Training in progress, step 800, checkpoint

checkpoint-800/README.md

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

checkpoint-800/adapter_config.json

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

checkpoint-800/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fc9c3bf696aec981b040faf7fa40379bc826255f7834dc74701cc30ac2ea708
+size 25200088

checkpoint-800/global_step800/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91ec8c61cc9522c6c05e9926c4d84a175ac41dcb0dfdc188d1e5c43cc248eb08
+size 18881328

checkpoint-800/global_step800/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2bb7ebfea599685c59c5a3d5b53ed545060412469dee3a427f500914b3ef540
+size 18881328

checkpoint-800/global_step800/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae90ff648a53d97eef03aedcd993cac136f5ec258f23434a5c41a93248bec285
+size 18881328

checkpoint-800/global_step800/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:722a6f4fb2315f8576561de51b58fb5746819b6665310e815cb772da15f063b0
+size 18881392

checkpoint-800/global_step800/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc3bea06eca0f05bf2c46e433883a5f44111d1038313762f6d573a591070f93f
+size 18881392

checkpoint-800/global_step800/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3389951c8ab0ef9e9ba18d5feeab7e59817e7b323132e0181b829a6a9dc5a1d7
+size 18881392

checkpoint-800/global_step800/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad3e2ce9a40f5b59091c9c283bc05fb847a6888226e697353ab077ec68cd27a4
+size 18881392

checkpoint-800/global_step800/bf16_zero_pp_rank_7_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a9031501428cdd8aeda1a8b59e3512340202db2f7459200d05284e00566d150
+size 18881392

checkpoint-800/global_step800/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4cc322f3de1beaa559966dae388da2aaee37a60409451938e0b078b5d97b3e05
+size 25379244

checkpoint-800/latest

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

checkpoint-800/rng_state_0.pth

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

checkpoint-800/rng_state_1.pth

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

checkpoint-800/rng_state_2.pth

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

checkpoint-800/rng_state_3.pth

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

checkpoint-800/rng_state_4.pth

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

checkpoint-800/rng_state_5.pth

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

checkpoint-800/rng_state_6.pth

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

checkpoint-800/rng_state_7.pth

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

checkpoint-800/scheduler.pt

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

checkpoint-800/special_tokens_map.json

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

checkpoint-800/tokenizer_config.json

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

checkpoint-800/trainer_state.json

@@ -0,0 +1,1489 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.7104795737122558,
+  "eval_steps": 50,
+  "global_step": 800,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008880994671403197,
+      "grad_norm": 0.045356735587120056,
+      "learning_rate": 4.999451708687114e-06,
+      "logits/chosen": 14.352750778198242,
+      "logits/rejected": 14.82281494140625,
+      "logps/chosen": -0.2592294216156006,
+      "logps/rejected": -0.32852867245674133,
+      "loss": 0.9315,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.38884416222572327,
+      "rewards/margins": 0.10394889116287231,
+      "rewards/rejected": -0.4927930235862732,
+      "step": 10
+    },
+    {
+      "epoch": 0.017761989342806393,
+      "grad_norm": 0.05255189165472984,
+      "learning_rate": 4.997807075247147e-06,
+      "logits/chosen": 14.80792236328125,
+      "logits/rejected": 15.017183303833008,
+      "logps/chosen": -0.2825874388217926,
+      "logps/rejected": -0.36373966932296753,
+      "loss": 0.9318,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.4238811433315277,
+      "rewards/margins": 0.12172831594944,
+      "rewards/rejected": -0.5456094741821289,
+      "step": 20
+    },
+    {
+      "epoch": 0.02664298401420959,
+      "grad_norm": 0.056027185171842575,
+      "learning_rate": 4.9950668210706795e-06,
+      "logits/chosen": 14.45336627960205,
+      "logits/rejected": 15.261484146118164,
+      "logps/chosen": -0.2638034522533417,
+      "logps/rejected": -0.37216562032699585,
+      "loss": 0.9211,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.3957051932811737,
+      "rewards/margins": 0.1625431776046753,
+      "rewards/rejected": -0.5582484006881714,
+      "step": 30
+    },
+    {
+      "epoch": 0.035523978685612786,
+      "grad_norm": 0.06511653959751129,
+      "learning_rate": 4.9912321481237616e-06,
+      "logits/chosen": 14.59851360321045,
+      "logits/rejected": 15.112770080566406,
+      "logps/chosen": -0.28972768783569336,
+      "logps/rejected": -0.36043626070022583,
+      "loss": 0.941,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.43459147214889526,
+      "rewards/margins": 0.10606291145086288,
+      "rewards/rejected": -0.5406544208526611,
+      "step": 40
+    },
+    {
+      "epoch": 0.04440497335701599,
+      "grad_norm": 0.0637577474117279,
+      "learning_rate": 4.986304738420684e-06,
+      "logits/chosen": 14.678179740905762,
+      "logits/rejected": 15.114399909973145,
+      "logps/chosen": -0.3033604919910431,
+      "logps/rejected": -0.3262741267681122,
+      "loss": 0.929,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.45504075288772583,
+      "rewards/margins": 0.03437047079205513,
+      "rewards/rejected": -0.4894111752510071,
+      "step": 50
+    },
+    {
+      "epoch": 0.04440497335701599,
+      "eval_logits/chosen": 14.936029434204102,
+      "eval_logits/rejected": 14.780267715454102,
+      "eval_logps/chosen": -0.29386037588119507,
+      "eval_logps/rejected": -0.3304942548274994,
+      "eval_loss": 0.9458721876144409,
+      "eval_rewards/accuracies": 0.49450549483299255,
+      "eval_rewards/chosen": -0.4407905340194702,
+      "eval_rewards/margins": 0.05495081841945648,
+      "eval_rewards/rejected": -0.49574142694473267,
+      "eval_runtime": 27.7436,
+      "eval_samples_per_second": 26.24,
+      "eval_steps_per_second": 3.28,
+      "step": 50
+    },
+    {
+      "epoch": 0.05328596802841918,
+      "grad_norm": 0.06470987200737,
+      "learning_rate": 4.980286753286196e-06,
+      "logits/chosen": 14.19079303741455,
+      "logits/rejected": 14.986845016479492,
+      "logps/chosen": -0.26126712560653687,
+      "logps/rejected": -0.31976616382598877,
+      "loss": 0.9335,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.3919006288051605,
+      "rewards/margins": 0.08774860948324203,
+      "rewards/rejected": -0.47964924573898315,
+      "step": 60
+    },
+    {
+      "epoch": 0.06216696269982238,
+      "grad_norm": 0.06163545697927475,
+      "learning_rate": 4.973180832407471e-06,
+      "logits/chosen": 14.062408447265625,
+      "logits/rejected": 15.050743103027344,
+      "logps/chosen": -0.2698076367378235,
+      "logps/rejected": -0.37131980061531067,
+      "loss": 0.9234,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.4047114849090576,
+      "rewards/margins": 0.1522682160139084,
+      "rewards/rejected": -0.556979775428772,
+      "step": 70
+    },
+    {
+      "epoch": 0.07104795737122557,
+      "grad_norm": 0.05943402647972107,
+      "learning_rate": 4.964990092676263e-06,
+      "logits/chosen": 14.523780822753906,
+      "logits/rejected": 15.173608779907227,
+      "logps/chosen": -0.2835150957107544,
+      "logps/rejected": -0.35379332304000854,
+      "loss": 0.9317,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.4252726137638092,
+      "rewards/margins": 0.10541732609272003,
+      "rewards/rejected": -0.5306899547576904,
+      "step": 80
+    },
+    {
+      "epoch": 0.07992895204262877,
+      "grad_norm": 0.05566830188035965,
+      "learning_rate": 4.9557181268217225e-06,
+      "logits/chosen": 14.752237319946289,
+      "logits/rejected": 15.213434219360352,
+      "logps/chosen": -0.25490278005599976,
+      "logps/rejected": -0.31673288345336914,
+      "loss": 0.9276,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.38235417008399963,
+      "rewards/margins": 0.09274514764547348,
+      "rewards/rejected": -0.4750993847846985,
+      "step": 90
+    },
+    {
+      "epoch": 0.08880994671403197,
+      "grad_norm": 0.07879115641117096,
+      "learning_rate": 4.9453690018345144e-06,
+      "logits/chosen": 14.362078666687012,
+      "logits/rejected": 14.708786010742188,
+      "logps/chosen": -0.27969443798065186,
+      "logps/rejected": -0.3294224143028259,
+      "loss": 0.9448,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": -0.41954168677330017,
+      "rewards/margins": 0.0745919868350029,
+      "rewards/rejected": -0.4941336512565613,
+      "step": 100
+    },
+    {
+      "epoch": 0.08880994671403197,
+      "eval_logits/chosen": 14.668691635131836,
+      "eval_logits/rejected": 14.53217601776123,
+      "eval_logps/chosen": -0.2837528884410858,
+      "eval_logps/rejected": -0.33105531334877014,
+      "eval_loss": 0.9382757544517517,
+      "eval_rewards/accuracies": 0.5164835453033447,
+      "eval_rewards/chosen": -0.4256293475627899,
+      "eval_rewards/margins": 0.07095365226268768,
+      "eval_rewards/rejected": -0.4965830445289612,
+      "eval_runtime": 26.9204,
+      "eval_samples_per_second": 27.043,
+      "eval_steps_per_second": 3.38,
+      "step": 100
+    },
+    {
+      "epoch": 0.09769094138543517,
+      "grad_norm": 0.0677201971411705,
+      "learning_rate": 4.933947257182901e-06,
+      "logits/chosen": 14.2677640914917,
+      "logits/rejected": 14.437828063964844,
+      "logps/chosen": -0.26914283633232117,
+      "logps/rejected": -0.3498677909374237,
+      "loss": 0.9307,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.40371423959732056,
+      "rewards/margins": 0.12108743190765381,
+      "rewards/rejected": -0.5248016715049744,
+      "step": 110
+    },
+    {
+      "epoch": 0.10657193605683836,
+      "grad_norm": 0.08429163694381714,
+      "learning_rate": 4.921457902821578e-06,
+      "logits/chosen": 13.978253364562988,
+      "logits/rejected": 14.673884391784668,
+      "logps/chosen": -0.2842218279838562,
+      "logps/rejected": -0.35421326756477356,
+      "loss": 0.9046,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.4263327121734619,
+      "rewards/margins": 0.10498716682195663,
+      "rewards/rejected": -0.5313198566436768,
+      "step": 120
+    },
+    {
+      "epoch": 0.11545293072824156,
+      "grad_norm": 0.08969741314649582,
+      "learning_rate": 4.907906416994146e-06,
+      "logits/chosen": 13.98906135559082,
+      "logits/rejected": 14.70555305480957,
+      "logps/chosen": -0.26557397842407227,
+      "logps/rejected": -0.3823702037334442,
+      "loss": 0.9151,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.3983609974384308,
+      "rewards/margins": 0.1751943826675415,
+      "rewards/rejected": -0.5735553503036499,
+      "step": 130
+    },
+    {
+      "epoch": 0.12433392539964476,
+      "grad_norm": 0.10470724105834961,
+      "learning_rate": 4.893298743830168e-06,
+      "logits/chosen": 13.948530197143555,
+      "logits/rejected": 14.77311897277832,
+      "logps/chosen": -0.2675064504146576,
+      "logps/rejected": -0.3631269335746765,
+      "loss": 0.9053,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.4012596607208252,
+      "rewards/margins": 0.14343078434467316,
+      "rewards/rejected": -0.5446904301643372,
+      "step": 140
+    },
+    {
+      "epoch": 0.13321492007104796,
+      "grad_norm": 0.09238290041685104,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": 12.939355850219727,
+      "logits/rejected": 13.387840270996094,
+      "logps/chosen": -0.27794820070266724,
+      "logps/rejected": -0.3520648777484894,
+      "loss": 0.9097,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.41692233085632324,
+      "rewards/margins": 0.11117497831583023,
+      "rewards/rejected": -0.5280972719192505,
+      "step": 150
+    },
+    {
+      "epoch": 0.13321492007104796,
+      "eval_logits/chosen": 13.123927116394043,
+      "eval_logits/rejected": 13.085403442382812,
+      "eval_logps/chosen": -0.28104570508003235,
+      "eval_logps/rejected": -0.3562033474445343,
+      "eval_loss": 0.9194123148918152,
+      "eval_rewards/accuracies": 0.5824176073074341,
+      "eval_rewards/chosen": -0.42156851291656494,
+      "eval_rewards/margins": 0.11273646354675293,
+      "eval_rewards/rejected": -0.5343050360679626,
+      "eval_runtime": 26.928,
+      "eval_samples_per_second": 27.035,
+      "eval_steps_per_second": 3.379,
+      "step": 150
+    },
+    {
+      "epoch": 0.14209591474245115,
+      "grad_norm": 0.14459910988807678,
+      "learning_rate": 4.860940925593703e-06,
+      "logits/chosen": 11.937938690185547,
+      "logits/rejected": 13.300163269042969,
+      "logps/chosen": -0.24840429425239563,
+      "logps/rejected": -0.4205872118473053,
+      "loss": 0.8929,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.37260642647743225,
+      "rewards/margins": 0.2582743763923645,
+      "rewards/rejected": -0.6308808326721191,
+      "step": 160
+    },
+    {
+      "epoch": 0.15097690941385436,
+      "grad_norm": 0.13749755918979645,
+      "learning_rate": 4.84320497372973e-06,
+      "logits/chosen": 12.549962997436523,
+      "logits/rejected": 12.858027458190918,
+      "logps/chosen": -0.2944340109825134,
+      "logps/rejected": -0.38480544090270996,
+      "loss": 0.895,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.44165101647377014,
+      "rewards/margins": 0.1355571448802948,
+      "rewards/rejected": -0.5772081613540649,
+      "step": 170
+    },
+    {
+      "epoch": 0.15985790408525755,
+      "grad_norm": 0.14530803263187408,
+      "learning_rate": 4.824441214720629e-06,
+      "logits/chosen": 10.892537117004395,
+      "logits/rejected": 11.756756782531738,
+      "logps/chosen": -0.2693363130092621,
+      "logps/rejected": -0.3874126374721527,
+      "loss": 0.8801,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.4040044844150543,
+      "rewards/margins": 0.17711447179317474,
+      "rewards/rejected": -0.5811189413070679,
+      "step": 180
+    },
+    {
+      "epoch": 0.16873889875666073,
+      "grad_norm": 0.20053793489933014,
+      "learning_rate": 4.804657878971252e-06,
+      "logits/chosen": 9.238119125366211,
+      "logits/rejected": 10.233763694763184,
+      "logps/chosen": -0.269479900598526,
+      "logps/rejected": -0.4149019122123718,
+      "loss": 0.8776,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.4042198657989502,
+      "rewards/margins": 0.21813304722309113,
+      "rewards/rejected": -0.6223528385162354,
+      "step": 190
+    },
+    {
+      "epoch": 0.17761989342806395,
+      "grad_norm": 0.20254959166049957,
+      "learning_rate": 4.783863644106502e-06,
+      "logits/chosen": 9.261235237121582,
+      "logits/rejected": 9.7451753616333,
+      "logps/chosen": -0.29611462354660034,
+      "logps/rejected": -0.44530659914016724,
+      "loss": 0.8596,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.4441719651222229,
+      "rewards/margins": 0.22378793358802795,
+      "rewards/rejected": -0.6679598689079285,
+      "step": 200
+    },
+    {
+      "epoch": 0.17761989342806395,
+      "eval_logits/chosen": 8.231317520141602,
+      "eval_logits/rejected": 8.285564422607422,
+      "eval_logps/chosen": -0.3201982080936432,
+      "eval_logps/rejected": -0.4862101078033447,
+      "eval_loss": 0.8602121472358704,
+      "eval_rewards/accuracies": 0.6153846383094788,
+      "eval_rewards/chosen": -0.480297327041626,
+      "eval_rewards/margins": 0.2490178644657135,
+      "eval_rewards/rejected": -0.7293152213096619,
+      "eval_runtime": 26.9054,
+      "eval_samples_per_second": 27.058,
+      "eval_steps_per_second": 3.382,
+      "step": 200
+    },
+    {
+      "epoch": 0.18650088809946713,
+      "grad_norm": 0.25219231843948364,
+      "learning_rate": 4.762067631165049e-06,
+      "logits/chosen": 7.822943687438965,
+      "logits/rejected": 8.847550392150879,
+      "logps/chosen": -0.3111446797847748,
+      "logps/rejected": -0.5253105163574219,
+      "loss": 0.8553,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -0.46671706438064575,
+      "rewards/margins": 0.3212486803531647,
+      "rewards/rejected": -0.787965714931488,
+      "step": 210
+    },
+    {
+      "epoch": 0.19538188277087035,
+      "grad_norm": 0.29052653908729553,
+      "learning_rate": 4.7392794005985324e-06,
+      "logits/chosen": 7.404467582702637,
+      "logits/rejected": 7.140469551086426,
+      "logps/chosen": -0.35531798005104065,
+      "logps/rejected": -0.4979163110256195,
+      "loss": 0.8204,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.5329769849777222,
+      "rewards/margins": 0.21389751136302948,
+      "rewards/rejected": -0.7468745112419128,
+      "step": 220
+    },
+    {
+      "epoch": 0.20426287744227353,
+      "grad_norm": 0.30769258737564087,
+      "learning_rate": 4.715508948078037e-06,
+      "logits/chosen": 5.096299648284912,
+      "logits/rejected": 5.143233299255371,
+      "logps/chosen": -0.3684440553188324,
+      "logps/rejected": -0.557721734046936,
+      "loss": 0.7884,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.5526660084724426,
+      "rewards/margins": 0.2839165925979614,
+      "rewards/rejected": -0.8365826606750488,
+      "step": 230
+    },
+    {
+      "epoch": 0.21314387211367672,
+      "grad_norm": 0.3165770471096039,
+      "learning_rate": 4.690766700109659e-06,
+      "logits/chosen": 5.376433372497559,
+      "logits/rejected": 4.738249778747559,
+      "logps/chosen": -0.4181212782859802,
+      "logps/rejected": -0.6782273054122925,
+      "loss": 0.7919,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.627181887626648,
+      "rewards/margins": 0.39015907049179077,
+      "rewards/rejected": -1.0173410177230835,
+      "step": 240
+    },
+    {
+      "epoch": 0.22202486678507993,
+      "grad_norm": 0.6854519844055176,
+      "learning_rate": 4.665063509461098e-06,
+      "logits/chosen": 3.3769917488098145,
+      "logits/rejected": 2.865615129470825,
+      "logps/chosen": -0.42344069480895996,
+      "logps/rejected": -0.8046137094497681,
+      "loss": 0.7456,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.6351610422134399,
+      "rewards/margins": 0.5717595219612122,
+      "rewards/rejected": -1.2069203853607178,
+      "step": 250
+    },
+    {
+      "epoch": 0.22202486678507993,
+      "eval_logits/chosen": 2.8929049968719482,
+      "eval_logits/rejected": 2.0500316619873047,
+      "eval_logps/chosen": -0.46059882640838623,
+      "eval_logps/rejected": -0.8680218458175659,
+      "eval_loss": 0.7499477863311768,
+      "eval_rewards/accuracies": 0.6373626589775085,
+      "eval_rewards/chosen": -0.6908981204032898,
+      "eval_rewards/margins": 0.6111345887184143,
+      "eval_rewards/rejected": -1.3020328283309937,
+      "eval_runtime": 26.9489,
+      "eval_samples_per_second": 27.014,
+      "eval_steps_per_second": 3.377,
+      "step": 250
+    },
+    {
+      "epoch": 0.23090586145648312,
+      "grad_norm": 1.4715200662612915,
+      "learning_rate": 4.638410650401267e-06,
+      "logits/chosen": 1.1702392101287842,
+      "logits/rejected": 0.6887636184692383,
+      "logps/chosen": -0.4449438154697418,
+      "logps/rejected": -1.0861380100250244,
+      "loss": 0.6853,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -0.6674157381057739,
+      "rewards/margins": 0.9617912173271179,
+      "rewards/rejected": -1.6292070150375366,
+      "step": 260
+    },
+    {
+      "epoch": 0.23978685612788633,
+      "grad_norm": 1.1773816347122192,
+      "learning_rate": 4.610819813755038e-06,
+      "logits/chosen": 2.005816698074341,
+      "logits/rejected": 0.7508751749992371,
+      "logps/chosen": -0.520461916923523,
+      "logps/rejected": -1.2372539043426514,
+      "loss": 0.6514,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.7806928157806396,
+      "rewards/margins": 1.0751880407333374,
+      "rewards/rejected": -1.8558809757232666,
+      "step": 270
+    },
+    {
+      "epoch": 0.24866785079928952,
+      "grad_norm": 0.4426538348197937,
+      "learning_rate": 4.582303101775249e-06,
+      "logits/chosen": 1.3614282608032227,
+      "logits/rejected": 0.17441503703594208,
+      "logps/chosen": -0.628559410572052,
+      "logps/rejected": -1.3841784000396729,
+      "loss": 0.6447,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -0.9428391456604004,
+      "rewards/margins": 1.1334283351898193,
+      "rewards/rejected": -2.076267719268799,
+      "step": 280
+    },
+    {
+      "epoch": 0.25754884547069273,
+      "grad_norm": 0.781704843044281,
+      "learning_rate": 4.55287302283426e-06,
+      "logits/chosen": 2.0829949378967285,
+      "logits/rejected": 1.0815263986587524,
+      "logps/chosen": -0.6394428014755249,
+      "logps/rejected": -1.8771930932998657,
+      "loss": 0.5863,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.9591643214225769,
+      "rewards/margins": 1.8566251993179321,
+      "rewards/rejected": -2.8157896995544434,
+      "step": 290
+    },
+    {
+      "epoch": 0.2664298401420959,
+      "grad_norm": 1.1327613592147827,
+      "learning_rate": 4.522542485937369e-06,
+      "logits/chosen": 1.197505235671997,
+      "logits/rejected": 0.16260084509849548,
+      "logps/chosen": -0.6966903805732727,
+      "logps/rejected": -2.2350592613220215,
+      "loss": 0.5354,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.0450356006622314,
+      "rewards/margins": 2.30755352973938,
+      "rewards/rejected": -3.3525891304016113,
+      "step": 300
+    },
+    {
+      "epoch": 0.2664298401420959,
+      "eval_logits/chosen": 1.2078615427017212,
+      "eval_logits/rejected": 0.2717524468898773,
+      "eval_logps/chosen": -0.7531170845031738,
+      "eval_logps/rejected": -2.0594394207000732,
+      "eval_loss": 0.5640697479248047,
+      "eval_rewards/accuracies": 0.6703296899795532,
+      "eval_rewards/chosen": -1.1296755075454712,
+      "eval_rewards/margins": 1.9594837427139282,
+      "eval_rewards/rejected": -3.0891592502593994,
+      "eval_runtime": 26.9506,
+      "eval_samples_per_second": 27.012,
+      "eval_steps_per_second": 3.377,
+      "step": 300
+    },
+    {
+      "epoch": 0.2753108348134991,
+      "grad_norm": 0.8920393586158752,
+      "learning_rate": 4.491324795060491e-06,
+      "logits/chosen": 1.12120521068573,
+      "logits/rejected": 0.3509993851184845,
+      "logps/chosen": -0.7844308018684387,
+      "logps/rejected": -2.2886409759521484,
+      "loss": 0.5357,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -1.176646113395691,
+      "rewards/margins": 2.256315231323242,
+      "rewards/rejected": -3.4329617023468018,
+      "step": 310
+    },
+    {
+      "epoch": 0.2841918294849023,
+      "grad_norm": 0.6658002734184265,
+      "learning_rate": 4.4592336433146e-06,
+      "logits/chosen": 1.750349760055542,
+      "logits/rejected": 0.7707412242889404,
+      "logps/chosen": -0.7995473146438599,
+      "logps/rejected": -2.2184109687805176,
+      "loss": 0.5412,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -1.199320912361145,
+      "rewards/margins": 2.128295421600342,
+      "rewards/rejected": -3.3276162147521973,
+      "step": 320
+    },
+    {
+      "epoch": 0.29307282415630553,
+      "grad_norm": 1.2708265781402588,
+      "learning_rate": 4.426283106939474e-06,
+      "logits/chosen": 1.8776671886444092,
+      "logits/rejected": 1.2588229179382324,
+      "logps/chosen": -0.894809901714325,
+      "logps/rejected": -2.4718618392944336,
+      "loss": 0.5502,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.342214822769165,
+      "rewards/margins": 2.3655781745910645,
+      "rewards/rejected": -3.7077927589416504,
+      "step": 330
+    },
+    {
+      "epoch": 0.3019538188277087,
+      "grad_norm": 1.2345410585403442,
+      "learning_rate": 4.3924876391293915e-06,
+      "logits/chosen": 1.9461545944213867,
+      "logits/rejected": 0.9410673975944519,
+      "logps/chosen": -0.891791045665741,
+      "logps/rejected": -2.2217679023742676,
+      "loss": 0.5519,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -1.3376867771148682,
+      "rewards/margins": 1.9949653148651123,
+      "rewards/rejected": -3.3326523303985596,
+      "step": 340
+    },
+    {
+      "epoch": 0.3108348134991119,
+      "grad_norm": 11.159003257751465,
+      "learning_rate": 4.357862063693486e-06,
+      "logits/chosen": 2.396707534790039,
+      "logits/rejected": 1.4882147312164307,
+      "logps/chosen": -1.1443694829940796,
+      "logps/rejected": -2.867492914199829,
+      "loss": 0.4806,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.7165542840957642,
+      "rewards/margins": 2.5846848487854004,
+      "rewards/rejected": -4.301239490509033,
+      "step": 350
+    },
+    {
+      "epoch": 0.3108348134991119,
+      "eval_logits/chosen": 1.7898317575454712,
+      "eval_logits/rejected": 1.109007477760315,
+      "eval_logps/chosen": -1.3312609195709229,
+      "eval_logps/rejected": -3.0454280376434326,
+      "eval_loss": 0.4698469638824463,
+      "eval_rewards/accuracies": 0.8131868243217468,
+      "eval_rewards/chosen": -1.9968912601470947,
+      "eval_rewards/margins": 2.5712504386901855,
+      "eval_rewards/rejected": -4.568141937255859,
+      "eval_runtime": 26.9108,
+      "eval_samples_per_second": 27.052,
+      "eval_steps_per_second": 3.382,
+      "step": 350
+    },
+    {
+      "epoch": 0.3197158081705151,
+      "grad_norm": 2.2653048038482666,
+      "learning_rate": 4.322421568553529e-06,
+      "logits/chosen": 0.36641037464141846,
+      "logits/rejected": -0.4621815085411072,
+      "logps/chosen": -1.5972144603729248,
+      "logps/rejected": -3.1685004234313965,
+      "loss": 0.472,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -2.3958218097686768,
+      "rewards/margins": 2.356929063796997,
+      "rewards/rejected": -4.752750396728516,
+      "step": 360
+    },
+    {
+      "epoch": 0.3285968028419183,
+      "grad_norm": 2.6831283569335938,
+      "learning_rate": 4.286181699082008e-06,
+      "logits/chosen": 1.18975830078125,
+      "logits/rejected": 0.7668083310127258,
+      "logps/chosen": -2.1462299823760986,
+      "logps/rejected": -3.7489593029022217,
+      "loss": 0.4169,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -3.2193446159362793,
+      "rewards/margins": 2.4040939807891846,
+      "rewards/rejected": -5.623438835144043,
+      "step": 370
+    },
+    {
+      "epoch": 0.33747779751332146,
+      "grad_norm": 3.5498828887939453,
+      "learning_rate": 4.249158351283414e-06,
+      "logits/chosen": 1.8030259609222412,
+      "logits/rejected": 1.4276998043060303,
+      "logps/chosen": -2.189352512359619,
+      "logps/rejected": -3.8717925548553467,
+      "loss": 0.4156,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -3.2840285301208496,
+      "rewards/margins": 2.523660182952881,
+      "rewards/rejected": -5.807689189910889,
+      "step": 380
+    },
+    {
+      "epoch": 0.3463587921847247,
+      "grad_norm": 1.46839439868927,
+      "learning_rate": 4.211367764821722e-06,
+      "logits/chosen": 2.2764127254486084,
+      "logits/rejected": 1.9199844598770142,
+      "logps/chosen": -2.4463610649108887,
+      "logps/rejected": -4.317194938659668,
+      "loss": 0.4085,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -3.669541597366333,
+      "rewards/margins": 2.806250810623169,
+      "rewards/rejected": -6.475791931152344,
+      "step": 390
+    },
+    {
+      "epoch": 0.3552397868561279,
+      "grad_norm": 1.6744616031646729,
+      "learning_rate": 4.172826515897146e-06,
+      "logits/chosen": 2.0364136695861816,
+      "logits/rejected": 1.4645214080810547,
+      "logps/chosen": -2.5738539695739746,
+      "logps/rejected": -4.41655158996582,
+      "loss": 0.3723,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -3.8607802391052246,
+      "rewards/margins": 2.764047145843506,
+      "rewards/rejected": -6.6248273849487305,
+      "step": 400
+    },
+    {
+      "epoch": 0.3552397868561279,
+      "eval_logits/chosen": 2.0340981483459473,
+      "eval_logits/rejected": 1.376452922821045,
+      "eval_logps/chosen": -2.6443424224853516,
+      "eval_logps/rejected": -4.638274192810059,
+      "eval_loss": 0.38738909363746643,
+      "eval_rewards/accuracies": 0.901098906993866,
+      "eval_rewards/chosen": -3.9665138721466064,
+      "eval_rewards/margins": 2.9908969402313232,
+      "eval_rewards/rejected": -6.95741081237793,
+      "eval_runtime": 26.943,
+      "eval_samples_per_second": 27.02,
+      "eval_steps_per_second": 3.377,
+      "step": 400
+    },
+    {
+      "epoch": 0.3641207815275311,
+      "grad_norm": 24.103384017944336,
+      "learning_rate": 4.133551509975264e-06,
+      "logits/chosen": 1.3653905391693115,
+      "logits/rejected": 0.7307125329971313,
+      "logps/chosen": -2.450911045074463,
+      "logps/rejected": -4.609705924987793,
+      "loss": 0.4031,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -3.6763663291931152,
+      "rewards/margins": 3.2381927967071533,
+      "rewards/rejected": -6.914559364318848,
+      "step": 410
+    },
+    {
+      "epoch": 0.37300177619893427,
+      "grad_norm": 2.102792263031006,
+      "learning_rate": 4.093559974371725e-06,
+      "logits/chosen": 1.6425203084945679,
+      "logits/rejected": 0.9890750050544739,
+      "logps/chosen": -2.1248934268951416,
+      "logps/rejected": -3.8592541217803955,
+      "loss": 0.3609,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -3.187340497970581,
+      "rewards/margins": 2.601541042327881,
+      "rewards/rejected": -5.788880825042725,
+      "step": 420
+    },
+    {
+      "epoch": 0.38188277087033745,
+      "grad_norm": 2.970210552215576,
+      "learning_rate": 4.052869450695776e-06,
+      "logits/chosen": 2.258129358291626,
+      "logits/rejected": 1.7126191854476929,
+      "logps/chosen": -2.9822497367858887,
+      "logps/rejected": -4.995323657989502,
+      "loss": 0.3669,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -4.473374366760254,
+      "rewards/margins": 3.01961088180542,
+      "rewards/rejected": -7.492985725402832,
+      "step": 430
+    },
+    {
+      "epoch": 0.3907637655417407,
+      "grad_norm": 2.3307077884674072,
+      "learning_rate": 4.011497787155938e-06,
+      "logits/chosen": 1.5054914951324463,
+      "logits/rejected": 0.6257806420326233,
+      "logps/chosen": -3.2783126831054688,
+      "logps/rejected": -5.625805854797363,
+      "loss": 0.3428,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.917469024658203,
+      "rewards/margins": 3.521239757537842,
+      "rewards/rejected": -8.438708305358887,
+      "step": 440
+    },
+    {
+      "epoch": 0.3996447602131439,
+      "grad_norm": 3.2838215827941895,
+      "learning_rate": 3.969463130731183e-06,
+      "logits/chosen": 1.711268424987793,
+      "logits/rejected": 1.4157356023788452,
+      "logps/chosen": -3.0163674354553223,
+      "logps/rejected": -5.020781517028809,
+      "loss": 0.3589,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -4.524550437927246,
+      "rewards/margins": 3.006621837615967,
+      "rewards/rejected": -7.531172752380371,
+      "step": 450
+    },
+    {
+      "epoch": 0.3996447602131439,
+      "eval_logits/chosen": 2.209981679916382,
+      "eval_logits/rejected": 1.6256438493728638,
+      "eval_logps/chosen": -2.841517925262451,
+      "eval_logps/rejected": -5.154296875,
+      "eval_loss": 0.35665163397789,
+      "eval_rewards/accuracies": 0.9120879173278809,
+      "eval_rewards/chosen": -4.262276649475098,
+      "eval_rewards/margins": 3.4691689014434814,
+      "eval_rewards/rejected": -7.7314453125,
+      "eval_runtime": 26.9493,
+      "eval_samples_per_second": 27.014,
+      "eval_steps_per_second": 3.377,
+      "step": 450
+    },
+    {
+      "epoch": 0.40852575488454707,
+      "grad_norm": 3.4476094245910645,
+      "learning_rate": 3.92678391921108e-06,
+      "logits/chosen": 1.3581907749176025,
+      "logits/rejected": 0.8621677160263062,
+      "logps/chosen": -2.831627130508423,
+      "logps/rejected": -5.023026943206787,
+      "loss": 0.3338,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.247440338134766,
+      "rewards/margins": 3.287100315093994,
+      "rewards/rejected": -7.534541130065918,
+      "step": 460
+    },
+    {
+      "epoch": 0.41740674955595025,
+      "grad_norm": 2.6653027534484863,
+      "learning_rate": 3.88347887310836e-06,
+      "logits/chosen": 1.8693492412567139,
+      "logits/rejected": 1.605238676071167,
+      "logps/chosen": -2.7347512245178223,
+      "logps/rejected": -5.385977745056152,
+      "loss": 0.3348,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.1021270751953125,
+      "rewards/margins": 3.976839542388916,
+      "rewards/rejected": -8.07896614074707,
+      "step": 470
+    },
+    {
+      "epoch": 0.42628774422735344,
+      "grad_norm": 6.658248424530029,
+      "learning_rate": 3.839566987447492e-06,
+      "logits/chosen": 1.8902003765106201,
+      "logits/rejected": 1.4942703247070312,
+      "logps/chosen": -3.1413326263427734,
+      "logps/rejected": -5.365391731262207,
+      "loss": 0.3171,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.71199893951416,
+      "rewards/margins": 3.3360886573791504,
+      "rewards/rejected": -8.048087120056152,
+      "step": 480
+    },
+    {
+      "epoch": 0.4351687388987567,
+      "grad_norm": 3.2576091289520264,
+      "learning_rate": 3.795067523432826e-06,
+      "logits/chosen": 2.301971673965454,
+      "logits/rejected": 1.5673840045928955,
+      "logps/chosen": -2.789170980453491,
+      "logps/rejected": -5.354689598083496,
+      "loss": 0.3238,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.1837568283081055,
+      "rewards/margins": 3.8482773303985596,
+      "rewards/rejected": -8.032033920288086,
+      "step": 490
+    },
+    {
+      "epoch": 0.44404973357015987,
+      "grad_norm": 5.094764709472656,
+      "learning_rate": 3.7500000000000005e-06,
+      "logits/chosen": 2.506347179412842,
+      "logits/rejected": 2.2316441535949707,
+      "logps/chosen": -3.1029722690582275,
+      "logps/rejected": -5.757994651794434,
+      "loss": 0.3052,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.654458522796631,
+      "rewards/margins": 3.9825336933135986,
+      "rewards/rejected": -8.636991500854492,
+      "step": 500
+    },
+    {
+      "epoch": 0.44404973357015987,
+      "eval_logits/chosen": 2.332798480987549,
+      "eval_logits/rejected": 1.8098194599151611,
+      "eval_logps/chosen": -3.269892692565918,
+      "eval_logps/rejected": -5.909496307373047,
+      "eval_loss": 0.324949711561203,
+      "eval_rewards/accuracies": 0.9120879173278809,
+      "eval_rewards/chosen": -4.904839038848877,
+      "eval_rewards/margins": 3.9594056606292725,
+      "eval_rewards/rejected": -8.86424446105957,
+      "eval_runtime": 26.9415,
+      "eval_samples_per_second": 27.021,
+      "eval_steps_per_second": 3.378,
+      "step": 500
+    },
+    {
+      "epoch": 0.45293072824156305,
+      "grad_norm": 4.973895072937012,
+      "learning_rate": 3.7043841852542884e-06,
+      "logits/chosen": 1.1408557891845703,
+      "logits/rejected": 0.9078874588012695,
+      "logps/chosen": -3.2813212871551514,
+      "logps/rejected": -5.6853814125061035,
+      "loss": 0.3185,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.9219818115234375,
+      "rewards/margins": 3.6060900688171387,
+      "rewards/rejected": -8.528071403503418,
+      "step": 510
+    },
+    {
+      "epoch": 0.46181172291296624,
+      "grad_norm": 2.726701021194458,
+      "learning_rate": 3.658240087799655e-06,
+      "logits/chosen": 1.5950994491577148,
+      "logits/rejected": 1.124607801437378,
+      "logps/chosen": -3.333927631378174,
+      "logps/rejected": -6.0137128829956055,
+      "loss": 0.3148,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.000891208648682,
+      "rewards/margins": 4.019678592681885,
+      "rewards/rejected": -9.020570755004883,
+      "step": 520
+    },
+    {
+      "epoch": 0.4706927175843694,
+      "grad_norm": 4.8865556716918945,
+      "learning_rate": 3.611587947962319e-06,
+      "logits/chosen": 2.257068157196045,
+      "logits/rejected": 1.7986671924591064,
+      "logps/chosen": -3.1947333812713623,
+      "logps/rejected": -6.262570381164551,
+      "loss": 0.3191,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.792099952697754,
+      "rewards/margins": 4.601754665374756,
+      "rewards/rejected": -9.393855094909668,
+      "step": 530
+    },
+    {
+      "epoch": 0.47957371225577267,
+      "grad_norm": 4.325016021728516,
+      "learning_rate": 3.564448228912682e-06,
+      "logits/chosen": 2.3096823692321777,
+      "logits/rejected": 1.7991451025009155,
+      "logps/chosen": -3.3641560077667236,
+      "logps/rejected": -6.142219543457031,
+      "loss": 0.276,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.046233654022217,
+      "rewards/margins": 4.167095184326172,
+      "rewards/rejected": -9.213329315185547,
+      "step": 540
+    },
+    {
+      "epoch": 0.48845470692717585,
+      "grad_norm": 3.15216326713562,
+      "learning_rate": 3.516841607689501e-06,
+      "logits/chosen": 1.4415160417556763,
+      "logits/rejected": 1.1298859119415283,
+      "logps/chosen": -2.851362466812134,
+      "logps/rejected": -6.210903167724609,
+      "loss": 0.3257,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.277044296264648,
+      "rewards/margins": 5.039311408996582,
+      "rewards/rejected": -9.316354751586914,
+      "step": 550
+    },
+    {
+      "epoch": 0.48845470692717585,
+      "eval_logits/chosen": 2.296088695526123,
+      "eval_logits/rejected": 1.8609188795089722,
+      "eval_logps/chosen": -3.422206401824951,
+      "eval_logps/rejected": -6.41662073135376,
+      "eval_loss": 0.30456680059432983,
+      "eval_rewards/accuracies": 0.9120879173278809,
+      "eval_rewards/chosen": -5.133309841156006,
+      "eval_rewards/margins": 4.491621971130371,
+      "eval_rewards/rejected": -9.624931335449219,
+      "eval_runtime": 26.9409,
+      "eval_samples_per_second": 27.022,
+      "eval_steps_per_second": 3.378,
+      "step": 550
+    },
+    {
+      "epoch": 0.49733570159857904,
+      "grad_norm": 3.456071615219116,
+      "learning_rate": 3.4687889661302577e-06,
+      "logits/chosen": 1.6153348684310913,
+      "logits/rejected": 1.1999633312225342,
+      "logps/chosen": -3.450087785720825,
+      "logps/rejected": -6.294098854064941,
+      "loss": 0.3094,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.175131797790527,
+      "rewards/margins": 4.266016960144043,
+      "rewards/rejected": -9.44114875793457,
+      "step": 560
+    },
+    {
+      "epoch": 0.5062166962699822,
+      "grad_norm": 3.7813830375671387,
+      "learning_rate": 3.4203113817116955e-06,
+      "logits/chosen": 2.5460948944091797,
+      "logits/rejected": 2.240358829498291,
+      "logps/chosen": -3.441761016845703,
+      "logps/rejected": -6.200081825256348,
+      "loss": 0.2967,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -5.1626410484313965,
+      "rewards/margins": 4.137481212615967,
+      "rewards/rejected": -9.300122261047363,
+      "step": 570
+    },
+    {
+      "epoch": 0.5150976909413855,
+      "grad_norm": 3.561509847640991,
+      "learning_rate": 3.3714301183045382e-06,
+      "logits/chosen": 2.236506223678589,
+      "logits/rejected": 2.0224223136901855,
+      "logps/chosen": -3.6373534202575684,
+      "logps/rejected": -6.733517646789551,
+      "loss": 0.2781,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.45603084564209,
+      "rewards/margins": 4.6442461013793945,
+      "rewards/rejected": -10.1002779006958,
+      "step": 580
+    },
+    {
+      "epoch": 0.5239786856127886,
+      "grad_norm": 2.540649175643921,
+      "learning_rate": 3.3221666168464584e-06,
+      "logits/chosen": 1.9122869968414307,
+      "logits/rejected": 1.8370224237442017,
+      "logps/chosen": -3.241629123687744,
+      "logps/rejected": -6.666001319885254,
+      "loss": 0.2678,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.862443447113037,
+      "rewards/margins": 5.1365580558776855,
+      "rewards/rejected": -9.999002456665039,
+      "step": 590
+    },
+    {
+      "epoch": 0.5328596802841918,
+      "grad_norm": 2.05529522895813,
+      "learning_rate": 3.272542485937369e-06,
+      "logits/chosen": 1.9268839359283447,
+      "logits/rejected": 1.4725837707519531,
+      "logps/chosen": -3.7067363262176514,
+      "logps/rejected": -6.972158908843994,
+      "loss": 0.264,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.560103893280029,
+      "rewards/margins": 4.898133754730225,
+      "rewards/rejected": -10.458237648010254,
+      "step": 600
+    },
+    {
+      "epoch": 0.5328596802841918,
+      "eval_logits/chosen": 2.3668525218963623,
+      "eval_logits/rejected": 1.9882280826568604,
+      "eval_logps/chosen": -3.6040353775024414,
+      "eval_logps/rejected": -6.91050386428833,
+      "eval_loss": 0.29209136962890625,
+      "eval_rewards/accuracies": 0.9120879173278809,
+      "eval_rewards/chosen": -5.406052589416504,
+      "eval_rewards/margins": 4.95970344543457,
+      "eval_rewards/rejected": -10.365756034851074,
+      "eval_runtime": 26.9486,
+      "eval_samples_per_second": 27.014,
+      "eval_steps_per_second": 3.377,
+      "step": 600
+    },
+    {
+      "epoch": 0.5417406749555951,
+      "grad_norm": 5.438882350921631,
+      "learning_rate": 3.222579492361179e-06,
+      "logits/chosen": 2.0183475017547607,
+      "logits/rejected": 1.7007204294204712,
+      "logps/chosen": -3.7757949829101562,
+      "logps/rejected": -6.875540256500244,
+      "loss": 0.2787,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.663692951202393,
+      "rewards/margins": 4.6496171951293945,
+      "rewards/rejected": -10.313310623168945,
+      "step": 610
+    },
+    {
+      "epoch": 0.5506216696269982,
+      "grad_norm": 6.464421272277832,
+      "learning_rate": 3.1722995515381644e-06,
+      "logits/chosen": 1.8897826671600342,
+      "logits/rejected": 1.6919243335723877,
+      "logps/chosen": -3.738862991333008,
+      "logps/rejected": -7.225765228271484,
+      "loss": 0.296,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.6082940101623535,
+      "rewards/margins": 5.230353355407715,
+      "rewards/rejected": -10.83864688873291,
+      "step": 620
+    },
+    {
+      "epoch": 0.5595026642984015,
+      "grad_norm": 3.983937978744507,
+      "learning_rate": 3.121724717912138e-06,
+      "logits/chosen": 1.8497775793075562,
+      "logits/rejected": 1.3537144660949707,
+      "logps/chosen": -3.6511616706848145,
+      "logps/rejected": -7.310211181640625,
+      "loss": 0.2604,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.476742267608643,
+      "rewards/margins": 5.488574028015137,
+      "rewards/rejected": -10.965316772460938,
+      "step": 630
+    },
+    {
+      "epoch": 0.5683836589698046,
+      "grad_norm": 9.850502014160156,
+      "learning_rate": 3.0708771752766397e-06,
+      "logits/chosen": 2.073599338531494,
+      "logits/rejected": 1.3638606071472168,
+      "logps/chosen": -4.1297831535339355,
+      "logps/rejected": -7.344725131988525,
+      "loss": 0.2868,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.194674491882324,
+      "rewards/margins": 4.822413444519043,
+      "rewards/rejected": -11.017088890075684,
+      "step": 640
+    },
+    {
+      "epoch": 0.5772646536412078,
+      "grad_norm": 3.2157247066497803,
+      "learning_rate": 3.019779227044398e-06,
+      "logits/chosen": 1.492789387702942,
+      "logits/rejected": 1.2012196779251099,
+      "logps/chosen": -3.458716630935669,
+      "logps/rejected": -6.776049613952637,
+      "loss": 0.2489,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.188075542449951,
+      "rewards/margins": 4.975998878479004,
+      "rewards/rejected": -10.164074897766113,
+      "step": 650
+    },
+    {
+      "epoch": 0.5772646536412078,
+      "eval_logits/chosen": 2.4555883407592773,
+      "eval_logits/rejected": 2.1057684421539307,
+      "eval_logps/chosen": -3.6452677249908447,
+      "eval_logps/rejected": -7.163515090942383,
+      "eval_loss": 0.2778012156486511,
+      "eval_rewards/accuracies": 0.901098906993866,
+      "eval_rewards/chosen": -5.467901706695557,
+      "eval_rewards/margins": 5.277371406555176,
+      "eval_rewards/rejected": -10.745272636413574,
+      "eval_runtime": 26.906,
+      "eval_samples_per_second": 27.057,
+      "eval_steps_per_second": 3.382,
+      "step": 650
+    },
+    {
+      "epoch": 0.5861456483126111,
+      "grad_norm": 3.6115524768829346,
+      "learning_rate": 2.9684532864643123e-06,
+      "logits/chosen": 2.066650390625,
+      "logits/rejected": 1.7030436992645264,
+      "logps/chosen": -3.875563859939575,
+      "logps/rejected": -7.133286476135254,
+      "loss": 0.2752,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.813345432281494,
+      "rewards/margins": 4.88658332824707,
+      "rewards/rejected": -10.699929237365723,
+      "step": 660
+    },
+    {
+      "epoch": 0.5950266429840142,
+      "grad_norm": 3.460318088531494,
+      "learning_rate": 2.9169218667902562e-06,
+      "logits/chosen": 2.433854341506958,
+      "logits/rejected": 2.045027256011963,
+      "logps/chosen": -3.6893928050994873,
+      "logps/rejected": -6.720344543457031,
+      "loss": 0.2562,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.534089088439941,
+      "rewards/margins": 4.5464277267456055,
+      "rewards/rejected": -10.080517768859863,
+      "step": 670
+    },
+    {
+      "epoch": 0.6039076376554174,
+      "grad_norm": 2.232542037963867,
+      "learning_rate": 2.8652075714060296e-06,
+      "logits/chosen": 2.3069093227386475,
+      "logits/rejected": 1.7321109771728516,
+      "logps/chosen": -3.936506748199463,
+      "logps/rejected": -7.7827630043029785,
+      "loss": 0.2333,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -5.904759883880615,
+      "rewards/margins": 5.769384860992432,
+      "rewards/rejected": -11.674144744873047,
+      "step": 680
+    },
+    {
+      "epoch": 0.6127886323268206,
+      "grad_norm": 3.947690010070801,
+      "learning_rate": 2.813333083910761e-06,
+      "logits/chosen": 1.467885136604309,
+      "logits/rejected": 1.0833615064620972,
+      "logps/chosen": -3.2870895862579346,
+      "logps/rejected": -7.22244930267334,
+      "loss": 0.2739,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.930634498596191,
+      "rewards/margins": 5.90303897857666,
+      "rewards/rejected": -10.833673477172852,
+      "step": 690
+    },
+    {
+      "epoch": 0.6216696269982238,
+      "grad_norm": 4.934174060821533,
+      "learning_rate": 2.761321158169134e-06,
+      "logits/chosen": 2.440274477005005,
+      "logits/rejected": 2.157214403152466,
+      "logps/chosen": -4.241659641265869,
+      "logps/rejected": -7.748563289642334,
+      "loss": 0.2859,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -6.362490177154541,
+      "rewards/margins": 5.260354995727539,
+      "rewards/rejected": -11.622844696044922,
+      "step": 700
+    },
+    {
+      "epoch": 0.6216696269982238,
+      "eval_logits/chosen": 2.5609161853790283,
+      "eval_logits/rejected": 2.258920431137085,
+      "eval_logps/chosen": -3.965893507003784,
+      "eval_logps/rejected": -7.77095365524292,
+      "eval_loss": 0.26121068000793457,
+      "eval_rewards/accuracies": 0.9340659379959106,
+      "eval_rewards/chosen": -5.948840618133545,
+      "eval_rewards/margins": 5.707590579986572,
+      "eval_rewards/rejected": -11.656431198120117,
+      "eval_runtime": 26.9101,
+      "eval_samples_per_second": 27.053,
+      "eval_steps_per_second": 3.382,
+      "step": 700
+    },
+    {
+      "epoch": 0.6305506216696269,
+      "grad_norm": 5.9008684158325195,
+      "learning_rate": 2.70919460833079e-06,
+      "logits/chosen": 1.6100571155548096,
+      "logits/rejected": 1.341528296470642,
+      "logps/chosen": -3.8634166717529297,
+      "logps/rejected": -7.63654088973999,
+      "loss": 0.2351,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.795124053955078,
+      "rewards/margins": 5.659686088562012,
+      "rewards/rejected": -11.454811096191406,
+      "step": 710
+    },
+    {
+      "epoch": 0.6394316163410302,
+      "grad_norm": 2.7195751667022705,
+      "learning_rate": 2.6569762988232838e-06,
+      "logits/chosen": 2.1212353706359863,
+      "logits/rejected": 1.864898443222046,
+      "logps/chosen": -3.8487918376922607,
+      "logps/rejected": -7.823777198791504,
+      "loss": 0.2329,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.77318811416626,
+      "rewards/margins": 5.9624762535095215,
+      "rewards/rejected": -11.735665321350098,
+      "step": 720
+    },
+    {
+      "epoch": 0.6483126110124334,
+      "grad_norm": 4.423059463500977,
+      "learning_rate": 2.604689134322999e-06,
+      "logits/chosen": 2.0979533195495605,
+      "logits/rejected": 1.7213973999023438,
+      "logps/chosen": -3.9874777793884277,
+      "logps/rejected": -7.617144584655762,
+      "loss": 0.2482,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.981216907501221,
+      "rewards/margins": 5.4444990158081055,
+      "rewards/rejected": -11.4257173538208,
+      "step": 730
+    },
+    {
+      "epoch": 0.6571936056838366,
+      "grad_norm": 4.327184200286865,
+      "learning_rate": 2.5523560497083927e-06,
+      "logits/chosen": 2.2398109436035156,
+      "logits/rejected": 1.913095474243164,
+      "logps/chosen": -3.72918701171875,
+      "logps/rejected": -7.094988822937012,
+      "loss": 0.223,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.593780517578125,
+      "rewards/margins": 5.048702716827393,
+      "rewards/rejected": -10.64248275756836,
+      "step": 740
+    },
+    {
+      "epoch": 0.6660746003552398,
+      "grad_norm": 3.1207029819488525,
+      "learning_rate": 2.5e-06,
+      "logits/chosen": 1.7178529500961304,
+      "logits/rejected": 1.7194137573242188,
+      "logps/chosen": -3.9564640522003174,
+      "logps/rejected": -7.360040187835693,
+      "loss": 0.2227,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.934695243835449,
+      "rewards/margins": 5.10536527633667,
+      "rewards/rejected": -11.040060043334961,
+      "step": 750
+    },
+    {
+      "epoch": 0.6660746003552398,
+      "eval_logits/chosen": 2.521843671798706,
+      "eval_logits/rejected": 2.2902743816375732,
+      "eval_logps/chosen": -4.141489028930664,
+      "eval_logps/rejected": -8.121171951293945,
+      "eval_loss": 0.25929248332977295,
+      "eval_rewards/accuracies": 0.9120879173278809,
+      "eval_rewards/chosen": -6.212233543395996,
+      "eval_rewards/margins": 5.969525337219238,
+      "eval_rewards/rejected": -12.181758880615234,
+      "eval_runtime": 26.9435,
+      "eval_samples_per_second": 27.019,
+      "eval_steps_per_second": 3.377,
+      "step": 750
+    },
+    {
+      "epoch": 0.6749555950266429,
+      "grad_norm": 2.5788235664367676,
+      "learning_rate": 2.447643950291608e-06,
+      "logits/chosen": 1.7696739435195923,
+      "logits/rejected": 1.624455213546753,
+      "logps/chosen": -4.3122711181640625,
+      "logps/rejected": -8.191080093383789,
+      "loss": 0.1991,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.468405723571777,
+      "rewards/margins": 5.81821346282959,
+      "rewards/rejected": -12.286620140075684,
+      "step": 760
+    },
+    {
+      "epoch": 0.6838365896980462,
+      "grad_norm": 4.544955730438232,
+      "learning_rate": 2.3953108656770018e-06,
+      "logits/chosen": 1.9560623168945312,
+      "logits/rejected": 1.7740411758422852,
+      "logps/chosen": -4.327980041503906,
+      "logps/rejected": -8.547250747680664,
+      "loss": 0.2301,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.491971492767334,
+      "rewards/margins": 6.328906059265137,
+      "rewards/rejected": -12.82087516784668,
+      "step": 770
+    },
+    {
+      "epoch": 0.6927175843694494,
+      "grad_norm": 3.706414222717285,
+      "learning_rate": 2.3430237011767166e-06,
+      "logits/chosen": 2.664365291595459,
+      "logits/rejected": 2.338026285171509,
+      "logps/chosen": -3.8961780071258545,
+      "logps/rejected": -7.658777713775635,
+      "loss": 0.2398,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.84426736831665,
+      "rewards/margins": 5.643899917602539,
+      "rewards/rejected": -11.488167762756348,
+      "step": 780
+    },
+    {
+      "epoch": 0.7015985790408525,
+      "grad_norm": 1.7868493795394897,
+      "learning_rate": 2.290805391669212e-06,
+      "logits/chosen": 3.1498961448669434,
+      "logits/rejected": 2.860450267791748,
+      "logps/chosen": -4.245728492736816,
+      "logps/rejected": -8.142782211303711,
+      "loss": 0.184,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -6.368593215942383,
+      "rewards/margins": 5.845582008361816,
+      "rewards/rejected": -12.214174270629883,
+      "step": 790
+    },
+    {
+      "epoch": 0.7104795737122558,
+      "grad_norm": 4.593838214874268,
+      "learning_rate": 2.238678841830867e-06,
+      "logits/chosen": 2.4697837829589844,
+      "logits/rejected": 2.062006950378418,
+      "logps/chosen": -4.312434673309326,
+      "logps/rejected": -7.9613471031188965,
+      "loss": 0.2373,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.468652248382568,
+      "rewards/margins": 5.4733686447143555,
+      "rewards/rejected": -11.942021369934082,
+      "step": 800
+    },
+    {
+      "epoch": 0.7104795737122558,
+      "eval_logits/chosen": 2.5936477184295654,
+      "eval_logits/rejected": 2.3626792430877686,
+      "eval_logps/chosen": -3.92742919921875,
+      "eval_logps/rejected": -8.105939865112305,
+      "eval_loss": 0.24979564547538757,
+      "eval_rewards/accuracies": 0.9230769276618958,
+      "eval_rewards/chosen": -5.891143321990967,
+      "eval_rewards/margins": 6.267765045166016,
+      "eval_rewards/rejected": -12.15890884399414,
+      "eval_runtime": 26.9514,
+      "eval_samples_per_second": 27.012,
+      "eval_steps_per_second": 3.376,
+      "step": 800
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 1500,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 50,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.94866406062293e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-800/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71bc9387fe632b190c857fa4052c3667f7da2a0079c5f2572bee765dfa764ce3
+size 7224

checkpoint-800/zero_to_fp32.py

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