
+      "logps/rejected": -7.298794746398926,
+      "loss": 0.2484,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -6.02122163772583,
+      "rewards/margins": 4.926970481872559,
+      "rewards/rejected": -10.94819164276123,
+      "step": 670
+    },
+    {
+      "epoch": 0.6039076376554174,
+      "grad_norm": 2.3615477085113525,
+      "learning_rate": 2.8652075714060296e-06,
+      "logits/chosen": 2.495004177093506,
+      "logits/rejected": 1.9873936176300049,
+      "logps/chosen": -4.2625041007995605,
+      "logps/rejected": -8.186586380004883,
+      "loss": 0.2144,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -6.393756866455078,
+      "rewards/margins": 5.886124610900879,
+      "rewards/rejected": -12.279881477355957,
+      "step": 680
+    },
+    {
+      "epoch": 0.6127886323268206,
+      "grad_norm": 3.497316837310791,
+      "learning_rate": 2.813333083910761e-06,
+      "logits/chosen": 1.540621042251587,
+      "logits/rejected": 1.2002273797988892,
+      "logps/chosen": -3.71490740776062,
+      "logps/rejected": -8.136199951171875,
+      "loss": 0.2861,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.572361469268799,
+      "rewards/margins": 6.631939888000488,
+      "rewards/rejected": -12.204300880432129,
+      "step": 690
+    },
+    {
+      "epoch": 0.6216696269982238,
+      "grad_norm": 3.2540247440338135,
+      "learning_rate": 2.761321158169134e-06,
+      "logits/chosen": 2.776721477508545,
+      "logits/rejected": 2.475888729095459,
+      "logps/chosen": -4.3385329246521,
+      "logps/rejected": -7.8828301429748535,
+      "loss": 0.2885,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -6.507800102233887,
+      "rewards/margins": 5.316445350646973,
+      "rewards/rejected": -11.824244499206543,
+      "step": 700
+    },
+    {
+      "epoch": 0.6216696269982238,
+      "eval_logits/chosen": 2.4267516136169434,
+      "eval_logits/rejected": 2.1245739459991455,
+      "eval_logps/chosen": -3.9584598541259766,
+      "eval_logps/rejected": -7.546706199645996,
+      "eval_loss": 0.26512712240219116,
+      "eval_rewards/accuracies": 0.9340659379959106,
+      "eval_rewards/chosen": -5.937689781188965,
+      "eval_rewards/margins": 5.382368564605713,
+      "eval_rewards/rejected": -11.320058822631836,
+      "eval_runtime": 25.2589,
+      "eval_samples_per_second": 28.822,
+      "eval_steps_per_second": 3.603,
+      "step": 700
+    },
+    {
+      "epoch": 0.6305506216696269,
+      "grad_norm": 2.582273483276367,
+      "learning_rate": 2.70919460833079e-06,
+      "logits/chosen": 2.2102553844451904,
+      "logits/rejected": 1.9504516124725342,
+      "logps/chosen": -3.7470879554748535,
+      "logps/rejected": -7.761659145355225,
+      "loss": 0.2078,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.620632171630859,
+      "rewards/margins": 6.02185583114624,
+      "rewards/rejected": -11.642488479614258,
+      "step": 710
+    },
+    {
+      "epoch": 0.6394316163410302,
+      "grad_norm": 2.2855663299560547,
+      "learning_rate": 2.6569762988232838e-06,
+      "logits/chosen": 2.4659409523010254,
+      "logits/rejected": 2.0434811115264893,
+      "logps/chosen": -3.4346442222595215,
+      "logps/rejected": -7.470824241638184,
+      "loss": 0.2346,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.151965618133545,
+      "rewards/margins": 6.054270267486572,
+      "rewards/rejected": -11.206236839294434,
+      "step": 720
+    },
+    {
+      "epoch": 0.6483126110124334,
+      "grad_norm": 2.038733959197998,
+      "learning_rate": 2.604689134322999e-06,
+      "logits/chosen": 2.270310878753662,
+      "logits/rejected": 1.9651508331298828,
+      "logps/chosen": -3.721379518508911,
+      "logps/rejected": -7.776650905609131,
+      "loss": 0.2354,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.582068920135498,
+      "rewards/margins": 6.0829057693481445,
+      "rewards/rejected": -11.6649751663208,
+      "step": 730
+    },
+    {
+      "epoch": 0.6571936056838366,
+      "grad_norm": 2.948915481567383,
+      "learning_rate": 2.5523560497083927e-06,
+      "logits/chosen": 2.358057737350464,
+      "logits/rejected": 2.063586711883545,
+      "logps/chosen": -3.6883864402770996,
+      "logps/rejected": -7.352984428405762,
+      "loss": 0.231,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.5325798988342285,
+      "rewards/margins": 5.496896266937256,
+      "rewards/rejected": -11.0294771194458,
+      "step": 740
+    },
+    {
+      "epoch": 0.6660746003552398,
+      "grad_norm": 3.661870002746582,
+      "learning_rate": 2.5e-06,
+      "logits/chosen": 2.6712796688079834,
+      "logits/rejected": 2.393817901611328,
+      "logps/chosen": -4.065141201019287,
+      "logps/rejected": -7.9232306480407715,
+      "loss": 0.2332,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -6.097712516784668,
+      "rewards/margins": 5.787134170532227,
+      "rewards/rejected": -11.884846687316895,
+      "step": 750
+    },
+    {
+      "epoch": 0.6660746003552398,
+      "eval_logits/chosen": 2.4871790409088135,
+      "eval_logits/rejected": 2.2864174842834473,
+      "eval_logps/chosen": -4.176412105560303,
+      "eval_logps/rejected": -8.009041786193848,
+      "eval_loss": 0.2521994411945343,
+      "eval_rewards/accuracies": 0.9340659379959106,
+      "eval_rewards/chosen": -6.264617443084717,
+      "eval_rewards/margins": 5.7489447593688965,
+      "eval_rewards/rejected": -12.01356315612793,
+      "eval_runtime": 25.2624,
+      "eval_samples_per_second": 28.818,
+      "eval_steps_per_second": 3.602,
+      "step": 750
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 1500,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 50,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.7692224572998287e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-750/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1a9f338a2c311ad932497119af4f36e393c3e914b2390e2dd2e31575d923108
+size 7224

checkpoint-750/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)