Upload folder using huggingface_hub

README.md

@@ -0,0 +1,202 @@
+---
+base_model: meta-llama/Llama-3.2-11B-Vision-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 0.15.2

adapter_config.json

@@ -0,0 +1,31 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "/data1/chenjiazun/LLM/Llama-3.2-11B-Vision-Instruct",
+  "bias": "none",
+  "corda_config": null,
+  "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": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": [],
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": "^(language_model).*\\.(down_proj|k_proj|q_proj|o_proj|up_proj|gate_proj|v_proj)$",
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2fb289ad31dc4f00f54d338eb2e57db970469609ae7e53227e2b12d5b3f6f843
+size 52511776

additional_config.json

@@ -0,0 +1 @@
+{"lora_dtype": null, "lorap_lr_ratio": null, "lorap_emb_lr": 1e-06}

args.json

@@ -0,0 +1,438 @@
+{
+  "model": "/data1/chenjiazun/LLM/Llama-3.2-11B-Vision-Instruct",
+  "model_type": "llama3_2_vision",
+  "model_revision": null,
+  "task_type": "causal_lm",
+  "torch_dtype": "bfloat16",
+  "attn_impl": null,
+  "num_labels": null,
+  "problem_type": null,
+  "rope_scaling": null,
+  "device_map": null,
+  "max_memory": {},
+  "local_repo_path": null,
+  "template": "llama3_2_vision",
+  "system": "###Task You are an expert at converting visual questions into effective search queries. \nYour goal is to create a comprehensive search query that will help find the most relevant information. \nFor each image-based question, you must create a search query that combines: \n1. Key visual elements from the image (objects, text, logos, scenes, actions, etc.) \n2. The core question being asked \n3. Potential answer terms or relevant context \nFor example: \n- If asking about a logo: include company name, industry, and visual description \n- If asking about an object: include its appearance, category, and possible brands/models \n- If asking about an event/scene: include location hints, activities, and time period clues",
+  "max_length": 4096,
+  "truncation_strategy": "delete",
+  "max_pixels": null,
+  "agent_template": null,
+  "norm_bbox": null,
+  "response_prefix": null,
+  "padding_side": "right",
+  "loss_scale": "last_round",
+  "sequence_parallel_size": 1,
+  "use_chat_template": true,
+  "template_backend": "swift",
+  "dataset": [
+    "task3_dpo_gt6_task2_gt30.3+task3gt5_0.1vp.json"
+  ],
+  "val_dataset": [],
+  "split_dataset_ratio": 0.01,
+  "data_seed": 42,
+  "dataset_num_proc": 4,
+  "dataset_shuffle": true,
+  "val_dataset_shuffle": false,
+  "streaming": false,
+  "interleave_prob": null,
+  "stopping_strategy": "first_exhausted",
+  "shuffle_buffer_size": 1000,
+  "enable_cache": false,
+  "download_mode": "reuse_dataset_if_exists",
+  "columns": {},
+  "strict": false,
+  "remove_unused_columns": true,
+  "model_name": [
+    null,
+    null
+  ],
+  "model_author": [
+    null,
+    null
+  ],
+  "custom_dataset_info": [],
+  "quant_method": null,
+  "quant_bits": null,
+  "hqq_axis": null,
+  "bnb_4bit_compute_dtype": "bfloat16",
+  "bnb_4bit_quant_type": "nf4",
+  "bnb_4bit_use_double_quant": true,
+  "bnb_4bit_quant_storage": null,
+  "max_new_tokens": 64,
+  "temperature": 0.9,
+  "top_k": 50,
+  "top_p": 0.9,
+  "repetition_penalty": 1.0,
+  "num_beams": 1,
+  "stream": false,
+  "stop_words": [],
+  "logprobs": false,
+  "top_logprobs": null,
+  "ckpt_dir": null,
+  "load_dataset_config": null,
+  "lora_modules": [],
+  "tuner_backend": "peft",
+  "train_type": "lora",
+  "adapters": [],
+  "external_plugins": [],
+  "seed": 42,
+  "model_kwargs": {},
+  "load_args": false,
+  "load_data_args": false,
+  "use_hf": false,
+  "hub_token": null,
+  "custom_register_path": [],
+  "ignore_args_error": false,
+  "use_swift_lora": false,
+  "output_dir": "/data2/kdd/crag/cjz/output/task3_dpo_gt6_task2_gt30.3+task3gt5_0.1vp/v0-20250611-011957",
+  "overwrite_output_dir": false,
+  "do_train": false,
+  "do_eval": false,
+  "do_predict": false,
+  "eval_strategy": "steps",
+  "prediction_loss_only": false,
+  "per_device_train_batch_size": 1,
+  "per_device_eval_batch_size": 1,
+  "per_gpu_train_batch_size": null,
+  "per_gpu_eval_batch_size": null,
+  "gradient_accumulation_steps": 32,
+  "eval_accumulation_steps": null,
+  "eval_delay": 0,
+  "torch_empty_cache_steps": null,
+  "learning_rate": 0.0002,
+  "weight_decay": 0.1,
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.95,
+  "adam_epsilon": 1e-08,
+  "max_grad_norm": 1.0,
+  "num_train_epochs": 1.0,
+  "max_steps": -1,
+  "lr_scheduler_type": "cosine",
+  "lr_scheduler_kwargs": null,
+  "warmup_ratio": 0.05,
+  "warmup_steps": 0,
+  "log_level": "passive",
+  "log_level_replica": "warning",
+  "log_on_each_node": true,
+  "logging_dir": "/data2/kdd/crag/cjz/output/task3_dpo_gt6_task2_gt30.3+task3gt5_0.1vp/v0-20250611-011957/runs",
+  "logging_strategy": "steps",
+  "logging_first_step": true,
+  "logging_steps": 1,
+  "logging_nan_inf_filter": true,
+  "save_strategy": "steps",
+  "save_steps": 50.0,
+  "save_total_limit": null,
+  "save_safetensors": true,
+  "save_on_each_node": false,
+  "save_only_model": false,
+  "restore_callback_states_from_checkpoint": false,
+  "no_cuda": false,
+  "use_cpu": false,
+  "use_mps_device": false,
+  "jit_mode_eval": false,
+  "use_ipex": false,
+  "bf16": true,
+  "fp16": false,
+  "fp16_opt_level": "O1",
+  "half_precision_backend": "auto",
+  "bf16_full_eval": false,
+  "fp16_full_eval": false,
+  "tf32": null,
+  "local_rank": 0,
+  "ddp_backend": null,
+  "tpu_num_cores": null,
+  "tpu_metrics_debug": false,
+  "debug": null,
+  "dataloader_drop_last": false,
+  "eval_steps": 300.0,
+  "dataloader_num_workers": 4,
+  "dataloader_prefetch_factor": null,
+  "past_index": -1,
+  "run_name": null,
+  "disable_tqdm": null,
+  "label_names": null,
+  "load_best_model_at_end": false,
+  "metric_for_best_model": "loss",
+  "greater_is_better": false,
+  "ignore_data_skip": false,
+  "fsdp": "",
+  "fsdp_min_num_params": 0,
+  "fsdp_config": null,
+  "tp_size": 0,
+  "fsdp_transformer_layer_cls_to_wrap": null,
+  "accelerator_config": {
+    "dispatch_batches": false
+  },
+  "deepspeed": {
+    "fp16": {
+      "enabled": "auto",
+      "loss_scale": 0,
+      "loss_scale_window": 1000,
+      "initial_scale_power": 16,
+      "hysteresis": 2,
+      "min_loss_scale": 1
+    },
+    "bf16": {
+      "enabled": "auto"
+    },
+    "zero_optimization": {
+      "stage": 2,
+      "offload_optimizer": {
+        "device": "none",
+        "pin_memory": true
+      },
+      "allgather_partitions": true,
+      "allgather_bucket_size": 200000000.0,
+      "overlap_comm": false,
+      "reduce_scatter": true,
+      "reduce_bucket_size": 200000000.0,
+      "contiguous_gradients": true
+    },
+    "gradient_accumulation_steps": "auto",
+    "gradient_clipping": "auto",
+    "steps_per_print": 2000,
+    "train_batch_size": "auto",
+    "train_micro_batch_size_per_gpu": "auto",
+    "wall_clock_breakdown": false
+  },
+  "label_smoothing_factor": 0.0,
+  "optim": "adamw_torch",
+  "optim_args": null,
+  "adafactor": false,
+  "group_by_length": false,
+  "length_column_name": "length",
+  "report_to": [
+    "tensorboard"
+  ],
+  "ddp_find_unused_parameters": null,
+  "ddp_bucket_cap_mb": null,
+  "ddp_broadcast_buffers": null,
+  "dataloader_pin_memory": true,
+  "dataloader_persistent_workers": false,
+  "skip_memory_metrics": true,
+  "use_legacy_prediction_loop": false,
+  "push_to_hub": false,
+  "resume_from_checkpoint": null,
+  "hub_model_id": null,
+  "hub_strategy": "every_save",
+  "hub_private_repo": null,
+  "hub_always_push": false,
+  "gradient_checkpointing": true,
+  "gradient_checkpointing_kwargs": null,
+  "include_inputs_for_metrics": false,
+  "include_for_metrics": [],
+  "eval_do_concat_batches": true,
+  "fp16_backend": "auto",
+  "push_to_hub_model_id": null,
+  "push_to_hub_organization": null,
+  "push_to_hub_token": null,
+  "mp_parameters": "",
+  "auto_find_batch_size": false,
+  "full_determinism": false,
+  "torchdynamo": null,
+  "ray_scope": "last",
+  "ddp_timeout": 1800,
+  "torch_compile": false,
+  "torch_compile_backend": null,
+  "torch_compile_mode": null,
+  "include_tokens_per_second": false,
+  "include_num_input_tokens_seen": false,
+  "neftune_noise_alpha": null,
+  "optim_target_modules": null,
+  "batch_eval_metrics": false,
+  "eval_on_start": false,
+  "use_liger_kernel": false,
+  "eval_use_gather_object": false,
+  "average_tokens_across_devices": false,
+  "sortish_sampler": false,
+  "predict_with_generate": false,
+  "generation_max_length": null,
+  "generation_num_beams": null,
+  "generation_config": null,
+  "check_model": true,
+  "acc_strategy": "token",
+  "train_dataloader_shuffle": true,
+  "metric_warmup_step": 0,
+  "fsdp_num": 1,
+  "acc_steps": 1,
+  "eval_use_evalscope": false,
+  "eval_datasets": [],
+  "eval_limit": null,
+  "eval_datasets_args": null,
+  "eval_generation_config": null,
+  "freeze_parameters": [
+    "vision_model",
+    "multi_modal_projector"
+  ],
+  "freeze_parameters_ratio": 0.0,
+  "trainable_parameters": [],
+  "freeze_llm": false,
+  "freeze_vit": true,
+  "freeze_aligner": true,
+  "target_modules": [
+    "all-linear"
+  ],
+  "target_regex": null,
+  "modules_to_save": [],
+  "lora_rank": 8,
+  "lora_alpha": 32,
+  "lora_dropout": 0.05,
+  "lora_bias": "none",
+  "lora_dtype": null,
+  "lorap_lr_ratio": null,
+  "use_rslora": false,
+  "use_dora": false,
+  "lora_ga_batch_size": 2,
+  "lora_ga_iters": 2,
+  "lora_ga_max_length": 1024,
+  "lora_ga_direction": "ArB2r",
+  "lora_ga_scale": "stable",
+  "lora_ga_stable_gamma": 16,
+  "init_weights": true,
+  "fourier_n_frequency": 2000,
+  "fourier_scaling": 300.0,
+  "boft_block_size": 4,
+  "boft_block_num": 0,
+  "boft_n_butterfly_factor": 1,
+  "boft_dropout": 0.0,
+  "vera_rank": 256,
+  "vera_projection_prng_key": 0,
+  "vera_dropout": 0.0,
+  "vera_d_initial": 0.1,
+  "adapter_act": "gelu",
+  "adapter_length": 128,
+  "use_galore": false,
+  "galore_target_modules": null,
+  "galore_rank": 128,
+  "galore_update_proj_gap": 50,
+  "galore_scale": 1.0,
+  "galore_proj_type": "std",
+  "galore_optim_per_parameter": false,
+  "galore_with_embedding": false,
+  "galore_quantization": false,
+  "galore_proj_quant": false,
+  "galore_proj_bits": 4,
+  "galore_proj_group_size": 256,
+  "galore_cos_threshold": 0.4,
+  "galore_gamma_proj": 2,
+  "galore_queue_size": 5,
+  "adalora_target_r": 8,
+  "adalora_init_r": 12,
+  "adalora_tinit": 0,
+  "adalora_tfinal": 0,
+  "adalora_deltaT": 1,
+  "adalora_beta1": 0.85,
+  "adalora_beta2": 0.85,
+  "adalora_orth_reg_weight": 0.5,
+  "llamapro_num_new_blocks": 4,
+  "llamapro_num_groups": null,
+  "lisa_activated_layers": 0,
+  "lisa_step_interval": 20,
+  "reft_layer_key": null,
+  "reft_layers": null,
+  "reft_rank": 4,
+  "reft_intervention_type": "LoreftIntervention",
+  "reft_args": null,
+  "swanlab_token": null,
+  "swanlab_project": null,
+  "swanlab_workspace": null,
+  "swanlab_exp_name": null,
+  "swanlab_mode": "cloud",
+  "add_version": true,
+  "resume_only_model": false,
+  "create_checkpoint_symlink": false,
+  "packing": false,
+  "lazy_tokenize": true,
+  "loss_type": "sigmoid",
+  "optimizer": null,
+  "metric": null,
+  "zero_hpz_partition_size": null,
+  "reward_model": null,
+  "reward_adapters": [],
+  "reward_model_type": null,
+  "reward_model_revision": null,
+  "num_ppo_epochs": 4,
+  "whiten_rewards": false,
+  "kl_coef": 0.05,
+  "cliprange": 0.2,
+  "vf_coef": 0.1,
+  "cliprange_value": 0.2,
+  "gamma": 1.0,
+  "lam": 0.95,
+  "num_mini_batches": 1,
+  "local_rollout_forward_batch_size": 64,
+  "num_sample_generations": 10,
+  "response_length": 512,
+  "missing_eos_penalty": null,
+  "epsilon": 0.2,
+  "epsilon_high": null,
+  "num_infer_workers": 1,
+  "vllm_device": [
+    "auto"
+  ],
+  "vllm_gpu_memory_utilization": 0.9,
+  "vllm_max_model_len": null,
+  "vllm_max_num_seqs": 256,
+  "vllm_enforce_eager": false,
+  "vllm_limit_mm_per_prompt": null,
+  "vllm_enable_prefix_caching": true,
+  "cosine_min_len_value_wrong": -0.5,
+  "cosine_max_len_value_wrong": 0.0,
+  "cosine_min_len_value_correct": 1.0,
+  "cosine_max_len_value_correct": 0.5,
+  "cosine_max_len": null,
+  "repetition_n_grams": 3,
+  "repetition_max_penalty": -1.0,
+  "use_lmdeploy": false,
+  "lmdeploy_device": "auto",
+  "lmdeploy_session_len": null,
+  "lmdeploy_cache_max_entry_count": 0.8,
+  "async_generate": false,
+  "tensor_parallel_size": 1,
+  "sleep_level": 0,
+  "move_model_batches": null,
+  "offload_optimizer": false,
+  "offload_model": false,
+  "gc_collect_after_offload": false,
+  "multi_turn_func": null,
+  "dynamic_sample": false,
+  "max_resample_times": 3,
+  "overlong_filter": false,
+  "soft_max_length": null,
+  "soft_cache_length": null,
+  "scale_rewards": true,
+  "wandb_log_unique_prompts": null,
+  "vllm_server_host": null,
+  "vllm_server_port": 8000,
+  "vllm_server_timeout": 240.0,
+  "num_generations": 8,
+  "max_completion_length": 512,
+  "ds3_gather_for_generation": true,
+  "reward_funcs": [],
+  "reward_weights": null,
+  "log_completions": false,
+  "use_vllm": false,
+  "num_iterations": 1,
+  "rlhf_type": "dpo",
+  "ref_model": null,
+  "ref_model_type": null,
+  "ref_model_revision": null,
+  "beta": 0.1,
+  "label_smoothing": 0,
+  "rpo_alpha": 1.0,
+  "cpo_alpha": 1.0,
+  "simpo_gamma": 1,
+  "desirable_weight": 1.0,
+  "undesirable_weight": 1.0,
+  "center_rewards_coefficient": null,
+  "rank": 0,
+  "global_world_size": 1,
+  "local_world_size": 1,
+  "model_suffix": "Llama-3.2-11B-Vision-Instruct",
+  "model_info": "ModelInfo(model_type='llama3_2_vision', model_dir='/data1/chenjiazun/LLM/Llama-3.2-11B-Vision-Instruct', torch_dtype=torch.bfloat16, max_model_len=131072, quant_method=None, quant_bits=None, rope_scaling={'factor': 8.0, 'high_freq_factor': 4.0, 'low_freq_factor': 1.0, 'original_max_position_embeddings': 8192, 'rope_type': 'llama3'}, config=None, task_type='causal_lm', num_labels=None)",
+  "model_meta": "ModelMeta(model_type='llama3_2_vision', model_groups=[ModelGroup(models=[Model(ms_model_id='LLM-Research/Llama-3.2-11B-Vision-Instruct', hf_model_id='meta-llama/Llama-3.2-11B-Vision-Instruct', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='LLM-Research/Llama-3.2-90B-Vision-Instruct', hf_model_id='meta-llama/Llama-3.2-90B-Vision-Instruct', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='LLM-Research/Llama-3.2-11B-Vision', hf_model_id='meta-llama/Llama-3.2-11B-Vision', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='LLM-Research/Llama-3.2-90B-Vision', hf_model_id='meta-llama/Llama-3.2-90B-Vision', model_path=None, ms_revision=None, hf_revision=None)], ignore_patterns=None, requires=None, tags=[])], template='llama3_2_vision', get_function=<function get_model_tokenizer_llama3_2_vision at 0x7f4f05f4ac20>, model_arch='llama3_2_vision', architectures=['MllamaForConditionalGeneration'], additional_saved_files=[], torch_dtype=None, is_multimodal=True, is_reward=False, task_type=None, ignore_patterns=None, requires=['transformers>=4.45'], tags=[])",
+  "model_dir": "/data1/chenjiazun/LLM/Llama-3.2-11B-Vision-Instruct",
+  "hub": "<class 'swift.hub.hub.MSHub'>",
+  "evaluation_strategy": "steps",
+  "training_args": "DPOConfig(output_dir='/data2/kdd/crag/cjz/output/task3_dpo_gt6_task2_gt30.3+task3gt5_0.1vp/v0-20250611-011957', overwrite_output_dir=False, do_train=False, do_eval=True, do_predict=False, eval_strategy=<IntervalStrategy.STEPS: 'steps'>, prediction_loss_only=False, per_device_train_batch_size=1, per_device_eval_batch_size=1, per_gpu_train_batch_size=None, per_gpu_eval_batch_size=None, gradient_accumulation_steps=32, eval_accumulation_steps=None, eval_delay=0, torch_empty_cache_steps=None, learning_rate=0.0002, weight_decay=0.1, adam_beta1=0.9, adam_beta2=0.95, adam_epsilon=1e-08, max_grad_norm=1.0, num_train_epochs=1.0, max_steps=-1, lr_scheduler_type=<SchedulerType.COSINE: 'cosine'>, lr_scheduler_kwargs=None, warmup_ratio=0.05, warmup_steps=0, log_level='passive', log_level_replica='warning', log_on_each_node=True, logging_dir='/data2/kdd/crag/cjz/output/task3_dpo_gt6_task2_gt30.3+task3gt5_0.1vp/v0-20250611-011957/runs', logging_strategy=<IntervalStrategy.STEPS: 'steps'>, logging_first_step=True, logging_steps=1, logging_nan_inf_filter=True, save_strategy=<SaveStrategy.STEPS: 'steps'>, save_steps=50, save_total_limit=None, save_safetensors=True, save_on_each_node=False, save_only_model=False, restore_callback_states_from_checkpoint=False, no_cuda=False, use_cpu=False, use_mps_device=False, seed=42, data_seed=42, jit_mode_eval=False, use_ipex=False, bf16=True, fp16=False, fp16_opt_level='O1', half_precision_backend='auto', bf16_full_eval=False, fp16_full_eval=False, tf32=None, local_rank=0, ddp_backend=None, tpu_num_cores=None, tpu_metrics_debug=False, debug=[], dataloader_drop_last=False, eval_steps=300, dataloader_num_workers=4, dataloader_prefetch_factor=10, past_index=-1, run_name='/data2/kdd/crag/cjz/output/task3_dpo_gt6_task2_gt30.3+task3gt5_0.1vp/v0-20250611-011957', disable_tqdm=False, remove_unused_columns=False, label_names=None, load_best_model_at_end=False, metric_for_best_model='loss', greater_is_better=False, ignore_data_skip=False, fsdp=[], fsdp_min_num_params=0, fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}, tp_size=0, fsdp_transformer_layer_cls_to_wrap=None, accelerator_config=AcceleratorConfig(split_batches=False, dispatch_batches=False, even_batches=True, use_seedable_sampler=True, non_blocking=False, gradient_accumulation_kwargs=None, use_configured_state=False), deepspeed={'fp16': {'enabled': 'auto', 'loss_scale': 0, 'loss_scale_window': 1000, 'initial_scale_power': 16, 'hysteresis': 2, 'min_loss_scale': 1}, 'bf16': {'enabled': 'auto'}, 'zero_optimization': {'stage': 2, 'offload_optimizer': {'device': 'none', 'pin_memory': True}, 'allgather_partitions': True, 'allgather_bucket_size': 200000000.0, 'overlap_comm': False, 'reduce_scatter': True, 'reduce_bucket_size': 200000000.0, 'contiguous_gradients': True}, 'gradient_accumulation_steps': 'auto', 'gradient_clipping': 'auto', 'steps_per_print': 2000, 'train_batch_size': 'auto', 'train_micro_batch_size_per_gpu': 'auto', 'wall_clock_breakdown': False}, label_smoothing_factor=0.0, optim=<OptimizerNames.ADAMW_TORCH: 'adamw_torch'>, optim_args=None, adafactor=False, group_by_length=False, length_column_name='length', report_to=['tensorboard'], ddp_find_unused_parameters=None, ddp_bucket_cap_mb=None, ddp_broadcast_buffers=None, dataloader_pin_memory=True, dataloader_persistent_workers=False, skip_memory_metrics=True, use_legacy_prediction_loop=False, push_to_hub=False, resume_from_checkpoint=None, hub_model_id=None, hub_strategy=<HubStrategy.EVERY_SAVE: 'every_save'>, hub_token=None, hub_private_repo=None, hub_always_push=False, gradient_checkpointing=True, gradient_checkpointing_kwargs=None, include_inputs_for_metrics=False, include_for_metrics=[], eval_do_concat_batches=True, fp16_backend='auto', push_to_hub_model_id=None, push_to_hub_organization=None, push_to_hub_token=None, mp_parameters='', auto_find_batch_size=False, full_determinism=False, torchdynamo=None, ray_scope='last', ddp_timeout=1800, torch_compile=False, torch_compile_backend=None, torch_compile_mode=None, include_tokens_per_second=None, include_num_input_tokens_seen=None, neftune_noise_alpha=None, optim_target_modules=None, batch_eval_metrics=False, eval_on_start=False, use_liger_kernel=False, eval_use_gather_object=False, average_tokens_across_devices=None, model_init_kwargs=None, ref_model_init_kwargs=None, model_adapter_name=None, ref_adapter_name=None, force_use_ref_model=False, disable_dropout=True, use_logits_to_keep=False, dataset_num_proc=4, padding_value=None, label_pad_token_id=None, max_prompt_length=512, max_completion_length=512, max_length=4096, truncation_mode='keep_end', padding_free=False, precompute_ref_log_probs=False, precompute_ref_batch_size=None, tools=None, loss_type='sigmoid', beta=0.1, f_divergence_type=<FDivergenceType.REVERSE_KL: 'reverse_kl'>, f_alpha_divergence_coef=1.0, reference_free=False, label_smoothing=0, use_weighting=False, rpo_alpha=1.0, discopop_tau=0.05, sync_ref_model=False, ref_model_mixup_alpha=0.6, ref_model_sync_steps=512, generate_during_eval=False, check_model=True, acc_strategy='token', train_dataloader_shuffle=True, metric_warmup_step=0, fsdp_num=1, acc_steps=1, eval_use_evalscope=False, eval_datasets=[], eval_limit=None, eval_datasets_args=None, eval_generation_config=None, train_type='lora', optimizer=None, local_repo_path=None, galore_config=None)"
+}

latest

@@ -0,0 +1 @@
+global_step150

rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:835f869ea325fd6edf27b48b589309fb66641cb92b45f2fc13d1bb6e8814106c
+size 14244

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d85d3d27fd8ffd845d095fc8135f039e37018a2a4dc1db4126f1132ada08658
+size 1064

trainer_state.json

@@ -0,0 +1,2734 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.1430316755564826,
+  "eval_steps": 300,
+  "global_step": 150,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0009535445037098841,
+      "grad_norm": 9.156327247619629,
+      "learning_rate": 3.7735849056603773e-06,
+      "logits/chosen": -0.6435090899467468,
+      "logits/rejected": -0.6501147150993347,
+      "logps/chosen": -12.8789701461792,
+      "logps/rejected": -12.67190170288086,
+      "loss": 1.3995760679244995,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7064292430877686,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": 0.0,
+      "rewards/margins": 0.0,
+      "rewards/rejected": 0.0,
+      "step": 1,
+      "train_speed(iter/s)": 0.005503
+    },
+    {
+      "epoch": 0.0019070890074197682,
+      "grad_norm": 3.750561237335205,
+      "learning_rate": 7.547169811320755e-06,
+      "logits/chosen": -0.5811247229576111,
+      "logits/rejected": -0.5784763693809509,
+      "logps/chosen": -11.550155639648438,
+      "logps/rejected": -10.463687896728516,
+      "loss": 1.1846946477890015,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4915475845336914,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": 0.0,
+      "rewards/margins": 0.0,
+      "rewards/rejected": 0.0,
+      "step": 2,
+      "train_speed(iter/s)": 0.005627
+    },
+    {
+      "epoch": 0.002860633511129652,
+      "grad_norm": 9.728832244873047,
+      "learning_rate": 1.1320754716981132e-05,
+      "logits/chosen": -0.6211333274841309,
+      "logits/rejected": -0.6250367760658264,
+      "logps/chosen": -13.265661239624023,
+      "logps/rejected": -17.363025665283203,
+      "loss": 1.4022446870803833,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7087438106536865,
+      "rewards/accuracies": 0.46875,
+      "rewards/chosen": 0.0019078836776316166,
+      "rewards/margins": -0.0005585900507867336,
+      "rewards/rejected": 0.002466473262757063,
+      "step": 3,
+      "train_speed(iter/s)": 0.005589
+    },
+    {
+      "epoch": 0.0038141780148395364,
+      "grad_norm": 4.456181049346924,
+      "learning_rate": 1.509433962264151e-05,
+      "logits/chosen": -0.6295092105865479,
+      "logits/rejected": -0.6310409307479858,
+      "logps/chosen": -20.259197235107422,
+      "logps/rejected": -15.924739837646484,
+      "loss": 1.4261250495910645,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7309765815734863,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.003216111334040761,
+      "rewards/margins": -0.003784709610044956,
+      "rewards/rejected": 0.0005685988580808043,
+      "step": 4,
+      "train_speed(iter/s)": 0.005576
+    },
+    {
+      "epoch": 0.004767722518549421,
+      "grad_norm": 5.137070178985596,
+      "learning_rate": 1.8867924528301888e-05,
+      "logits/chosen": -0.6352660059928894,
+      "logits/rejected": -0.6393146514892578,
+      "logps/chosen": -11.320538520812988,
+      "logps/rejected": -18.213350296020508,
+      "loss": 1.3472224473953247,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.659798800945282,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 0.009231260977685452,
+      "rewards/margins": 0.011925339698791504,
+      "rewards/rejected": -0.0026940787211060524,
+      "step": 5,
+      "train_speed(iter/s)": 0.005563
+    },
+    {
+      "epoch": 0.005721267022259304,
+      "grad_norm": 4.738046169281006,
+      "learning_rate": 2.2641509433962265e-05,
+      "logits/chosen": -0.621586799621582,
+      "logits/rejected": -0.6225624680519104,
+      "logps/chosen": -13.36612319946289,
+      "logps/rejected": -12.092449188232422,
+      "loss": 1.2847044467926025,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.593052089214325,
+      "rewards/accuracies": 0.46875,
+      "rewards/chosen": 0.001901240088045597,
+      "rewards/margins": 0.0050404006615281105,
+      "rewards/rejected": -0.0031391603406518698,
+      "step": 6,
+      "train_speed(iter/s)": 0.005557
+    },
+    {
+      "epoch": 0.0066748115259691885,
+      "grad_norm": 4.152891159057617,
+      "learning_rate": 2.641509433962264e-05,
+      "logits/chosen": -0.6192730665206909,
+      "logits/rejected": -0.6235961318016052,
+      "logps/chosen": -12.630643844604492,
+      "logps/rejected": -15.917936325073242,
+      "loss": 1.2920165061950684,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5905699133872986,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": 0.016993187367916107,
+      "rewards/margins": -0.008502829819917679,
+      "rewards/rejected": 0.025496022775769234,
+      "step": 7,
+      "train_speed(iter/s)": 0.005538
+    },
+    {
+      "epoch": 0.007628356029679073,
+      "grad_norm": 4.160374641418457,
+      "learning_rate": 3.018867924528302e-05,
+      "logits/chosen": -0.6182655692100525,
+      "logits/rejected": -0.6189552545547485,
+      "logps/chosen": -10.698919296264648,
+      "logps/rejected": -14.148299217224121,
+      "loss": 1.0840333700180054,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3833664357662201,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.006827078759670258,
+      "rewards/margins": 0.02289848029613495,
+      "rewards/rejected": -0.0297255702316761,
+      "step": 8,
+      "train_speed(iter/s)": 0.005529
+    },
+    {
+      "epoch": 0.008581900533388956,
+      "grad_norm": 7.751194953918457,
+      "learning_rate": 3.39622641509434e-05,
+      "logits/chosen": -0.5885738134384155,
+      "logits/rejected": -0.5893712043762207,
+      "logps/chosen": -14.677619934082031,
+      "logps/rejected": -19.348114013671875,
+      "loss": 1.4759101867675781,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7508934140205383,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": -0.06757626682519913,
+      "rewards/margins": -0.03109188750386238,
+      "rewards/rejected": -0.036484383046627045,
+      "step": 9,
+      "train_speed(iter/s)": 0.005546
+    },
+    {
+      "epoch": 0.009535445037098841,
+      "grad_norm": 5.425357818603516,
+      "learning_rate": 3.7735849056603776e-05,
+      "logits/chosen": -0.6194073557853699,
+      "logits/rejected": -0.6175965666770935,
+      "logps/chosen": -16.864004135131836,
+      "logps/rejected": -13.315808296203613,
+      "loss": 1.264159083366394,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5726930499076843,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.03342372551560402,
+      "rewards/margins": 0.05060018226504326,
+      "rewards/rejected": -0.08402392268180847,
+      "step": 10,
+      "train_speed(iter/s)": 0.005559
+    },
+    {
+      "epoch": 0.010488989540808725,
+      "grad_norm": 4.181962013244629,
+      "learning_rate": 4.150943396226415e-05,
+      "logits/chosen": -0.6295148730278015,
+      "logits/rejected": -0.6311144232749939,
+      "logps/chosen": -17.622665405273438,
+      "logps/rejected": -22.323240280151367,
+      "loss": 1.1960029602050781,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4701622426509857,
+      "rewards/accuracies": 0.46875,
+      "rewards/chosen": -0.05372938886284828,
+      "rewards/margins": -0.0011130962520837784,
+      "rewards/rejected": -0.05261629447340965,
+      "step": 11,
+      "train_speed(iter/s)": 0.00555
+    },
+    {
+      "epoch": 0.011442534044518608,
+      "grad_norm": 3.3646111488342285,
+      "learning_rate": 4.528301886792453e-05,
+      "logits/chosen": -0.6567115783691406,
+      "logits/rejected": -0.6585782170295715,
+      "logps/chosen": -11.29936408996582,
+      "logps/rejected": -18.29203987121582,
+      "loss": 1.1115467548370361,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4559721350669861,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.08929000794887543,
+      "rewards/margins": 0.12125876545906067,
+      "rewards/rejected": -0.03196876123547554,
+      "step": 12,
+      "train_speed(iter/s)": 0.005551
+    },
+    {
+      "epoch": 0.012396078548228493,
+      "grad_norm": 3.79887318611145,
+      "learning_rate": 4.9056603773584906e-05,
+      "logits/chosen": -0.5698614716529846,
+      "logits/rejected": -0.5705659985542297,
+      "logps/chosen": -16.54460906982422,
+      "logps/rejected": -16.52007484436035,
+      "loss": 1.2756099700927734,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.6095594167709351,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": 0.08876940608024597,
+      "rewards/margins": 0.0944020226597786,
+      "rewards/rejected": -0.005632609128952026,
+      "step": 13,
+      "train_speed(iter/s)": 0.005543
+    },
+    {
+      "epoch": 0.013349623051938377,
+      "grad_norm": 3.3852715492248535,
+      "learning_rate": 5.283018867924528e-05,
+      "logits/chosen": -0.5822104811668396,
+      "logits/rejected": -0.5813897848129272,
+      "logps/chosen": -15.234231948852539,
+      "logps/rejected": -15.407049179077148,
+      "loss": 1.2322038412094116,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.6042034029960632,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.1384495347738266,
+      "rewards/margins": 0.16144631803035736,
+      "rewards/rejected": -0.022996775805950165,
+      "step": 14,
+      "train_speed(iter/s)": 0.005541
+    },
+    {
+      "epoch": 0.014303167555648262,
+      "grad_norm": 2.7666451930999756,
+      "learning_rate": 5.660377358490566e-05,
+      "logits/chosen": -0.5926915407180786,
+      "logits/rejected": -0.5929235816001892,
+      "logps/chosen": -14.370649337768555,
+      "logps/rejected": -14.672292709350586,
+      "loss": 1.268752098083496,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5840609669685364,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": 0.13987205922603607,
+      "rewards/margins": 0.04084888845682144,
+      "rewards/rejected": 0.09902316331863403,
+      "step": 15,
+      "train_speed(iter/s)": 0.005545
+    },
+    {
+      "epoch": 0.015256712059358146,
+      "grad_norm": 3.996248960494995,
+      "learning_rate": 6.037735849056604e-05,
+      "logits/chosen": -0.6161852478981018,
+      "logits/rejected": -0.6143717169761658,
+      "logps/chosen": -17.775665283203125,
+      "logps/rejected": -16.12677001953125,
+      "loss": 1.3597179651260376,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7031959295272827,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": 0.14894011616706848,
+      "rewards/margins": 0.10105747729539871,
+      "rewards/rejected": 0.047882646322250366,
+      "step": 16,
+      "train_speed(iter/s)": 0.005542
+    },
+    {
+      "epoch": 0.01621025656306803,
+      "grad_norm": 3.343315601348877,
+      "learning_rate": 6.415094339622641e-05,
+      "logits/chosen": -0.5822757482528687,
+      "logits/rejected": -0.5849615931510925,
+      "logps/chosen": -11.210283279418945,
+      "logps/rejected": -17.080455780029297,
+      "loss": 1.1165493726730347,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4471588432788849,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.13627491891384125,
+      "rewards/margins": 0.07742398977279663,
+      "rewards/rejected": 0.058850936591625214,
+      "step": 17,
+      "train_speed(iter/s)": 0.005542
+    },
+    {
+      "epoch": 0.017163801066777912,
+      "grad_norm": 2.5265791416168213,
+      "learning_rate": 6.79245283018868e-05,
+      "logits/chosen": -0.5711070895195007,
+      "logits/rejected": -0.5730242729187012,
+      "logps/chosen": -14.197075843811035,
+      "logps/rejected": -15.04891586303711,
+      "loss": 1.1029906272888184,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.45255714654922485,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": 0.1637343168258667,
+      "rewards/margins": 0.1597798466682434,
+      "rewards/rejected": 0.0039544664323329926,
+      "step": 18,
+      "train_speed(iter/s)": 0.005535
+    },
+    {
+      "epoch": 0.0181173455704878,
+      "grad_norm": 4.558610916137695,
+      "learning_rate": 7.169811320754717e-05,
+      "logits/chosen": -0.572235643863678,
+      "logits/rejected": -0.5743283629417419,
+      "logps/chosen": -9.268120765686035,
+      "logps/rejected": -12.118350982666016,
+      "loss": 1.1668801307678223,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5335167646408081,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.19031290709972382,
+      "rewards/margins": 0.15231794118881226,
+      "rewards/rejected": 0.03799496963620186,
+      "step": 19,
+      "train_speed(iter/s)": 0.005519
+    },
+    {
+      "epoch": 0.019070890074197683,
+      "grad_norm": 3.7134947776794434,
+      "learning_rate": 7.547169811320755e-05,
+      "logits/chosen": -0.585778534412384,
+      "logits/rejected": -0.5890673398971558,
+      "logps/chosen": -8.275850296020508,
+      "logps/rejected": -16.348182678222656,
+      "loss": 1.0512793064117432,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.39501887559890747,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.1688304841518402,
+      "rewards/margins": 0.12996332347393036,
+      "rewards/rejected": 0.03886716440320015,
+      "step": 20,
+      "train_speed(iter/s)": 0.005525
+    },
+    {
+      "epoch": 0.020024434577907566,
+      "grad_norm": 3.5586955547332764,
+      "learning_rate": 7.924528301886794e-05,
+      "logits/chosen": -0.5529674291610718,
+      "logits/rejected": -0.5549667477607727,
+      "logps/chosen": -14.883947372436523,
+      "logps/rejected": -19.51848793029785,
+      "loss": 1.2289440631866455,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5555365085601807,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.06489060074090958,
+      "rewards/margins": 0.08113954961299896,
+      "rewards/rejected": -0.016248945146799088,
+      "step": 21,
+      "train_speed(iter/s)": 0.005523
+    },
+    {
+      "epoch": 0.02097797908161745,
+      "grad_norm": 2.970125913619995,
+      "learning_rate": 8.30188679245283e-05,
+      "logits/chosen": -0.5630563497543335,
+      "logits/rejected": -0.5648879408836365,
+      "logps/chosen": -9.709285736083984,
+      "logps/rejected": -14.605964660644531,
+      "loss": 1.0496439933776855,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4030095934867859,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": 0.13459333777427673,
+      "rewards/margins": 0.13257721066474915,
+      "rewards/rejected": 0.0020161233842372894,
+      "step": 22,
+      "train_speed(iter/s)": 0.005527
+    },
+    {
+      "epoch": 0.021931523585327333,
+      "grad_norm": 3.7449333667755127,
+      "learning_rate": 8.679245283018869e-05,
+      "logits/chosen": -0.5388307571411133,
+      "logits/rejected": -0.5379776954650879,
+      "logps/chosen": -20.206819534301758,
+      "logps/rejected": -10.312554359436035,
+      "loss": 1.3675460815429688,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7346486449241638,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.22365330159664154,
+      "rewards/margins": 0.19076848030090332,
+      "rewards/rejected": 0.032884806394577026,
+      "step": 23,
+      "train_speed(iter/s)": 0.005527
+    },
+    {
+      "epoch": 0.022885068089037217,
+      "grad_norm": 4.3556928634643555,
+      "learning_rate": 9.056603773584906e-05,
+      "logits/chosen": -0.5620133876800537,
+      "logits/rejected": -0.5651191473007202,
+      "logps/chosen": -13.681132316589355,
+      "logps/rejected": -17.737977981567383,
+      "loss": 1.427290439605713,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7979669570922852,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.20859572291374207,
+      "rewards/margins": 0.19208168983459473,
+      "rewards/rejected": 0.01651405170559883,
+      "step": 24,
+      "train_speed(iter/s)": 0.005528
+    },
+    {
+      "epoch": 0.023838612592747103,
+      "grad_norm": 2.59773850440979,
+      "learning_rate": 9.433962264150944e-05,
+      "logits/chosen": -0.5109421014785767,
+      "logits/rejected": -0.5130017995834351,
+      "logps/chosen": -11.422687530517578,
+      "logps/rejected": -19.77487564086914,
+      "loss": 1.0190178155899048,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3792160749435425,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.17286184430122375,
+      "rewards/margins": 0.1656832993030548,
+      "rewards/rejected": 0.007178528234362602,
+      "step": 25,
+      "train_speed(iter/s)": 0.005528
+    },
+    {
+      "epoch": 0.024792157096456987,
+      "grad_norm": 3.2465970516204834,
+      "learning_rate": 9.811320754716981e-05,
+      "logits/chosen": -0.4895856976509094,
+      "logits/rejected": -0.490243524312973,
+      "logps/chosen": -8.71623420715332,
+      "logps/rejected": -7.8305277824401855,
+      "loss": 1.0329375267028809,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3276180028915405,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": 0.13548454642295837,
+      "rewards/margins": 0.014407174661755562,
+      "rewards/rejected": 0.12107735872268677,
+      "step": 26,
+      "train_speed(iter/s)": 0.005523
+    },
+    {
+      "epoch": 0.02574570160016687,
+      "grad_norm": 2.9056291580200195,
+      "learning_rate": 0.0001018867924528302,
+      "logits/chosen": -0.5098599195480347,
+      "logits/rejected": -0.5105721950531006,
+      "logps/chosen": -17.795873641967773,
+      "logps/rejected": -13.080977439880371,
+      "loss": 1.1328891515731812,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4805879294872284,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.18526741862297058,
+      "rewards/margins": 0.11028175055980682,
+      "rewards/rejected": 0.07498566806316376,
+      "step": 27,
+      "train_speed(iter/s)": 0.005528
+    },
+    {
+      "epoch": 0.026699246103876754,
+      "grad_norm": 3.9831905364990234,
+      "learning_rate": 0.00010566037735849057,
+      "logits/chosen": -0.5249112844467163,
+      "logits/rejected": -0.5256497859954834,
+      "logps/chosen": -9.17713737487793,
+      "logps/rejected": -16.791439056396484,
+      "loss": 1.0309749841690063,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.49379563331604004,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.26130685210227966,
+      "rewards/margins": 0.4484291970729828,
+      "rewards/rejected": -0.1871223747730255,
+      "step": 28,
+      "train_speed(iter/s)": 0.00553
+    },
+    {
+      "epoch": 0.027652790607586637,
+      "grad_norm": 3.447692394256592,
+      "learning_rate": 0.00010943396226415095,
+      "logits/chosen": -0.46926626563072205,
+      "logits/rejected": -0.474046528339386,
+      "logps/chosen": -10.564407348632812,
+      "logps/rejected": -13.679093360900879,
+      "loss": 1.0266515016555786,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4402708113193512,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.25625136494636536,
+      "rewards/margins": 0.27123552560806274,
+      "rewards/rejected": -0.014984178356826305,
+      "step": 29,
+      "train_speed(iter/s)": 0.005529
+    },
+    {
+      "epoch": 0.028606335111296524,
+      "grad_norm": 2.3513646125793457,
+      "learning_rate": 0.00011320754716981132,
+      "logits/chosen": -0.4841839671134949,
+      "logits/rejected": -0.4823201894760132,
+      "logps/chosen": -16.52553367614746,
+      "logps/rejected": -12.728946685791016,
+      "loss": 1.051925539970398,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5132753252983093,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.28733503818511963,
+      "rewards/margins": 0.3806324601173401,
+      "rewards/rejected": -0.09329745918512344,
+      "step": 30,
+      "train_speed(iter/s)": 0.00553
+    },
+    {
+      "epoch": 0.029559879615006408,
+      "grad_norm": 2.8920440673828125,
+      "learning_rate": 0.0001169811320754717,
+      "logits/chosen": -0.4617501497268677,
+      "logits/rejected": -0.4629114866256714,
+      "logps/chosen": -9.827648162841797,
+      "logps/rejected": -11.787566184997559,
+      "loss": 1.0702807903289795,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4221873879432678,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.12683041393756866,
+      "rewards/margins": 0.20381762087345123,
+      "rewards/rejected": -0.07698723673820496,
+      "step": 31,
+      "train_speed(iter/s)": 0.005529
+    },
+    {
+      "epoch": 0.03051342411871629,
+      "grad_norm": 3.150108575820923,
+      "learning_rate": 0.00012075471698113207,
+      "logits/chosen": -0.46168380975723267,
+      "logits/rejected": -0.45957043766975403,
+      "logps/chosen": -14.481122970581055,
+      "logps/rejected": -12.509255409240723,
+      "loss": 1.1068114042282104,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4239238202571869,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": 0.07766711711883545,
+      "rewards/margins": 0.11046308279037476,
+      "rewards/rejected": -0.03279596567153931,
+      "step": 32,
+      "train_speed(iter/s)": 0.005524
+    },
+    {
+      "epoch": 0.03146696862242618,
+      "grad_norm": 4.185500621795654,
+      "learning_rate": 0.00012452830188679244,
+      "logits/chosen": -0.4712987542152405,
+      "logits/rejected": -0.47331178188323975,
+      "logps/chosen": -13.179000854492188,
+      "logps/rejected": -19.023120880126953,
+      "loss": 1.3572198152542114,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7039042711257935,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": 0.17201806604862213,
+      "rewards/margins": 0.16012105345726013,
+      "rewards/rejected": 0.011897009797394276,
+      "step": 33,
+      "train_speed(iter/s)": 0.005529
+    },
+    {
+      "epoch": 0.03242051312613606,
+      "grad_norm": 3.005161762237549,
+      "learning_rate": 0.00012830188679245283,
+      "logits/chosen": -0.4754801392555237,
+      "logits/rejected": -0.4769393503665924,
+      "logps/chosen": -11.273967742919922,
+      "logps/rejected": -16.496191024780273,
+      "loss": 1.0072283744812012,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.41234296560287476,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.22565320134162903,
+      "rewards/margins": 0.27443331480026245,
+      "rewards/rejected": -0.04878011345863342,
+      "step": 34,
+      "train_speed(iter/s)": 0.005526
+    },
+    {
+      "epoch": 0.033374057629845945,
+      "grad_norm": 5.103848457336426,
+      "learning_rate": 0.0001320754716981132,
+      "logits/chosen": -0.44018203020095825,
+      "logits/rejected": -0.44089382886886597,
+      "logps/chosen": -15.832134246826172,
+      "logps/rejected": -19.550304412841797,
+      "loss": 1.2559186220169067,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5793285965919495,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": 0.09596908837556839,
+      "rewards/margins": 0.11671226471662521,
+      "rewards/rejected": -0.020743174478411674,
+      "step": 35,
+      "train_speed(iter/s)": 0.005527
+    },
+    {
+      "epoch": 0.034327602133555825,
+      "grad_norm": 4.321141719818115,
+      "learning_rate": 0.0001358490566037736,
+      "logits/chosen": -0.49340876936912537,
+      "logits/rejected": -0.49693170189857483,
+      "logps/chosen": -17.06662940979004,
+      "logps/rejected": -20.88056755065918,
+      "loss": 1.2329245805740356,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.6244450807571411,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.2872624695301056,
+      "rewards/margins": 0.23183389008045197,
+      "rewards/rejected": 0.05542857572436333,
+      "step": 36,
+      "train_speed(iter/s)": 0.00553
+    },
+    {
+      "epoch": 0.03528114663726571,
+      "grad_norm": 3.112792491912842,
+      "learning_rate": 0.00013962264150943395,
+      "logits/chosen": -0.4647415578365326,
+      "logits/rejected": -0.4648529291152954,
+      "logps/chosen": -12.979642868041992,
+      "logps/rejected": -13.105558395385742,
+      "loss": 1.1428803205490112,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.48155656456947327,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": 0.19289334118366241,
+      "rewards/margins": 0.169374018907547,
+      "rewards/rejected": 0.023519329726696014,
+      "step": 37,
+      "train_speed(iter/s)": 0.005535
+    },
+    {
+      "epoch": 0.0362346911409756,
+      "grad_norm": 4.363394260406494,
+      "learning_rate": 0.00014339622641509434,
+      "logits/chosen": -0.47051119804382324,
+      "logits/rejected": -0.4702276289463043,
+      "logps/chosen": -11.824732780456543,
+      "logps/rejected": -18.102703094482422,
+      "loss": 0.9977712631225586,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.48080000281333923,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.35463425517082214,
+      "rewards/margins": 0.5215649008750916,
+      "rewards/rejected": -0.166930690407753,
+      "step": 38,
+      "train_speed(iter/s)": 0.005537
+    },
+    {
+      "epoch": 0.03718823564468548,
+      "grad_norm": 2.3291094303131104,
+      "learning_rate": 0.00014716981132075472,
+      "logits/chosen": -0.4471542239189148,
+      "logits/rejected": -0.44849908351898193,
+      "logps/chosen": -10.541170120239258,
+      "logps/rejected": -13.821127891540527,
+      "loss": 1.0517562627792358,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4245188236236572,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.17497971653938293,
+      "rewards/margins": 0.2404995709657669,
+      "rewards/rejected": -0.06551986932754517,
+      "step": 39,
+      "train_speed(iter/s)": 0.005533
+    },
+    {
+      "epoch": 0.038141780148395366,
+      "grad_norm": 2.5439131259918213,
+      "learning_rate": 0.0001509433962264151,
+      "logits/chosen": -0.43407440185546875,
+      "logits/rejected": -0.4330942928791046,
+      "logps/chosen": -12.606916427612305,
+      "logps/rejected": -8.916302680969238,
+      "loss": 1.1393225193023682,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.500526487827301,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.15790946781635284,
+      "rewards/margins": 0.28719791769981384,
+      "rewards/rejected": -0.129288449883461,
+      "step": 40,
+      "train_speed(iter/s)": 0.005533
+    },
+    {
+      "epoch": 0.039095324652105246,
+      "grad_norm": 2.752281904220581,
+      "learning_rate": 0.0001547169811320755,
+      "logits/chosen": -0.4207141399383545,
+      "logits/rejected": -0.42105770111083984,
+      "logps/chosen": -9.193354606628418,
+      "logps/rejected": -15.80713939666748,
+      "loss": 1.0188286304473877,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.41979092359542847,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.11082945019006729,
+      "rewards/margins": 0.3942355513572693,
+      "rewards/rejected": -0.2834060788154602,
+      "step": 41,
+      "train_speed(iter/s)": 0.005532
+    },
+    {
+      "epoch": 0.04004886915581513,
+      "grad_norm": 2.4327123165130615,
+      "learning_rate": 0.00015849056603773587,
+      "logits/chosen": -0.445608913898468,
+      "logits/rejected": -0.4462995231151581,
+      "logps/chosen": -8.890420913696289,
+      "logps/rejected": -9.104397773742676,
+      "loss": 1.0348408222198486,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4377897083759308,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.1634826809167862,
+      "rewards/margins": 0.28891459107398987,
+      "rewards/rejected": -0.12543192505836487,
+      "step": 42,
+      "train_speed(iter/s)": 0.005537
+    },
+    {
+      "epoch": 0.04100241365952501,
+      "grad_norm": 2.331282377243042,
+      "learning_rate": 0.00016226415094339625,
+      "logits/chosen": -0.4530089497566223,
+      "logits/rejected": -0.4580504298210144,
+      "logps/chosen": -7.02070426940918,
+      "logps/rejected": -22.712093353271484,
+      "loss": 0.876542329788208,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3533342182636261,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.29712867736816406,
+      "rewards/margins": 0.5244494676589966,
+      "rewards/rejected": -0.22732074558734894,
+      "step": 43,
+      "train_speed(iter/s)": 0.005538
+    },
+    {
+      "epoch": 0.0419559581632349,
+      "grad_norm": 2.865417003631592,
+      "learning_rate": 0.0001660377358490566,
+      "logits/chosen": -0.4700409770011902,
+      "logits/rejected": -0.4735233187675476,
+      "logps/chosen": -12.126070022583008,
+      "logps/rejected": -24.352678298950195,
+      "loss": 1.157395362854004,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5185694694519043,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.1286190301179886,
+      "rewards/margins": 0.2031334638595581,
+      "rewards/rejected": -0.07451442629098892,
+      "step": 44,
+      "train_speed(iter/s)": 0.005538
+    },
+    {
+      "epoch": 0.042909502666944786,
+      "grad_norm": 2.9107770919799805,
+      "learning_rate": 0.000169811320754717,
+      "logits/chosen": -0.45035916566848755,
+      "logits/rejected": -0.4494967758655548,
+      "logps/chosen": -17.021671295166016,
+      "logps/rejected": -16.504793167114258,
+      "loss": 1.2039690017700195,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5990548729896545,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.22049224376678467,
+      "rewards/margins": 0.25519850850105286,
+      "rewards/rejected": -0.03470626473426819,
+      "step": 45,
+      "train_speed(iter/s)": 0.005534
+    },
+    {
+      "epoch": 0.043863047170654666,
+      "grad_norm": 2.5569493770599365,
+      "learning_rate": 0.00017358490566037738,
+      "logits/chosen": -0.4576927423477173,
+      "logits/rejected": -0.4589836597442627,
+      "logps/chosen": -15.28840160369873,
+      "logps/rejected": -21.630834579467773,
+      "loss": 1.0875338315963745,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5480765104293823,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.2394683063030243,
+      "rewards/margins": 0.43596386909484863,
+      "rewards/rejected": -0.19649554789066315,
+      "step": 46,
+      "train_speed(iter/s)": 0.005534
+    },
+    {
+      "epoch": 0.04481659167436455,
+      "grad_norm": 2.6082468032836914,
+      "learning_rate": 0.00017735849056603776,
+      "logits/chosen": -0.44933271408081055,
+      "logits/rejected": -0.4535468816757202,
+      "logps/chosen": -13.067358016967773,
+      "logps/rejected": -25.21561622619629,
+      "loss": 0.9420796632766724,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.44500476121902466,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.3027465343475342,
+      "rewards/margins": 0.5848984122276306,
+      "rewards/rejected": -0.28215184807777405,
+      "step": 47,
+      "train_speed(iter/s)": 0.005534
+    },
+    {
+      "epoch": 0.04577013617807443,
+      "grad_norm": 2.8552000522613525,
+      "learning_rate": 0.00018113207547169812,
+      "logits/chosen": -0.44021791219711304,
+      "logits/rejected": -0.4409497380256653,
+      "logps/chosen": -13.639575958251953,
+      "logps/rejected": -19.479185104370117,
+      "loss": 1.0777249336242676,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5401443243026733,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.2517797350883484,
+      "rewards/margins": 0.4618395268917084,
+      "rewards/rejected": -0.21005980670452118,
+      "step": 48,
+      "train_speed(iter/s)": 0.00553
+    },
+    {
+      "epoch": 0.04672368068178432,
+      "grad_norm": 2.530268430709839,
+      "learning_rate": 0.0001849056603773585,
+      "logits/chosen": -0.47215527296066284,
+      "logits/rejected": -0.47504734992980957,
+      "logps/chosen": -11.48988151550293,
+      "logps/rejected": -36.38770294189453,
+      "loss": 0.9700489640235901,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.46057599782943726,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.21084292232990265,
+      "rewards/margins": 0.533352792263031,
+      "rewards/rejected": -0.32250988483428955,
+      "step": 49,
+      "train_speed(iter/s)": 0.005532
+    },
+    {
+      "epoch": 0.04767722518549421,
+      "grad_norm": 2.7035655975341797,
+      "learning_rate": 0.00018867924528301889,
+      "logits/chosen": -0.4868597388267517,
+      "logits/rejected": -0.4888378977775574,
+      "logps/chosen": -13.504924774169922,
+      "logps/rejected": -21.10767936706543,
+      "loss": 1.0414435863494873,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5301634669303894,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.18318726122379303,
+      "rewards/margins": 0.619285523891449,
+      "rewards/rejected": -0.43609827756881714,
+      "step": 50,
+      "train_speed(iter/s)": 0.005531
+    },
+    {
+      "epoch": 0.04863076968920409,
+      "grad_norm": 3.0783157348632812,
+      "learning_rate": 0.00019245283018867927,
+      "logits/chosen": -0.488979697227478,
+      "logits/rejected": -0.4920251965522766,
+      "logps/chosen": -6.389867305755615,
+      "logps/rejected": -23.881044387817383,
+      "loss": 0.7357016801834106,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.2519811689853668,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.22045277059078217,
+      "rewards/margins": 0.6349221467971802,
+      "rewards/rejected": -0.4144693613052368,
+      "step": 51,
+      "train_speed(iter/s)": 0.00552
+    },
+    {
+      "epoch": 0.049584314192913974,
+      "grad_norm": 2.936868906021118,
+      "learning_rate": 0.00019622641509433963,
+      "logits/chosen": -0.49481600522994995,
+      "logits/rejected": -0.4982725977897644,
+      "logps/chosen": -7.566017150878906,
+      "logps/rejected": -28.77280044555664,
+      "loss": 0.8173349499702454,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.2955441176891327,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.12177374958992004,
+      "rewards/margins": 0.6433938145637512,
+      "rewards/rejected": -0.5216200947761536,
+      "step": 52,
+      "train_speed(iter/s)": 0.005518
+    },
+    {
+      "epoch": 0.050537858696623854,
+      "grad_norm": 4.539004802703857,
+      "learning_rate": 0.0002,
+      "logits/chosen": -0.5527479648590088,
+      "logits/rejected": -0.5551312565803528,
+      "logps/chosen": -15.817987442016602,
+      "logps/rejected": -22.29538345336914,
+      "loss": 1.1207798719406128,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5884667038917542,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.18014809489250183,
+      "rewards/margins": 0.6673915982246399,
+      "rewards/rejected": -0.4872434735298157,
+      "step": 53,
+      "train_speed(iter/s)": 0.005519
+    },
+    {
+      "epoch": 0.05149140320033374,
+      "grad_norm": 4.143117904663086,
+      "learning_rate": 0.00019999950154813253,
+      "logits/chosen": -0.5354641675949097,
+      "logits/rejected": -0.5355905294418335,
+      "logps/chosen": -12.528202056884766,
+      "logps/rejected": -18.404970169067383,
+      "loss": 1.0412046909332275,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5650326609611511,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.17625540494918823,
+      "rewards/margins": 0.8302103877067566,
+      "rewards/rejected": -0.6539549827575684,
+      "step": 54,
+      "train_speed(iter/s)": 0.005522
+    },
+    {
+      "epoch": 0.05244494770404363,
+      "grad_norm": 3.329041004180908,
+      "learning_rate": 0.00019999800619749922,
+      "logits/chosen": -0.5789484977722168,
+      "logits/rejected": -0.5782615542411804,
+      "logps/chosen": -14.283157348632812,
+      "logps/rejected": -24.138996124267578,
+      "loss": 1.0249700546264648,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5574489831924438,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.08275524526834488,
+      "rewards/margins": 0.9348181486129761,
+      "rewards/rejected": -0.8520628809928894,
+      "step": 55,
+      "train_speed(iter/s)": 0.005525
+    },
+    {
+      "epoch": 0.05339849220775351,
+      "grad_norm": 13.451016426086426,
+      "learning_rate": 0.00019999551396300723,
+      "logits/chosen": -0.6104876399040222,
+      "logits/rejected": -0.6144217848777771,
+      "logps/chosen": -11.782422065734863,
+      "logps/rejected": -26.71551513671875,
+      "loss": 1.1726720333099365,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.6291560530662537,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.0043334998190402985,
+      "rewards/margins": 0.6636021137237549,
+      "rewards/rejected": -0.6679355502128601,
+      "step": 56,
+      "train_speed(iter/s)": 0.005523
+    },
+    {
+      "epoch": 0.054352036711463395,
+      "grad_norm": 7.804044246673584,
+      "learning_rate": 0.00019999202486950177,
+      "logits/chosen": -0.5772112011909485,
+      "logits/rejected": -0.5761610865592957,
+      "logps/chosen": -23.908967971801758,
+      "logps/rejected": -30.196250915527344,
+      "loss": 1.3278322219848633,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7636411786079407,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.15834051370620728,
+      "rewards/margins": 0.4628582298755646,
+      "rewards/rejected": -0.3045177161693573,
+      "step": 57,
+      "train_speed(iter/s)": 0.005522
+    },
+    {
+      "epoch": 0.055305581215173275,
+      "grad_norm": 7.5992913246154785,
+      "learning_rate": 0.00019998753895176575,
+      "logits/chosen": -0.6457486748695374,
+      "logits/rejected": -0.6466608047485352,
+      "logps/chosen": -10.734827041625977,
+      "logps/rejected": -32.13155746459961,
+      "loss": 1.0030261278152466,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5355030298233032,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -0.01070893183350563,
+      "rewards/margins": 0.7878642082214355,
+      "rewards/rejected": -0.7985730171203613,
+      "step": 58,
+      "train_speed(iter/s)": 0.005524
+    },
+    {
+      "epoch": 0.05625912571888316,
+      "grad_norm": 5.3810601234436035,
+      "learning_rate": 0.00019998205625451943,
+      "logits/chosen": -0.6111299991607666,
+      "logits/rejected": -0.6088098287582397,
+      "logps/chosen": -16.970266342163086,
+      "logps/rejected": -19.622106552124023,
+      "loss": 0.9686940908432007,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5118071436882019,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.15675117075443268,
+      "rewards/margins": 0.8334945440292358,
+      "rewards/rejected": -0.6767433881759644,
+      "step": 59,
+      "train_speed(iter/s)": 0.005525
+    },
+    {
+      "epoch": 0.05721267022259305,
+      "grad_norm": 3.480093002319336,
+      "learning_rate": 0.00019997557683242004,
+      "logits/chosen": -0.5985093712806702,
+      "logits/rejected": -0.6007335186004639,
+      "logps/chosen": -14.120916366577148,
+      "logps/rejected": -24.104907989501953,
+      "loss": 0.9003526568412781,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.466304212808609,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.13193026185035706,
+      "rewards/margins": 1.0070805549621582,
+      "rewards/rejected": -0.8751502633094788,
+      "step": 60,
+      "train_speed(iter/s)": 0.005522
+    },
+    {
+      "epoch": 0.05816621472630293,
+      "grad_norm": 3.3445444107055664,
+      "learning_rate": 0.00019996810075006117,
+      "logits/chosen": -0.6064149141311646,
+      "logits/rejected": -0.6075825691223145,
+      "logps/chosen": -10.57797622680664,
+      "logps/rejected": -21.275867462158203,
+      "loss": 0.9624521732330322,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.41281017661094666,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.06739869713783264,
+      "rewards/margins": 0.5023229122161865,
+      "rewards/rejected": -0.43492427468299866,
+      "step": 61,
+      "train_speed(iter/s)": 0.005519
+    },
+    {
+      "epoch": 0.059119759230012815,
+      "grad_norm": 5.101805210113525,
+      "learning_rate": 0.00019995962808197216,
+      "logits/chosen": -0.6019335985183716,
+      "logits/rejected": -0.6061650514602661,
+      "logps/chosen": -12.45846939086914,
+      "logps/rejected": -22.627965927124023,
+      "loss": 1.1021240949630737,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5154472589492798,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.0749291479587555,
+      "rewards/margins": 0.5026906728744507,
+      "rewards/rejected": -0.5776197910308838,
+      "step": 62,
+      "train_speed(iter/s)": 0.005522
+    },
+    {
+      "epoch": 0.060073303733722695,
+      "grad_norm": 3.7086844444274902,
+      "learning_rate": 0.0001999501589126174,
+      "logits/chosen": -0.598800539970398,
+      "logits/rejected": -0.5973880887031555,
+      "logps/chosen": -16.170019149780273,
+      "logps/rejected": -24.100147247314453,
+      "loss": 0.9728513956069946,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4919710159301758,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.14402300119400024,
+      "rewards/margins": 0.7554571628570557,
+      "rewards/rejected": -0.6114341616630554,
+      "step": 63,
+      "train_speed(iter/s)": 0.005522
+    },
+    {
+      "epoch": 0.06102684823743258,
+      "grad_norm": 6.82169246673584,
+      "learning_rate": 0.00019993969333639532,
+      "logits/chosen": -0.5916394591331482,
+      "logits/rejected": -0.5948237180709839,
+      "logps/chosen": -6.504903793334961,
+      "logps/rejected": -23.65030860900879,
+      "loss": 0.8641103506088257,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3815235197544098,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.06883495301008224,
+      "rewards/margins": 0.6778679490089417,
+      "rewards/rejected": -0.6090329885482788,
+      "step": 64,
+      "train_speed(iter/s)": 0.005524
+    },
+    {
+      "epoch": 0.06198039274114246,
+      "grad_norm": 6.41129207611084,
+      "learning_rate": 0.0001999282314576377,
+      "logits/chosen": -0.6117686033248901,
+      "logits/rejected": -0.6133745312690735,
+      "logps/chosen": -7.471805572509766,
+      "logps/rejected": -19.006999969482422,
+      "loss": 0.8599905371665955,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3879605829715729,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.15403376519680023,
+      "rewards/margins": 0.6219788193702698,
+      "rewards/rejected": -0.4679450988769531,
+      "step": 65,
+      "train_speed(iter/s)": 0.005525
+    },
+    {
+      "epoch": 0.06293393724485236,
+      "grad_norm": 3.6126596927642822,
+      "learning_rate": 0.00019991577339060842,
+      "logits/chosen": -0.6491844654083252,
+      "logits/rejected": -0.6502828001976013,
+      "logps/chosen": -9.777487754821777,
+      "logps/rejected": -19.9858455657959,
+      "loss": 0.9987360835075378,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4526561498641968,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.14287549257278442,
+      "rewards/margins": 0.528179407119751,
+      "rewards/rejected": -0.38530388474464417,
+      "step": 66,
+      "train_speed(iter/s)": 0.005524
+    },
+    {
+      "epoch": 0.06388748174856224,
+      "grad_norm": 3.5133907794952393,
+      "learning_rate": 0.0001999023192595024,
+      "logits/chosen": -0.6023973822593689,
+      "logits/rejected": -0.6058337092399597,
+      "logps/chosen": -11.287931442260742,
+      "logps/rejected": -23.294925689697266,
+      "loss": 0.949726939201355,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.49871301651000977,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.25335025787353516,
+      "rewards/margins": 0.7444449663162231,
+      "rewards/rejected": -0.491094708442688,
+      "step": 67,
+      "train_speed(iter/s)": 0.005525
+    },
+    {
+      "epoch": 0.06484102625227212,
+      "grad_norm": 4.242381572723389,
+      "learning_rate": 0.00019988786919844436,
+      "logits/chosen": -0.577968955039978,
+      "logits/rejected": -0.584266185760498,
+      "logps/chosen": -7.122470378875732,
+      "logps/rejected": -23.440967559814453,
+      "loss": 0.825219988822937,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3752996325492859,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.28059279918670654,
+      "rewards/margins": 0.8312418460845947,
+      "rewards/rejected": -0.5506490468978882,
+      "step": 68,
+      "train_speed(iter/s)": 0.005524
+    },
+    {
+      "epoch": 0.065794570755982,
+      "grad_norm": 4.247474670410156,
+      "learning_rate": 0.00019987242335148757,
+      "logits/chosen": -0.6062893271446228,
+      "logits/rejected": -0.6092509031295776,
+      "logps/chosen": -9.965551376342773,
+      "logps/rejected": -16.18874740600586,
+      "loss": 1.0371503829956055,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4371563792228699,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.08624501526355743,
+      "rewards/margins": 0.4600314795970917,
+      "rewards/rejected": -0.37378647923469543,
+      "step": 69,
+      "train_speed(iter/s)": 0.005525
+    },
+    {
+      "epoch": 0.06674811525969189,
+      "grad_norm": 3.118610143661499,
+      "learning_rate": 0.0001998559818726122,
+      "logits/chosen": -0.651878297328949,
+      "logits/rejected": -0.6511794328689575,
+      "logps/chosen": -17.028488159179688,
+      "logps/rejected": -18.37329864501953,
+      "loss": 1.1015472412109375,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5589602589607239,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.17374631762504578,
+      "rewards/margins": 0.5566114187240601,
+      "rewards/rejected": -0.3828650414943695,
+      "step": 70,
+      "train_speed(iter/s)": 0.005526
+    },
+    {
+      "epoch": 0.06770165976340177,
+      "grad_norm": 3.3582799434661865,
+      "learning_rate": 0.00019983854492572394,
+      "logits/chosen": -0.615595281124115,
+      "logits/rejected": -0.6146353483200073,
+      "logps/chosen": -12.519451141357422,
+      "logps/rejected": -18.15428924560547,
+      "loss": 0.9180172085762024,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3709501624107361,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.09964233636856079,
+      "rewards/margins": 0.6540214419364929,
+      "rewards/rejected": -0.5543791651725769,
+      "step": 71,
+      "train_speed(iter/s)": 0.005525
+    },
+    {
+      "epoch": 0.06865520426711165,
+      "grad_norm": 2.883702516555786,
+      "learning_rate": 0.00019982011268465243,
+      "logits/chosen": -0.5868794918060303,
+      "logits/rejected": -0.5905233025550842,
+      "logps/chosen": -14.138279914855957,
+      "logps/rejected": -27.953182220458984,
+      "loss": 1.0747126340866089,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.46605974435806274,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.03510252386331558,
+      "rewards/margins": 0.6189652681350708,
+      "rewards/rejected": -0.5838626623153687,
+      "step": 72,
+      "train_speed(iter/s)": 0.005526
+    },
+    {
+      "epoch": 0.06960874877082154,
+      "grad_norm": 2.9610211849212646,
+      "learning_rate": 0.00019980068533314934,
+      "logits/chosen": -0.6018360257148743,
+      "logits/rejected": -0.6016470789909363,
+      "logps/chosen": -14.77442455291748,
+      "logps/rejected": -16.293825149536133,
+      "loss": 1.1167826652526855,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5493913888931274,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.20179376006126404,
+      "rewards/margins": 0.5395517945289612,
+      "rewards/rejected": -0.3377581536769867,
+      "step": 73,
+      "train_speed(iter/s)": 0.005526
+    },
+    {
+      "epoch": 0.07056229327453142,
+      "grad_norm": 3.1142945289611816,
+      "learning_rate": 0.00019978026306488668,
+      "logits/chosen": -0.5901756286621094,
+      "logits/rejected": -0.5861356258392334,
+      "logps/chosen": -12.717107772827148,
+      "logps/rejected": -19.3599796295166,
+      "loss": 0.9646521210670471,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.42209792137145996,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.19932276010513306,
+      "rewards/margins": 0.5936742424964905,
+      "rewards/rejected": -0.3943515419960022,
+      "step": 74,
+      "train_speed(iter/s)": 0.005525
+    },
+    {
+      "epoch": 0.0715158377782413,
+      "grad_norm": 2.8381545543670654,
+      "learning_rate": 0.00019975884608345476,
+      "logits/chosen": -0.5791196227073669,
+      "logits/rejected": -0.5793704390525818,
+      "logps/chosen": -11.382536888122559,
+      "logps/rejected": -17.99052619934082,
+      "loss": 1.0251520872116089,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5087648630142212,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.225914865732193,
+      "rewards/margins": 0.544399619102478,
+      "rewards/rejected": -0.3184847831726074,
+      "step": 75,
+      "train_speed(iter/s)": 0.005523
+    },
+    {
+      "epoch": 0.0724693822819512,
+      "grad_norm": 2.720726251602173,
+      "learning_rate": 0.0001997364346023603,
+      "logits/chosen": -0.5865166187286377,
+      "logits/rejected": -0.5875246524810791,
+      "logps/chosen": -14.62844467163086,
+      "logps/rejected": -26.72977638244629,
+      "loss": 0.9625909924507141,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5162026882171631,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.2656521499156952,
+      "rewards/margins": 0.8063244819641113,
+      "rewards/rejected": -0.5406723022460938,
+      "step": 76,
+      "train_speed(iter/s)": 0.005522
+    },
+    {
+      "epoch": 0.07342292678566108,
+      "grad_norm": 7.292218208312988,
+      "learning_rate": 0.0001997130288450242,
+      "logits/chosen": -0.5742957592010498,
+      "logits/rejected": -0.5747869610786438,
+      "logps/chosen": -15.201194763183594,
+      "logps/rejected": -21.86297607421875,
+      "loss": 1.2068696022033691,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7135031819343567,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.23197884857654572,
+      "rewards/margins": 0.7994663119316101,
+      "rewards/rejected": -0.5674874782562256,
+      "step": 77,
+      "train_speed(iter/s)": 0.005522
+    },
+    {
+      "epoch": 0.07437647128937096,
+      "grad_norm": 3.270038366317749,
+      "learning_rate": 0.00019968862904477935,
+      "logits/chosen": -0.5858986377716064,
+      "logits/rejected": -0.588924765586853,
+      "logps/chosen": -9.959943771362305,
+      "logps/rejected": -30.939224243164062,
+      "loss": 0.8423591256141663,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.42530590295791626,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.16566459834575653,
+      "rewards/margins": 1.2258960008621216,
+      "rewards/rejected": -1.060231328010559,
+      "step": 78,
+      "train_speed(iter/s)": 0.005518
+    },
+    {
+      "epoch": 0.07533001579308084,
+      "grad_norm": 4.3024821281433105,
+      "learning_rate": 0.00019966323544486818,
+      "logits/chosen": -0.6145219802856445,
+      "logits/rejected": -0.6170623898506165,
+      "logps/chosen": -15.870341300964355,
+      "logps/rejected": -31.39044952392578,
+      "loss": 0.8412056565284729,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4725872874259949,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.3251492381095886,
+      "rewards/margins": 1.2906752824783325,
+      "rewards/rejected": -0.9655259251594543,
+      "step": 79,
+      "train_speed(iter/s)": 0.005516
+    },
+    {
+      "epoch": 0.07628356029679073,
+      "grad_norm": 5.5750732421875,
+      "learning_rate": 0.00019963684829844052,
+      "logits/chosen": -0.6042959094047546,
+      "logits/rejected": -0.6049582362174988,
+      "logps/chosen": -13.382628440856934,
+      "logps/rejected": -21.927295684814453,
+      "loss": 0.993065357208252,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.49278005957603455,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.10083693265914917,
+      "rewards/margins": 0.7778795957565308,
+      "rewards/rejected": -0.6770426630973816,
+      "step": 80,
+      "train_speed(iter/s)": 0.005517
+    },
+    {
+      "epoch": 0.07723710480050061,
+      "grad_norm": 7.816519260406494,
+      "learning_rate": 0.0001996094678685508,
+      "logits/chosen": -0.5581300258636475,
+      "logits/rejected": -0.5570173263549805,
+      "logps/chosen": -16.26945686340332,
+      "logps/rejected": -28.130666732788086,
+      "loss": 1.1555508375167847,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5590921640396118,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -0.22734299302101135,
+      "rewards/margins": 0.652872622013092,
+      "rewards/rejected": -0.8802156448364258,
+      "step": 81,
+      "train_speed(iter/s)": 0.005518
+    },
+    {
+      "epoch": 0.07819064930421049,
+      "grad_norm": 4.555399417877197,
+      "learning_rate": 0.00019958109442815553,
+      "logits/chosen": -0.562245786190033,
+      "logits/rejected": -0.5638425946235657,
+      "logps/chosen": -17.669172286987305,
+      "logps/rejected": -31.455379486083984,
+      "loss": 1.0409901142120361,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5925040245056152,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.18633463978767395,
+      "rewards/margins": 1.1558425426483154,
+      "rewards/rejected": -1.3421772718429565,
+      "step": 82,
+      "train_speed(iter/s)": 0.005517
+    },
+    {
+      "epoch": 0.07914419380792038,
+      "grad_norm": 4.788554668426514,
+      "learning_rate": 0.00019955172826011062,
+      "logits/chosen": -0.53816819190979,
+      "logits/rejected": -0.539014458656311,
+      "logps/chosen": -13.409234046936035,
+      "logps/rejected": -24.46169090270996,
+      "loss": 1.0504765510559082,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5079448819160461,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -0.030599184334278107,
+      "rewards/margins": 0.6909205317497253,
+      "rewards/rejected": -0.7215197086334229,
+      "step": 83,
+      "train_speed(iter/s)": 0.005518
+    },
+    {
+      "epoch": 0.08009773831163026,
+      "grad_norm": 2.9610979557037354,
+      "learning_rate": 0.00019952136965716846,
+      "logits/chosen": -0.5609460473060608,
+      "logits/rejected": -0.5607710480690002,
+      "logps/chosen": -15.43799114227295,
+      "logps/rejected": -20.685916900634766,
+      "loss": 0.9568982720375061,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.49247080087661743,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.2509080469608307,
+      "rewards/margins": 0.9044961333274841,
+      "rewards/rejected": -0.653588056564331,
+      "step": 84,
+      "train_speed(iter/s)": 0.005516
+    },
+    {
+      "epoch": 0.08105128281534014,
+      "grad_norm": 2.7049379348754883,
+      "learning_rate": 0.00019949001892197515,
+      "logits/chosen": -0.5431854128837585,
+      "logits/rejected": -0.5459212064743042,
+      "logps/chosen": -11.692042350769043,
+      "logps/rejected": -25.70047378540039,
+      "loss": 0.9204345941543579,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5127474069595337,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.16273778676986694,
+      "rewards/margins": 1.052358865737915,
+      "rewards/rejected": -0.8896210193634033,
+      "step": 85,
+      "train_speed(iter/s)": 0.005518
+    },
+    {
+      "epoch": 0.08200482731905002,
+      "grad_norm": 2.835888624191284,
+      "learning_rate": 0.00019945767636706728,
+      "logits/chosen": -0.5524388551712036,
+      "logits/rejected": -0.5518356561660767,
+      "logps/chosen": -9.604469299316406,
+      "logps/rejected": -18.570831298828125,
+      "loss": 0.7963011860847473,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.35101598501205444,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.20881126821041107,
+      "rewards/margins": 0.7979729771614075,
+      "rewards/rejected": -0.58916175365448,
+      "step": 86,
+      "train_speed(iter/s)": 0.005516
+    },
+    {
+      "epoch": 0.08295837182275992,
+      "grad_norm": 4.458975315093994,
+      "learning_rate": 0.00019942434231486902,
+      "logits/chosen": -0.5035311579704285,
+      "logits/rejected": -0.5041376352310181,
+      "logps/chosen": -12.906057357788086,
+      "logps/rejected": -21.71013641357422,
+      "loss": 0.9929911494255066,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5292322039604187,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.23509997129440308,
+      "rewards/margins": 0.7336819171905518,
+      "rewards/rejected": -0.49858203530311584,
+      "step": 87,
+      "train_speed(iter/s)": 0.005515
+    },
+    {
+      "epoch": 0.0839119163264698,
+      "grad_norm": 5.752819061279297,
+      "learning_rate": 0.0001993900170976888,
+      "logits/chosen": -0.5372301340103149,
+      "logits/rejected": -0.5387623310089111,
+      "logps/chosen": -12.41868782043457,
+      "logps/rejected": -23.16730499267578,
+      "loss": 0.9672287106513977,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4671844244003296,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.22129932045936584,
+      "rewards/margins": 0.8583503365516663,
+      "rewards/rejected": -0.637050986289978,
+      "step": 88,
+      "train_speed(iter/s)": 0.005516
+    },
+    {
+      "epoch": 0.08486546083017968,
+      "grad_norm": 3.0927765369415283,
+      "learning_rate": 0.00019935470105771598,
+      "logits/chosen": -0.5962197780609131,
+      "logits/rejected": -0.5928555130958557,
+      "logps/chosen": -15.36233139038086,
+      "logps/rejected": -16.967832565307617,
+      "loss": 1.0067566633224487,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5986248850822449,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.20106852054595947,
+      "rewards/margins": 0.8472704291343689,
+      "rewards/rejected": -0.6462019681930542,
+      "step": 89,
+      "train_speed(iter/s)": 0.005517
+    },
+    {
+      "epoch": 0.08581900533388957,
+      "grad_norm": 4.051328659057617,
+      "learning_rate": 0.00019931839454701743,
+      "logits/chosen": -0.5495755672454834,
+      "logits/rejected": -0.54953533411026,
+      "logps/chosen": -9.681739807128906,
+      "logps/rejected": -17.38517951965332,
+      "loss": 0.9176310896873474,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4568890631198883,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.18029510974884033,
+      "rewards/margins": 0.7410764694213867,
+      "rewards/rejected": -0.5607813596725464,
+      "step": 90,
+      "train_speed(iter/s)": 0.005516
+    },
+    {
+      "epoch": 0.08677254983759945,
+      "grad_norm": 2.651430368423462,
+      "learning_rate": 0.00019928109792753418,
+      "logits/chosen": -0.5709578990936279,
+      "logits/rejected": -0.5723389983177185,
+      "logps/chosen": -12.097832679748535,
+      "logps/rejected": -25.470287322998047,
+      "loss": 0.8452543020248413,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4146232008934021,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.2571614384651184,
+      "rewards/margins": 0.9069947600364685,
+      "rewards/rejected": -0.6498332619667053,
+      "step": 91,
+      "train_speed(iter/s)": 0.005515
+    },
+    {
+      "epoch": 0.08772609434130933,
+      "grad_norm": 3.9205963611602783,
+      "learning_rate": 0.0001992428115710776,
+      "logits/chosen": -0.5633900761604309,
+      "logits/rejected": -0.5629602074623108,
+      "logps/chosen": -15.341146469116211,
+      "logps/rejected": -22.676523208618164,
+      "loss": 1.0960146188735962,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.6228011846542358,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.2320929765701294,
+      "rewards/margins": 0.802864670753479,
+      "rewards/rejected": -0.5707716941833496,
+      "step": 92,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.08867963884501923,
+      "grad_norm": 2.4844846725463867,
+      "learning_rate": 0.00019920353585932578,
+      "logits/chosen": -0.5643666982650757,
+      "logits/rejected": -0.5697072148323059,
+      "logps/chosen": -7.112787246704102,
+      "logps/rejected": -29.775066375732422,
+      "loss": 0.7670912742614746,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3726654350757599,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.11930885910987854,
+      "rewards/margins": 1.221017837524414,
+      "rewards/rejected": -1.101709008216858,
+      "step": 93,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.0896331833487291,
+      "grad_norm": 3.3409717082977295,
+      "learning_rate": 0.00019916327118381982,
+      "logits/chosen": -0.5507174730300903,
+      "logits/rejected": -0.5550180673599243,
+      "logps/chosen": -6.935158729553223,
+      "logps/rejected": -26.466045379638672,
+      "loss": 0.8178773522377014,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3691931962966919,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": -0.014170083217322826,
+      "rewards/margins": 0.906183123588562,
+      "rewards/rejected": -0.9203532338142395,
+      "step": 94,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.09058672785243899,
+      "grad_norm": 3.8606507778167725,
+      "learning_rate": 0.0001991220179459597,
+      "logits/chosen": -0.5692034959793091,
+      "logits/rejected": -0.5723109841346741,
+      "logps/chosen": -9.636994361877441,
+      "logps/rejected": -28.736644744873047,
+      "loss": 0.7489418387413025,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.38759085536003113,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.1789592206478119,
+      "rewards/margins": 1.2420873641967773,
+      "rewards/rejected": -1.0631282329559326,
+      "step": 95,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.09154027235614887,
+      "grad_norm": 4.69579553604126,
+      "learning_rate": 0.00019907977655700054,
+      "logits/chosen": -0.5835409760475159,
+      "logits/rejected": -0.5857175588607788,
+      "logps/chosen": -8.839982032775879,
+      "logps/rejected": -26.350032806396484,
+      "loss": 0.8248448371887207,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4012177586555481,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.14620812237262726,
+      "rewards/margins": 1.2952075004577637,
+      "rewards/rejected": -1.1489993333816528,
+      "step": 96,
+      "train_speed(iter/s)": 0.005512
+    },
+    {
+      "epoch": 0.09249381685985876,
+      "grad_norm": 3.4665541648864746,
+      "learning_rate": 0.00019903654743804833,
+      "logits/chosen": -0.6227341294288635,
+      "logits/rejected": -0.6252981424331665,
+      "logps/chosen": -11.844514846801758,
+      "logps/rejected": -29.124967575073242,
+      "loss": 0.8466665744781494,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3924159109592438,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.03820016235113144,
+      "rewards/margins": 1.3871092796325684,
+      "rewards/rejected": -1.425309419631958,
+      "step": 97,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.09344736136356864,
+      "grad_norm": 13.193551063537598,
+      "learning_rate": 0.00019899233102005573,
+      "logits/chosen": -0.5675249695777893,
+      "logits/rejected": -0.571427583694458,
+      "logps/chosen": -11.932504653930664,
+      "logps/rejected": -29.481807708740234,
+      "loss": 1.3480544090270996,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.8751646280288696,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.04504670575261116,
+      "rewards/margins": 1.2038631439208984,
+      "rewards/rejected": -1.248909831047058,
+      "step": 98,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.09440090586727852,
+      "grad_norm": 13.958789825439453,
+      "learning_rate": 0.00019894712774381787,
+      "logits/chosen": -0.5426734685897827,
+      "logits/rejected": -0.5455094575881958,
+      "logps/chosen": -14.97989273071289,
+      "logps/rejected": -29.963254928588867,
+      "loss": 0.9666093587875366,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5926772356033325,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -0.12036357820034027,
+      "rewards/margins": 1.3575993776321411,
+      "rewards/rejected": -1.4779632091522217,
+      "step": 99,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.09535445037098841,
+      "grad_norm": 3.8765738010406494,
+      "learning_rate": 0.00019890093805996793,
+      "logits/chosen": -0.5699120163917542,
+      "logits/rejected": -0.5716601014137268,
+      "logps/chosen": -18.304710388183594,
+      "logps/rejected": -37.15324783325195,
+      "loss": 1.0460050106048584,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.6231843829154968,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -0.23249366879463196,
+      "rewards/margins": 1.3045672178268433,
+      "rewards/rejected": -1.5370607376098633,
+      "step": 100,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.0963079948746983,
+      "grad_norm": 3.2104878425598145,
+      "learning_rate": 0.00019885376242897258,
+      "logits/chosen": -0.5706620812416077,
+      "logits/rejected": -0.5722963213920593,
+      "logps/chosen": -14.833114624023438,
+      "logps/rejected": -30.274829864501953,
+      "loss": 1.1376781463623047,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.653939425945282,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.14056257903575897,
+      "rewards/margins": 1.1054493188858032,
+      "rewards/rejected": -0.9648867845535278,
+      "step": 101,
+      "train_speed(iter/s)": 0.005506
+    },
+    {
+      "epoch": 0.09726153937840817,
+      "grad_norm": 2.326752185821533,
+      "learning_rate": 0.00019880560132112742,
+      "logits/chosen": -0.5489917993545532,
+      "logits/rejected": -0.5558469891548157,
+      "logps/chosen": -10.081119537353516,
+      "logps/rejected": -31.342273712158203,
+      "loss": 0.9489866495132446,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5341838002204895,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.08325084298849106,
+      "rewards/margins": 1.2551041841506958,
+      "rewards/rejected": -1.1718534231185913,
+      "step": 102,
+      "train_speed(iter/s)": 0.005507
+    },
+    {
+      "epoch": 0.09821508388211805,
+      "grad_norm": 2.702747344970703,
+      "learning_rate": 0.0001987564552165524,
+      "logits/chosen": -0.5335314869880676,
+      "logits/rejected": -0.5375723838806152,
+      "logps/chosen": -6.767879962921143,
+      "logps/rejected": -26.971349716186523,
+      "loss": 0.7666488885879517,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3699600100517273,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.2037293165922165,
+      "rewards/margins": 1.2131260633468628,
+      "rewards/rejected": -1.0093967914581299,
+      "step": 103,
+      "train_speed(iter/s)": 0.005507
+    },
+    {
+      "epoch": 0.09916862838582795,
+      "grad_norm": 3.298539638519287,
+      "learning_rate": 0.0001987063246051868,
+      "logits/chosen": -0.5586405992507935,
+      "logits/rejected": -0.5575414299964905,
+      "logps/chosen": -10.250675201416016,
+      "logps/rejected": -19.98495864868164,
+      "loss": 0.9064714312553406,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.38871869444847107,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.07911191880702972,
+      "rewards/margins": 0.7202980518341064,
+      "rewards/rejected": -0.6411861181259155,
+      "step": 104,
+      "train_speed(iter/s)": 0.005508
+    },
+    {
+      "epoch": 0.10012217288953783,
+      "grad_norm": 3.7112255096435547,
+      "learning_rate": 0.00019865520998678458,
+      "logits/chosen": -0.5400758385658264,
+      "logits/rejected": -0.5405173301696777,
+      "logps/chosen": -16.44561004638672,
+      "logps/rejected": -24.691865921020508,
+      "loss": 1.1246851682662964,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.6655758023262024,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.11685307323932648,
+      "rewards/margins": 0.8305923938751221,
+      "rewards/rejected": -0.713739275932312,
+      "step": 105,
+      "train_speed(iter/s)": 0.005507
+    },
+    {
+      "epoch": 0.10107571739324771,
+      "grad_norm": 6.40952205657959,
+      "learning_rate": 0.0001986031118709093,
+      "logits/chosen": -0.548525333404541,
+      "logits/rejected": -0.5509157180786133,
+      "logps/chosen": -11.731019020080566,
+      "logps/rejected": -22.46083641052246,
+      "loss": 1.37913179397583,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.7638791799545288,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -0.029659561812877655,
+      "rewards/margins": 0.5513206124305725,
+      "rewards/rejected": -0.5809801816940308,
+      "step": 106,
+      "train_speed(iter/s)": 0.005505
+    },
+    {
+      "epoch": 0.1020292618969576,
+      "grad_norm": 5.338200092315674,
+      "learning_rate": 0.00019855003077692897,
+      "logits/chosen": -0.5491806268692017,
+      "logits/rejected": -0.5485914945602417,
+      "logps/chosen": -10.374083518981934,
+      "logps/rejected": -22.91812515258789,
+      "loss": 0.9197418689727783,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.42541149258613586,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.23835767805576324,
+      "rewards/margins": 0.8718117475509644,
+      "rewards/rejected": -0.6334540843963623,
+      "step": 107,
+      "train_speed(iter/s)": 0.005505
+    },
+    {
+      "epoch": 0.10298280640066748,
+      "grad_norm": 4.72066593170166,
+      "learning_rate": 0.00019849596723401107,
+      "logits/chosen": -0.5481430888175964,
+      "logits/rejected": -0.552554190158844,
+      "logps/chosen": -14.082160949707031,
+      "logps/rejected": -25.61027717590332,
+      "loss": 1.2021526098251343,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.6988036036491394,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.09246861189603806,
+      "rewards/margins": 0.8137391805648804,
+      "rewards/rejected": -0.7212705612182617,
+      "step": 108,
+      "train_speed(iter/s)": 0.005507
+    },
+    {
+      "epoch": 0.10393635090437736,
+      "grad_norm": 2.7633321285247803,
+      "learning_rate": 0.00019844092178111702,
+      "logits/chosen": -0.508384108543396,
+      "logits/rejected": -0.5140159130096436,
+      "logps/chosen": -6.925086975097656,
+      "logps/rejected": -24.077953338623047,
+      "loss": 0.8022588491439819,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.41396474838256836,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.16561280190944672,
+      "rewards/margins": 1.0861924886703491,
+      "rewards/rejected": -0.9205797910690308,
+      "step": 109,
+      "train_speed(iter/s)": 0.005509
+    },
+    {
+      "epoch": 0.10488989540808726,
+      "grad_norm": 2.4014365673065186,
+      "learning_rate": 0.00019838489496699704,
+      "logits/chosen": -0.5266930460929871,
+      "logits/rejected": -0.5241900682449341,
+      "logps/chosen": -15.143444061279297,
+      "logps/rejected": -18.07321548461914,
+      "loss": 0.8957412242889404,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5194482803344727,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.33800825476646423,
+      "rewards/margins": 1.0479741096496582,
+      "rewards/rejected": -0.7099659442901611,
+      "step": 110,
+      "train_speed(iter/s)": 0.005508
+    },
+    {
+      "epoch": 0.10584343991179714,
+      "grad_norm": 3.5473034381866455,
+      "learning_rate": 0.0001983278873501845,
+      "logits/chosen": -0.5651346445083618,
+      "logits/rejected": -0.5652989149093628,
+      "logps/chosen": -13.63066291809082,
+      "logps/rejected": -21.72479820251465,
+      "loss": 1.070034146308899,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5245512127876282,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.1012478768825531,
+      "rewards/margins": 0.5533490180969238,
+      "rewards/rejected": -0.4521011710166931,
+      "step": 111,
+      "train_speed(iter/s)": 0.005508
+    },
+    {
+      "epoch": 0.10679698441550702,
+      "grad_norm": 2.844266653060913,
+      "learning_rate": 0.00019826989949899048,
+      "logits/chosen": -0.5871210694313049,
+      "logits/rejected": -0.5863486528396606,
+      "logps/chosen": -10.202323913574219,
+      "logps/rejected": -17.968069076538086,
+      "loss": 0.9712742567062378,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4458525776863098,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.31141582131385803,
+      "rewards/margins": 0.6074751615524292,
+      "rewards/rejected": -0.29605934023857117,
+      "step": 112,
+      "train_speed(iter/s)": 0.005507
+    },
+    {
+      "epoch": 0.1077505289192169,
+      "grad_norm": 3.325552225112915,
+      "learning_rate": 0.00019821093199149804,
+      "logits/chosen": -0.5867654085159302,
+      "logits/rejected": -0.590225338935852,
+      "logps/chosen": -14.67109489440918,
+      "logps/rejected": -31.97191619873047,
+      "loss": 0.8784574270248413,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5371198058128357,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.31997063755989075,
+      "rewards/margins": 1.1458133459091187,
+      "rewards/rejected": -0.8258427381515503,
+      "step": 113,
+      "train_speed(iter/s)": 0.005509
+    },
+    {
+      "epoch": 0.10870407342292679,
+      "grad_norm": 2.5996289253234863,
+      "learning_rate": 0.00019815098541555646,
+      "logits/chosen": -0.6402957439422607,
+      "logits/rejected": -0.6378954648971558,
+      "logps/chosen": -12.175540924072266,
+      "logps/rejected": -20.870878219604492,
+      "loss": 0.9693875908851624,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5651832818984985,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.3929046392440796,
+      "rewards/margins": 1.0991772413253784,
+      "rewards/rejected": -0.7062726020812988,
+      "step": 114,
+      "train_speed(iter/s)": 0.005509
+    },
+    {
+      "epoch": 0.10965761792663667,
+      "grad_norm": 2.5734198093414307,
+      "learning_rate": 0.00019809006036877538,
+      "logits/chosen": -0.6097349524497986,
+      "logits/rejected": -0.6098358035087585,
+      "logps/chosen": -11.490409851074219,
+      "logps/rejected": -15.817056655883789,
+      "loss": 0.9781900644302368,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.502380907535553,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.2869223654270172,
+      "rewards/margins": 0.795809268951416,
+      "rewards/rejected": -0.5088868737220764,
+      "step": 115,
+      "train_speed(iter/s)": 0.00551
+    },
+    {
+      "epoch": 0.11061116243034655,
+      "grad_norm": 2.6200523376464844,
+      "learning_rate": 0.00019802815745851885,
+      "logits/chosen": -0.6539290547370911,
+      "logits/rejected": -0.6565437912940979,
+      "logps/chosen": -13.780235290527344,
+      "logps/rejected": -29.117767333984375,
+      "loss": 0.8663976192474365,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.406424343585968,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.24194149672985077,
+      "rewards/margins": 1.0130045413970947,
+      "rewards/rejected": -0.7710630893707275,
+      "step": 116,
+      "train_speed(iter/s)": 0.005509
+    },
+    {
+      "epoch": 0.11156470693405644,
+      "grad_norm": 2.858734130859375,
+      "learning_rate": 0.00019796527730189936,
+      "logits/chosen": -0.6368309259414673,
+      "logits/rejected": -0.6411804556846619,
+      "logps/chosen": -12.20506477355957,
+      "logps/rejected": -22.50167465209961,
+      "loss": 0.9522262811660767,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5067539215087891,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.35891181230545044,
+      "rewards/margins": 0.9037349224090576,
+      "rewards/rejected": -0.5448231101036072,
+      "step": 117,
+      "train_speed(iter/s)": 0.005508
+    },
+    {
+      "epoch": 0.11251825143776632,
+      "grad_norm": 3.5055670738220215,
+      "learning_rate": 0.0001979014205257715,
+      "logits/chosen": -0.6525743007659912,
+      "logits/rejected": -0.6540606021881104,
+      "logps/chosen": -15.675086975097656,
+      "logps/rejected": -24.1904239654541,
+      "loss": 1.0509792566299438,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5997871160507202,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.07162542641162872,
+      "rewards/margins": 1.1154685020446777,
+      "rewards/rejected": -1.0438430309295654,
+      "step": 118,
+      "train_speed(iter/s)": 0.005507
+    },
+    {
+      "epoch": 0.1134717959414762,
+      "grad_norm": 3.984374761581421,
+      "learning_rate": 0.0001978365877667258,
+      "logits/chosen": -0.6372765302658081,
+      "logits/rejected": -0.6411250233650208,
+      "logps/chosen": -10.17730712890625,
+      "logps/rejected": -22.074499130249023,
+      "loss": 1.0653222799301147,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.6034587025642395,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.198074609041214,
+      "rewards/margins": 1.0864827632904053,
+      "rewards/rejected": -0.8884081244468689,
+      "step": 119,
+      "train_speed(iter/s)": 0.005508
+    },
+    {
+      "epoch": 0.1144253404451861,
+      "grad_norm": 3.182776689529419,
+      "learning_rate": 0.00019777077967108255,
+      "logits/chosen": -0.6417007446289062,
+      "logits/rejected": -0.6392112970352173,
+      "logps/chosen": -19.301258087158203,
+      "logps/rejected": -18.364166259765625,
+      "loss": 1.108437180519104,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5851361751556396,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.22438707947731018,
+      "rewards/margins": 0.5671754479408264,
+      "rewards/rejected": -0.3427883982658386,
+      "step": 120,
+      "train_speed(iter/s)": 0.005508
+    },
+    {
+      "epoch": 0.11537888494889598,
+      "grad_norm": 3.537256956100464,
+      "learning_rate": 0.00019770399689488506,
+      "logits/chosen": -0.6460685729980469,
+      "logits/rejected": -0.6467642784118652,
+      "logps/chosen": -13.258935928344727,
+      "logps/rejected": -27.513343811035156,
+      "loss": 0.9472922682762146,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5081149935722351,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.23222729563713074,
+      "rewards/margins": 1.276731014251709,
+      "rewards/rejected": -1.0445036888122559,
+      "step": 121,
+      "train_speed(iter/s)": 0.005508
+    },
+    {
+      "epoch": 0.11633242945260586,
+      "grad_norm": 3.096590280532837,
+      "learning_rate": 0.00019763624010389334,
+      "logits/chosen": -0.6088275909423828,
+      "logits/rejected": -0.6121100783348083,
+      "logps/chosen": -13.26574420928955,
+      "logps/rejected": -19.00273323059082,
+      "loss": 0.9677691459655762,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.507895290851593,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.22311073541641235,
+      "rewards/margins": 0.7144712209701538,
+      "rewards/rejected": -0.49136048555374146,
+      "step": 122,
+      "train_speed(iter/s)": 0.005509
+    },
+    {
+      "epoch": 0.11728597395631574,
+      "grad_norm": 2.810216188430786,
+      "learning_rate": 0.0001975675099735774,
+      "logits/chosen": -0.6021947264671326,
+      "logits/rejected": -0.6030389070510864,
+      "logps/chosen": -8.321534156799316,
+      "logps/rejected": -17.207504272460938,
+      "loss": 0.8276495933532715,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3906245231628418,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.21257750689983368,
+      "rewards/margins": 0.894737958908081,
+      "rewards/rejected": -0.682160496711731,
+      "step": 123,
+      "train_speed(iter/s)": 0.005509
+    },
+    {
+      "epoch": 0.11823951846002563,
+      "grad_norm": 2.6153383255004883,
+      "learning_rate": 0.0001974978071891104,
+      "logits/chosen": -0.6311705112457275,
+      "logits/rejected": -0.6307083964347839,
+      "logps/chosen": -10.833940505981445,
+      "logps/rejected": -22.41739273071289,
+      "loss": 0.8693855404853821,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4562944173812866,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.15080919861793518,
+      "rewards/margins": 1.2266969680786133,
+      "rewards/rejected": -1.0758877992630005,
+      "step": 124,
+      "train_speed(iter/s)": 0.00551
+    },
+    {
+      "epoch": 0.11919306296373551,
+      "grad_norm": 3.493828535079956,
+      "learning_rate": 0.00019742713244536204,
+      "logits/chosen": -0.6434034109115601,
+      "logits/rejected": -0.6431565284729004,
+      "logps/chosen": -18.75763702392578,
+      "logps/rejected": -25.821609497070312,
+      "loss": 1.0967546701431274,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5587143898010254,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.0902448296546936,
+      "rewards/margins": 0.9879235029220581,
+      "rewards/rejected": -0.8976786732673645,
+      "step": 125,
+      "train_speed(iter/s)": 0.00551
+    },
+    {
+      "epoch": 0.12014660746744539,
+      "grad_norm": 2.5747878551483154,
+      "learning_rate": 0.00019735548644689147,
+      "logits/chosen": -0.6225179433822632,
+      "logits/rejected": -0.6244460940361023,
+      "logps/chosen": -15.244443893432617,
+      "logps/rejected": -17.300045013427734,
+      "loss": 1.0364809036254883,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5735288858413696,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.416117787361145,
+      "rewards/margins": 0.7515360713005066,
+      "rewards/rejected": -0.3354182839393616,
+      "step": 126,
+      "train_speed(iter/s)": 0.00551
+    },
+    {
+      "epoch": 0.12110015197115528,
+      "grad_norm": 3.198167562484741,
+      "learning_rate": 0.00019728286990794037,
+      "logits/chosen": -0.6006501317024231,
+      "logits/rejected": -0.6046801805496216,
+      "logps/chosen": -14.4312162399292,
+      "logps/rejected": -23.69890785217285,
+      "loss": 1.0553488731384277,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5887905955314636,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.162513867020607,
+      "rewards/margins": 0.9815673232078552,
+      "rewards/rejected": -0.8190534114837646,
+      "step": 127,
+      "train_speed(iter/s)": 0.00551
+    },
+    {
+      "epoch": 0.12205369647486516,
+      "grad_norm": 3.224283456802368,
+      "learning_rate": 0.00019720928355242568,
+      "logits/chosen": -0.5655272006988525,
+      "logits/rejected": -0.5726292133331299,
+      "logps/chosen": -13.809501647949219,
+      "logps/rejected": -30.130277633666992,
+      "loss": 0.8311694264411926,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3935176134109497,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.027349811047315598,
+      "rewards/margins": 1.382154941558838,
+      "rewards/rejected": -1.3548052310943604,
+      "step": 128,
+      "train_speed(iter/s)": 0.00551
+    },
+    {
+      "epoch": 0.12300724097857504,
+      "grad_norm": 2.575568199157715,
+      "learning_rate": 0.00019713472811393258,
+      "logits/chosen": -0.5765466094017029,
+      "logits/rejected": -0.5786842107772827,
+      "logps/chosen": -16.621131896972656,
+      "logps/rejected": -26.19173240661621,
+      "loss": 0.9337475895881653,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5804612636566162,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.37209561467170715,
+      "rewards/margins": 1.4012634754180908,
+      "rewards/rejected": -1.0291677713394165,
+      "step": 129,
+      "train_speed(iter/s)": 0.005512
+    },
+    {
+      "epoch": 0.12396078548228492,
+      "grad_norm": 3.0303828716278076,
+      "learning_rate": 0.00019705920433570694,
+      "logits/chosen": -0.5693878531455994,
+      "logits/rejected": -0.5738221406936646,
+      "logps/chosen": -7.376468658447266,
+      "logps/rejected": -25.63204574584961,
+      "loss": 0.9063543081283569,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5075121521949768,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.23270781338214874,
+      "rewards/margins": 1.1151927709579468,
+      "rewards/rejected": -0.8824849128723145,
+      "step": 130,
+      "train_speed(iter/s)": 0.005511
+    },
+    {
+      "epoch": 0.12491432998599482,
+      "grad_norm": 4.375239849090576,
+      "learning_rate": 0.0001969827129706482,
+      "logits/chosen": -0.5413441061973572,
+      "logits/rejected": -0.5445973873138428,
+      "logps/chosen": -9.677428245544434,
+      "logps/rejected": -23.95619010925293,
+      "loss": 0.982276439666748,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5150867104530334,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.23964405059814453,
+      "rewards/margins": 0.8764944076538086,
+      "rewards/rejected": -0.6368504762649536,
+      "step": 131,
+      "train_speed(iter/s)": 0.005511
+    },
+    {
+      "epoch": 0.1258678744897047,
+      "grad_norm": 2.136566162109375,
+      "learning_rate": 0.0001969052547813016,
+      "logits/chosen": -0.5252913236618042,
+      "logits/rejected": -0.5288749933242798,
+      "logps/chosen": -8.76069450378418,
+      "logps/rejected": -24.643917083740234,
+      "loss": 0.6882707476615906,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3399735689163208,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.3247303366661072,
+      "rewards/margins": 1.320786952972412,
+      "rewards/rejected": -0.9960566759109497,
+      "step": 132,
+      "train_speed(iter/s)": 0.005512
+    },
+    {
+      "epoch": 0.1268214189934146,
+      "grad_norm": 2.7579689025878906,
+      "learning_rate": 0.00019682683053985072,
+      "logits/chosen": -0.4897473454475403,
+      "logits/rejected": -0.49097952246665955,
+      "logps/chosen": -14.222724914550781,
+      "logps/rejected": -26.548240661621094,
+      "loss": 0.8395786881446838,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.49962103366851807,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.3120686113834381,
+      "rewards/margins": 1.3334513902664185,
+      "rewards/rejected": -1.0213829278945923,
+      "step": 133,
+      "train_speed(iter/s)": 0.005511
+    },
+    {
+      "epoch": 0.12777496349712447,
+      "grad_norm": 4.233667850494385,
+      "learning_rate": 0.00019674744102810978,
+      "logits/chosen": -0.49640408158302307,
+      "logits/rejected": -0.49791407585144043,
+      "logps/chosen": -11.952919006347656,
+      "logps/rejected": -20.992504119873047,
+      "loss": 0.9811766147613525,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.6130494475364685,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.26505932211875916,
+      "rewards/margins": 1.113063097000122,
+      "rewards/rejected": -0.8480038046836853,
+      "step": 134,
+      "train_speed(iter/s)": 0.005511
+    },
+    {
+      "epoch": 0.12872850800083435,
+      "grad_norm": 4.032670974731445,
+      "learning_rate": 0.00019666708703751576,
+      "logits/chosen": -0.5555613040924072,
+      "logits/rejected": -0.5562028288841248,
+      "logps/chosen": -12.461511611938477,
+      "logps/rejected": -29.6381893157959,
+      "loss": 0.9406546354293823,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5542497634887695,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.14334315061569214,
+      "rewards/margins": 1.2391164302825928,
+      "rewards/rejected": -1.0957733392715454,
+      "step": 135,
+      "train_speed(iter/s)": 0.005511
+    },
+    {
+      "epoch": 0.12968205250454423,
+      "grad_norm": 2.8314244747161865,
+      "learning_rate": 0.00019658576936912057,
+      "logits/chosen": -0.5166746973991394,
+      "logits/rejected": -0.5175175666809082,
+      "logps/chosen": -11.453944206237793,
+      "logps/rejected": -32.48845672607422,
+      "loss": 0.741032600402832,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.39420661330223083,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.2297428399324417,
+      "rewards/margins": 1.5928231477737427,
+      "rewards/rejected": -1.3630802631378174,
+      "step": 136,
+      "train_speed(iter/s)": 0.005511
+    },
+    {
+      "epoch": 0.1306355970082541,
+      "grad_norm": 3.786283016204834,
+      "learning_rate": 0.00019650348883358315,
+      "logits/chosen": -0.5285561084747314,
+      "logits/rejected": -0.529099702835083,
+      "logps/chosen": -14.34618091583252,
+      "logps/rejected": -24.930580139160156,
+      "loss": 1.0361278057098389,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.598253607749939,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.2112821340560913,
+      "rewards/margins": 1.2288541793823242,
+      "rewards/rejected": -1.017572045326233,
+      "step": 137,
+      "train_speed(iter/s)": 0.005512
+    },
+    {
+      "epoch": 0.131589141511964,
+      "grad_norm": 3.4468750953674316,
+      "learning_rate": 0.00019642024625116117,
+      "logits/chosen": -0.5355381369590759,
+      "logits/rejected": -0.5351199507713318,
+      "logps/chosen": -14.2181396484375,
+      "logps/rejected": -21.662395477294922,
+      "loss": 0.9722803831100464,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5212066769599915,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.12246156483888626,
+      "rewards/margins": 1.046801209449768,
+      "rewards/rejected": -0.9243396520614624,
+      "step": 138,
+      "train_speed(iter/s)": 0.005512
+    },
+    {
+      "epoch": 0.1325426860156739,
+      "grad_norm": 4.629439830780029,
+      "learning_rate": 0.0001963360424517031,
+      "logits/chosen": -0.5707206726074219,
+      "logits/rejected": -0.5748513340950012,
+      "logps/chosen": -17.805238723754883,
+      "logps/rejected": -29.64284896850586,
+      "loss": 1.254711389541626,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.743948221206665,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.11810468882322311,
+      "rewards/margins": 0.9775583744049072,
+      "rewards/rejected": -0.8594536185264587,
+      "step": 139,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.13349623051938378,
+      "grad_norm": 4.580717086791992,
+      "learning_rate": 0.0001962508782746397,
+      "logits/chosen": -0.580274224281311,
+      "logits/rejected": -0.581068217754364,
+      "logps/chosen": -16.117170333862305,
+      "logps/rejected": -23.330665588378906,
+      "loss": 0.9780559539794922,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.48741835355758667,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.07310809940099716,
+      "rewards/margins": 0.9340574741363525,
+      "rewards/rejected": -0.8609493374824524,
+      "step": 140,
+      "train_speed(iter/s)": 0.005514
+    },
+    {
+      "epoch": 0.13444977502309366,
+      "grad_norm": 3.511350631713867,
+      "learning_rate": 0.0001961647545689759,
+      "logits/chosen": -0.5712682604789734,
+      "logits/rejected": -0.5709258317947388,
+      "logps/chosen": -14.216093063354492,
+      "logps/rejected": -23.430875778198242,
+      "loss": 1.023219108581543,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5690780878067017,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.3575439751148224,
+      "rewards/margins": 0.9377648830413818,
+      "rewards/rejected": -0.5802209973335266,
+      "step": 141,
+      "train_speed(iter/s)": 0.005514
+    },
+    {
+      "epoch": 0.13540331952680354,
+      "grad_norm": 2.3146886825561523,
+      "learning_rate": 0.0001960776721932821,
+      "logits/chosen": -0.5722199082374573,
+      "logits/rejected": -0.5763861536979675,
+      "logps/chosen": -7.839071273803711,
+      "logps/rejected": -18.26845932006836,
+      "loss": 0.810951828956604,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3562580943107605,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": 0.24190367758274078,
+      "rewards/margins": 0.711327850818634,
+      "rewards/rejected": -0.4694240987300873,
+      "step": 142,
+      "train_speed(iter/s)": 0.005514
+    },
+    {
+      "epoch": 0.13635686403051342,
+      "grad_norm": 3.1744766235351562,
+      "learning_rate": 0.00019598963201568573,
+      "logits/chosen": -0.5400611758232117,
+      "logits/rejected": -0.5411737561225891,
+      "logps/chosen": -13.320467948913574,
+      "logps/rejected": -24.537696838378906,
+      "loss": 1.0167641639709473,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.47375380992889404,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.18297314643859863,
+      "rewards/margins": 0.6672495007514954,
+      "rewards/rejected": -0.48427632451057434,
+      "step": 143,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.1373104085342233,
+      "grad_norm": 2.385725259780884,
+      "learning_rate": 0.0001959006349138626,
+      "logits/chosen": -0.5638188123703003,
+      "logits/rejected": -0.5665206909179688,
+      "logps/chosen": -12.801736831665039,
+      "logps/rejected": -24.967641830444336,
+      "loss": 0.9001826047897339,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.45819178223609924,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.24550366401672363,
+      "rewards/margins": 0.9985946416854858,
+      "rewards/rejected": -0.7530909776687622,
+      "step": 144,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.13826395303793318,
+      "grad_norm": 4.4898362159729,
+      "learning_rate": 0.00019581068177502826,
+      "logits/chosen": -0.5841139554977417,
+      "logits/rejected": -0.5885876417160034,
+      "logps/chosen": -8.781448364257812,
+      "logps/rejected": -21.54075813293457,
+      "loss": 0.9311922192573547,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.46065330505371094,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.2900657057762146,
+      "rewards/margins": 0.7273497581481934,
+      "rewards/rejected": -0.43728405237197876,
+      "step": 145,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.1392174975416431,
+      "grad_norm": 2.794843912124634,
+      "learning_rate": 0.00019571977349592878,
+      "logits/chosen": -0.5932512879371643,
+      "logits/rejected": -0.5940472483634949,
+      "logps/chosen": -9.732755661010742,
+      "logps/rejected": -20.826379776000977,
+      "loss": 0.945945680141449,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.47954535484313965,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.2869046926498413,
+      "rewards/margins": 0.9880754947662354,
+      "rewards/rejected": -0.7011707425117493,
+      "step": 146,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.14017104204535297,
+      "grad_norm": 2.540703058242798,
+      "learning_rate": 0.00019562791098283225,
+      "logits/chosen": -0.6081781983375549,
+      "logits/rejected": -0.614071786403656,
+      "logps/chosen": -12.527294158935547,
+      "logps/rejected": -29.05521011352539,
+      "loss": 0.9200064539909363,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.4640219807624817,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.25594115257263184,
+      "rewards/margins": 0.8903652429580688,
+      "rewards/rejected": -0.634424090385437,
+      "step": 147,
+      "train_speed(iter/s)": 0.005512
+    },
+    {
+      "epoch": 0.14112458654906285,
+      "grad_norm": 2.9323384761810303,
+      "learning_rate": 0.0001955350951515195,
+      "logits/chosen": -0.6128370761871338,
+      "logits/rejected": -0.6138598322868347,
+      "logps/chosen": -12.749382019042969,
+      "logps/rejected": -25.719005584716797,
+      "loss": 0.9213395118713379,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.43978482484817505,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.32896676659584045,
+      "rewards/margins": 1.2176496982574463,
+      "rewards/rejected": -0.8886829018592834,
+      "step": 148,
+      "train_speed(iter/s)": 0.005513
+    },
+    {
+      "epoch": 0.14207813105277273,
+      "grad_norm": 4.304955959320068,
+      "learning_rate": 0.00019544132692727497,
+      "logits/chosen": -0.6183297634124756,
+      "logits/rejected": -0.6181198358535767,
+      "logps/chosen": -13.527120590209961,
+      "logps/rejected": -16.62553596496582,
+      "loss": 1.1245884895324707,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.5154122114181519,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": 0.12139306217432022,
+      "rewards/margins": 0.5662888884544373,
+      "rewards/rejected": -0.44489583373069763,
+      "step": 149,
+      "train_speed(iter/s)": 0.005512
+    },
+    {
+      "epoch": 0.1430316755564826,
+      "grad_norm": 3.4540951251983643,
+      "learning_rate": 0.00019534660724487764,
+      "logits/chosen": -0.6565526127815247,
+      "logits/rejected": -0.6572035551071167,
+      "logps/chosen": -8.482949256896973,
+      "logps/rejected": -25.549943923950195,
+      "loss": 0.6943767070770264,
+      "memory(GiB)": 46.35,
+      "nll_loss": 0.3159071207046509,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.13372917473316193,
+      "rewards/margins": 1.538865327835083,
+      "rewards/rejected": -1.4051361083984375,
+      "step": 150,
+      "train_speed(iter/s)": 0.005513
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 1048,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "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": 1.0844345149874504e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd5d2ba79d7611fe1fd77fd7b56c5388376cba0a585f422620fbd3f2425da83f
+size 8696

zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/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 gc
+import json
+import numpy as np
+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, weights_only=False)
+
+        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 tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # 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}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] 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")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+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 = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * 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
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[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}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        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 to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=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
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    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.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    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")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+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,
+                                                          lazy_mode=True)
+
+    # 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")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_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 by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # 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)