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|gate_proj|o_proj|up_proj|k_proj|v_proj|q_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 sha256:877b64be7d79ab80c6f8c771abacd982c2c37c3397db1e5180ead3a725e32462
+size 52511776

additional_config.json

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

args.json

@@ -0,0 +1,440 @@
+{
+  "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": "You are a helpful assistant that truthfully answers user questions about the provided image.",
+  "max_length": 2048,
+  "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": [
+    "json_files/task1_dpo_round2_0dot9_0603.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": "/data2/kdd/crag/cjz/output/task1_dpo_absgt10_noidk__pari0.5_0602_p2prompt/v1-20250603-021349/checkpoint-200",
+  "load_dataset_config": null,
+  "lora_modules": [],
+  "tuner_backend": "peft",
+  "train_type": "lora",
+  "adapters": [
+    "/data2/kdd/crag/cjz/output/task1_dpo_absgt10_noidk__pari0.5_0602_p2prompt/v1-20250603-021349/checkpoint-200"
+  ],
+  "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/task1_dpo_round2_0dot9_0603/v1-20250603-223558",
+  "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": 2,
+  "per_device_eval_batch_size": 2,
+  "per_gpu_train_batch_size": null,
+  "per_gpu_eval_batch_size": null,
+  "gradient_accumulation_steps": 16,
+  "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": 3.0,
+  "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/task1_dpo_round2_0dot9_0603/v1-20250603-223558/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": "/data2/kdd/crag/cjz/output/task1_dpo_absgt10_noidk__pari0.5_0602_p2prompt/v1-20250603-021349/checkpoint-200",
+  "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": true,
+  "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 0x7174495470a0>, 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/task1_dpo_round2_0dot9_0603/v1-20250603-223558', 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=2, per_device_eval_batch_size=2, per_gpu_train_batch_size=None, per_gpu_eval_batch_size=None, gradient_accumulation_steps=16, 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=3.0, 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/task1_dpo_round2_0dot9_0603/v1-20250603-223558/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/task1_dpo_round2_0dot9_0603/v1-20250603-223558', 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='/data2/kdd/crag/cjz/output/task1_dpo_absgt10_noidk__pari0.5_0602_p2prompt/v1-20250603-021349/checkpoint-200', 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=2048, 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 sha256:144a71cf3ee27ab02ab4817315e219784a2027923546fabe822e1b16d032b3df
+size 1064

trainer_state.json

@@ -0,0 +1,2734 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.8648648648648649,
+  "eval_steps": 300,
+  "global_step": 150,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.005765765765765766,
+      "grad_norm": 2.1474921703338623,
+      "learning_rate": 7.692307692307694e-06,
+      "logits/chosen": -0.4285973310470581,
+      "logits/rejected": -0.721319854259491,
+      "logps/chosen": -2.468954086303711,
+      "logps/rejected": -59.87730407714844,
+      "loss": 0.22946561872959137,
+      "memory(GiB)": 30.65,
+      "nll_loss": 0.13756254315376282,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.23724405467510223,
+      "rewards/margins": 3.9806408882141113,
+      "rewards/rejected": -3.743396759033203,
+      "step": 1,
+      "train_speed(iter/s)": 0.01022
+    },
+    {
+      "epoch": 0.011531531531531532,
+      "grad_norm": 2.328110456466675,
+      "learning_rate": 1.5384615384615387e-05,
+      "logits/chosen": -0.3982561528682709,
+      "logits/rejected": -0.7214521169662476,
+      "logps/chosen": -3.335645914077759,
+      "logps/rejected": -60.79209899902344,
+      "loss": 0.3770241439342499,
+      "memory(GiB)": 30.65,
+      "nll_loss": 0.23075783252716064,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.10687967389822006,
+      "rewards/margins": 3.9392333030700684,
+      "rewards/rejected": -3.832353353500366,
+      "step": 2,
+      "train_speed(iter/s)": 0.010488
+    },
+    {
+      "epoch": 0.017297297297297298,
+      "grad_norm": 14.568985939025879,
+      "learning_rate": 2.307692307692308e-05,
+      "logits/chosen": -0.43175598978996277,
+      "logits/rejected": -0.6921666264533997,
+      "logps/chosen": -5.190546035766602,
+      "logps/rejected": -51.33646011352539,
+      "loss": 0.9886025786399841,
+      "memory(GiB)": 34.1,
+      "nll_loss": 0.6371194124221802,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": -0.034887176007032394,
+      "rewards/margins": 3.2021734714508057,
+      "rewards/rejected": -3.237060785293579,
+      "step": 3,
+      "train_speed(iter/s)": 0.010581
+    },
+    {
+      "epoch": 0.023063063063063063,
+      "grad_norm": 13.817249298095703,
+      "learning_rate": 3.0769230769230774e-05,
+      "logits/chosen": -0.4616771340370178,
+      "logits/rejected": -0.6264235973358154,
+      "logps/chosen": -3.2552552223205566,
+      "logps/rejected": -51.25586700439453,
+      "loss": 0.5021163821220398,
+      "memory(GiB)": 37.58,
+      "nll_loss": 0.26375898718833923,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.06816040724515915,
+      "rewards/margins": 3.5297036170959473,
+      "rewards/rejected": -3.461543083190918,
+      "step": 4,
+      "train_speed(iter/s)": 0.010625
+    },
+    {
+      "epoch": 0.02882882882882883,
+      "grad_norm": 2.363553524017334,
+      "learning_rate": 3.846153846153846e-05,
+      "logits/chosen": -0.48805952072143555,
+      "logits/rejected": -0.7040677070617676,
+      "logps/chosen": -6.89398717880249,
+      "logps/rejected": -59.37879943847656,
+      "loss": 0.39657193422317505,
+      "memory(GiB)": 37.58,
+      "nll_loss": 0.2818652391433716,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.1999940723180771,
+      "rewards/margins": 4.230030059814453,
+      "rewards/rejected": -4.030035495758057,
+      "step": 5,
+      "train_speed(iter/s)": 0.010659
+    },
+    {
+      "epoch": 0.034594594594594595,
+      "grad_norm": 4.551424026489258,
+      "learning_rate": 4.615384615384616e-05,
+      "logits/chosen": -0.39471542835235596,
+      "logits/rejected": -0.6508359909057617,
+      "logps/chosen": -6.633285999298096,
+      "logps/rejected": -58.268653869628906,
+      "loss": 0.5636770129203796,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.41472452878952026,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.1396198272705078,
+      "rewards/margins": 3.7723348140716553,
+      "rewards/rejected": -3.6327152252197266,
+      "step": 6,
+      "train_speed(iter/s)": 0.010667
+    },
+    {
+      "epoch": 0.04036036036036036,
+      "grad_norm": 4.302145481109619,
+      "learning_rate": 5.384615384615385e-05,
+      "logits/chosen": -0.42404329776763916,
+      "logits/rejected": -0.7084805369377136,
+      "logps/chosen": -4.09792947769165,
+      "logps/rejected": -69.69088745117188,
+      "loss": 0.5202179551124573,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3875095546245575,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.018997177481651306,
+      "rewards/margins": 5.0437703132629395,
+      "rewards/rejected": -5.062767505645752,
+      "step": 7,
+      "train_speed(iter/s)": 0.010685
+    },
+    {
+      "epoch": 0.04612612612612613,
+      "grad_norm": 4.475785255432129,
+      "learning_rate": 6.153846153846155e-05,
+      "logits/chosen": -0.4411894977092743,
+      "logits/rejected": -0.6594015955924988,
+      "logps/chosen": -6.067828178405762,
+      "logps/rejected": -52.161441802978516,
+      "loss": 0.8079419136047363,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.5527109503746033,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.13843362033367157,
+      "rewards/margins": 3.638244390487671,
+      "rewards/rejected": -3.4998104572296143,
+      "step": 8,
+      "train_speed(iter/s)": 0.010699
+    },
+    {
+      "epoch": 0.05189189189189189,
+      "grad_norm": 7.412286281585693,
+      "learning_rate": 6.923076923076924e-05,
+      "logits/chosen": -0.4308490753173828,
+      "logits/rejected": -0.6868806481361389,
+      "logps/chosen": -4.5601372718811035,
+      "logps/rejected": -75.88018798828125,
+      "loss": 0.3704576790332794,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.23506566882133484,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.1938612163066864,
+      "rewards/margins": 5.914104461669922,
+      "rewards/rejected": -5.720243453979492,
+      "step": 9,
+      "train_speed(iter/s)": 0.010708
+    },
+    {
+      "epoch": 0.05765765765765766,
+      "grad_norm": 2.0841763019561768,
+      "learning_rate": 7.692307692307693e-05,
+      "logits/chosen": -0.4546598792076111,
+      "logits/rejected": -0.7517622113227844,
+      "logps/chosen": -1.471292495727539,
+      "logps/rejected": -70.88200378417969,
+      "loss": 0.2737431526184082,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.14067727327346802,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.13278257846832275,
+      "rewards/margins": 5.628582000732422,
+      "rewards/rejected": -5.495799541473389,
+      "step": 10,
+      "train_speed(iter/s)": 0.010721
+    },
+    {
+      "epoch": 0.06342342342342343,
+      "grad_norm": 5.912591457366943,
+      "learning_rate": 8.461538461538461e-05,
+      "logits/chosen": -0.4128813147544861,
+      "logits/rejected": -0.5927000045776367,
+      "logps/chosen": -7.319942474365234,
+      "logps/rejected": -61.27806091308594,
+      "loss": 0.5680095553398132,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3753523826599121,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.07616602629423141,
+      "rewards/margins": 4.830938816070557,
+      "rewards/rejected": -4.754772663116455,
+      "step": 11,
+      "train_speed(iter/s)": 0.010727
+    },
+    {
+      "epoch": 0.06918918918918919,
+      "grad_norm": 2.5374209880828857,
+      "learning_rate": 9.230769230769232e-05,
+      "logits/chosen": -0.349031925201416,
+      "logits/rejected": -0.6373917460441589,
+      "logps/chosen": -4.199922561645508,
+      "logps/rejected": -72.83607482910156,
+      "loss": 0.34027042984962463,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.22808796167373657,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.20608201622962952,
+      "rewards/margins": 5.424815654754639,
+      "rewards/rejected": -5.218733787536621,
+      "step": 12,
+      "train_speed(iter/s)": 0.010728
+    },
+    {
+      "epoch": 0.07495495495495495,
+      "grad_norm": 2.7130517959594727,
+      "learning_rate": 0.0001,
+      "logits/chosen": -0.3631379306316376,
+      "logits/rejected": -0.6601940989494324,
+      "logps/chosen": -3.3938565254211426,
+      "logps/rejected": -65.2313232421875,
+      "loss": 0.4514213800430298,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.30333614349365234,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.06806281208992004,
+      "rewards/margins": 4.726724624633789,
+      "rewards/rejected": -4.658661365509033,
+      "step": 13,
+      "train_speed(iter/s)": 0.010733
+    },
+    {
+      "epoch": 0.08072072072072072,
+      "grad_norm": 1.5474798679351807,
+      "learning_rate": 0.0001076923076923077,
+      "logits/chosen": -0.4096493124961853,
+      "logits/rejected": -0.6515597105026245,
+      "logps/chosen": -2.300206184387207,
+      "logps/rejected": -57.26811218261719,
+      "loss": 0.3060320019721985,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.1727571338415146,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.08898697793483734,
+      "rewards/margins": 4.1406354904174805,
+      "rewards/rejected": -4.0516486167907715,
+      "step": 14,
+      "train_speed(iter/s)": 0.010737
+    },
+    {
+      "epoch": 0.08648648648648649,
+      "grad_norm": 1.628075361251831,
+      "learning_rate": 0.00011538461538461538,
+      "logits/chosen": -0.4402551054954529,
+      "logits/rejected": -0.6672070622444153,
+      "logps/chosen": -2.9995224475860596,
+      "logps/rejected": -53.87057113647461,
+      "loss": 0.3575773239135742,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2217392474412918,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.20550209283828735,
+      "rewards/margins": 3.6874935626983643,
+      "rewards/rejected": -3.4819912910461426,
+      "step": 15,
+      "train_speed(iter/s)": 0.010742
+    },
+    {
+      "epoch": 0.09225225225225225,
+      "grad_norm": 1.838375449180603,
+      "learning_rate": 0.0001230769230769231,
+      "logits/chosen": -0.3547935485839844,
+      "logits/rejected": -0.6626885533332825,
+      "logps/chosen": -4.558413982391357,
+      "logps/rejected": -64.36922454833984,
+      "loss": 0.45015937089920044,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3128659129142761,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.14241908490657806,
+      "rewards/margins": 4.83186149597168,
+      "rewards/rejected": -4.689442157745361,
+      "step": 16,
+      "train_speed(iter/s)": 0.010745
+    },
+    {
+      "epoch": 0.09801801801801802,
+      "grad_norm": 3.119417667388916,
+      "learning_rate": 0.00013076923076923077,
+      "logits/chosen": -0.3334687352180481,
+      "logits/rejected": -0.6293997168540955,
+      "logps/chosen": -5.949204444885254,
+      "logps/rejected": -74.25395965576172,
+      "loss": 0.6850419044494629,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.5004001259803772,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.07516612857580185,
+      "rewards/margins": 4.788809776306152,
+      "rewards/rejected": -4.713644027709961,
+      "step": 17,
+      "train_speed(iter/s)": 0.010741
+    },
+    {
+      "epoch": 0.10378378378378378,
+      "grad_norm": 1.6356091499328613,
+      "learning_rate": 0.00013846153846153847,
+      "logits/chosen": -0.3526935875415802,
+      "logits/rejected": -0.6357190608978271,
+      "logps/chosen": -2.911073923110962,
+      "logps/rejected": -66.46131896972656,
+      "loss": 0.32070592045783997,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.18397927284240723,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.09694083034992218,
+      "rewards/margins": 5.153117656707764,
+      "rewards/rejected": -5.056177139282227,
+      "step": 18,
+      "train_speed(iter/s)": 0.010741
+    },
+    {
+      "epoch": 0.10954954954954955,
+      "grad_norm": 2.7403504848480225,
+      "learning_rate": 0.00014615384615384615,
+      "logits/chosen": -0.3763662278652191,
+      "logits/rejected": -0.586420476436615,
+      "logps/chosen": -4.417596340179443,
+      "logps/rejected": -58.99639892578125,
+      "loss": 0.4634058177471161,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.28715017437934875,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.1435782015323639,
+      "rewards/margins": 4.609218120574951,
+      "rewards/rejected": -4.465640068054199,
+      "step": 19,
+      "train_speed(iter/s)": 0.010743
+    },
+    {
+      "epoch": 0.11531531531531532,
+      "grad_norm": 1.982851266860962,
+      "learning_rate": 0.00015384615384615385,
+      "logits/chosen": -0.3701193332672119,
+      "logits/rejected": -0.6088087558746338,
+      "logps/chosen": -4.619842529296875,
+      "logps/rejected": -68.1020278930664,
+      "loss": 0.4344192445278168,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.26591166853904724,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.10610304027795792,
+      "rewards/margins": 5.044137954711914,
+      "rewards/rejected": -4.938034534454346,
+      "step": 20,
+      "train_speed(iter/s)": 0.010745
+    },
+    {
+      "epoch": 0.12108108108108108,
+      "grad_norm": 3.175961494445801,
+      "learning_rate": 0.00016153846153846155,
+      "logits/chosen": -0.40009674429893494,
+      "logits/rejected": -0.642030656337738,
+      "logps/chosen": -3.357761859893799,
+      "logps/rejected": -65.21231842041016,
+      "loss": 0.40142354369163513,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2889588177204132,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.1323939859867096,
+      "rewards/margins": 5.234187602996826,
+      "rewards/rejected": -5.1017937660217285,
+      "step": 21,
+      "train_speed(iter/s)": 0.010748
+    },
+    {
+      "epoch": 0.12684684684684686,
+      "grad_norm": 1.9400721788406372,
+      "learning_rate": 0.00016923076923076923,
+      "logits/chosen": -0.3025435209274292,
+      "logits/rejected": -0.6950021386146545,
+      "logps/chosen": -2.492130994796753,
+      "logps/rejected": -107.03312683105469,
+      "loss": 0.2920421361923218,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.19772115349769592,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.19073158502578735,
+      "rewards/margins": 8.599780082702637,
+      "rewards/rejected": -8.409048080444336,
+      "step": 22,
+      "train_speed(iter/s)": 0.010746
+    },
+    {
+      "epoch": 0.13261261261261262,
+      "grad_norm": 1.3099267482757568,
+      "learning_rate": 0.00017692307692307693,
+      "logits/chosen": -0.36498135328292847,
+      "logits/rejected": -0.6359634399414062,
+      "logps/chosen": -2.421377658843994,
+      "logps/rejected": -74.05368041992188,
+      "loss": 0.20869436860084534,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.11023169010877609,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.18500684201717377,
+      "rewards/margins": 5.958514213562012,
+      "rewards/rejected": -5.773507118225098,
+      "step": 23,
+      "train_speed(iter/s)": 0.010747
+    },
+    {
+      "epoch": 0.13837837837837838,
+      "grad_norm": 1.9334064722061157,
+      "learning_rate": 0.00018461538461538463,
+      "logits/chosen": -0.45037421584129333,
+      "logits/rejected": -0.7719848155975342,
+      "logps/chosen": -2.6665430068969727,
+      "logps/rejected": -94.5350570678711,
+      "loss": 0.3057625889778137,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2374470978975296,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.09573014080524445,
+      "rewards/margins": 8.040020942687988,
+      "rewards/rejected": -7.944291114807129,
+      "step": 24,
+      "train_speed(iter/s)": 0.010749
+    },
+    {
+      "epoch": 0.14414414414414414,
+      "grad_norm": 7.789314270019531,
+      "learning_rate": 0.00019230769230769233,
+      "logits/chosen": -0.3473276197910309,
+      "logits/rejected": -0.6793957352638245,
+      "logps/chosen": -4.235269069671631,
+      "logps/rejected": -109.01844024658203,
+      "loss": 0.42337244749069214,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.21255403757095337,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.009569963440299034,
+      "rewards/margins": 8.769816398620605,
+      "rewards/rejected": -8.760246276855469,
+      "step": 25,
+      "train_speed(iter/s)": 0.010747
+    },
+    {
+      "epoch": 0.1499099099099099,
+      "grad_norm": 1.970948338508606,
+      "learning_rate": 0.0002,
+      "logits/chosen": -0.40258127450942993,
+      "logits/rejected": -0.7479668259620667,
+      "logps/chosen": -2.0013341903686523,
+      "logps/rejected": -85.51154327392578,
+      "loss": 0.4197818636894226,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2574591636657715,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.16396881639957428,
+      "rewards/margins": 7.033238410949707,
+      "rewards/rejected": -6.869269847869873,
+      "step": 26,
+      "train_speed(iter/s)": 0.010746
+    },
+    {
+      "epoch": 0.15567567567567567,
+      "grad_norm": 11.3534574508667,
+      "learning_rate": 0.00019999796963348897,
+      "logits/chosen": -0.5246711373329163,
+      "logits/rejected": -0.7294141054153442,
+      "logps/chosen": -11.85606575012207,
+      "logps/rejected": -76.03211975097656,
+      "loss": 0.8875234127044678,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.575400710105896,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -0.2269383817911148,
+      "rewards/margins": 5.498793125152588,
+      "rewards/rejected": -5.725731372833252,
+      "step": 27,
+      "train_speed(iter/s)": 0.010747
+    },
+    {
+      "epoch": 0.16144144144144143,
+      "grad_norm": 2.0314371585845947,
+      "learning_rate": 0.00019999187861640362,
+      "logits/chosen": -0.5284671187400818,
+      "logits/rejected": -0.7846834063529968,
+      "logps/chosen": -4.434022426605225,
+      "logps/rejected": -87.89623260498047,
+      "loss": 0.3868662416934967,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.30320096015930176,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.07812386751174927,
+      "rewards/margins": 7.374299049377441,
+      "rewards/rejected": -7.296175479888916,
+      "step": 28,
+      "train_speed(iter/s)": 0.010748
+    },
+    {
+      "epoch": 0.16720720720720722,
+      "grad_norm": 6.089433193206787,
+      "learning_rate": 0.0001999817271960839,
+      "logits/chosen": -0.4792092442512512,
+      "logits/rejected": -0.7371792793273926,
+      "logps/chosen": -3.6791412830352783,
+      "logps/rejected": -80.48262023925781,
+      "loss": 0.34335166215896606,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.22308313846588135,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -0.049518827348947525,
+      "rewards/margins": 6.395692825317383,
+      "rewards/rejected": -6.445211410522461,
+      "step": 29,
+      "train_speed(iter/s)": 0.010751
+    },
+    {
+      "epoch": 0.17297297297297298,
+      "grad_norm": 1.8896535634994507,
+      "learning_rate": 0.00019996751578475186,
+      "logits/chosen": -0.44468259811401367,
+      "logits/rejected": -0.7023504972457886,
+      "logps/chosen": -3.048004388809204,
+      "logps/rejected": -80.6348648071289,
+      "loss": 0.30949145555496216,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.1800393909215927,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.15738406777381897,
+      "rewards/margins": 5.8682780265808105,
+      "rewards/rejected": -5.710893630981445,
+      "step": 30,
+      "train_speed(iter/s)": 0.010749
+    },
+    {
+      "epoch": 0.17873873873873874,
+      "grad_norm": 2.7516067028045654,
+      "learning_rate": 0.00019994924495949504,
+      "logits/chosen": -0.4754641652107239,
+      "logits/rejected": -0.7408227920532227,
+      "logps/chosen": -5.943354606628418,
+      "logps/rejected": -79.13115692138672,
+      "loss": 0.6435320973396301,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.4586246609687805,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.0629303902387619,
+      "rewards/margins": 6.110820293426514,
+      "rewards/rejected": -6.0478901863098145,
+      "step": 31,
+      "train_speed(iter/s)": 0.01075
+    },
+    {
+      "epoch": 0.1845045045045045,
+      "grad_norm": 5.854465007781982,
+      "learning_rate": 0.00019992691546224278,
+      "logits/chosen": -0.45127877593040466,
+      "logits/rejected": -0.7568551898002625,
+      "logps/chosen": -2.736678123474121,
+      "logps/rejected": -66.9885025024414,
+      "loss": 0.3652667701244354,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.16899552941322327,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.050895966589450836,
+      "rewards/margins": 4.933393478393555,
+      "rewards/rejected": -4.882497310638428,
+      "step": 32,
+      "train_speed(iter/s)": 0.01075
+    },
+    {
+      "epoch": 0.19027027027027027,
+      "grad_norm": 1.3124663829803467,
+      "learning_rate": 0.00019990052819973642,
+      "logits/chosen": -0.3465266227722168,
+      "logits/rejected": -0.6685288548469543,
+      "logps/chosen": -3.301164388656616,
+      "logps/rejected": -85.9378662109375,
+      "loss": 0.3046538233757019,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.1670866757631302,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.17874348163604736,
+      "rewards/margins": 6.709518909454346,
+      "rewards/rejected": -6.530775547027588,
+      "step": 33,
+      "train_speed(iter/s)": 0.010748
+    },
+    {
+      "epoch": 0.19603603603603603,
+      "grad_norm": 2.341381549835205,
+      "learning_rate": 0.00019987008424349225,
+      "logits/chosen": -0.38037875294685364,
+      "logits/rejected": -0.699614405632019,
+      "logps/chosen": -1.5970546007156372,
+      "logps/rejected": -74.28025817871094,
+      "loss": 0.20820346474647522,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.10679290443658829,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.11616384983062744,
+      "rewards/margins": 5.825808048248291,
+      "rewards/rejected": -5.709644794464111,
+      "step": 34,
+      "train_speed(iter/s)": 0.010749
+    },
+    {
+      "epoch": 0.2018018018018018,
+      "grad_norm": 5.144662380218506,
+      "learning_rate": 0.00019983558482975799,
+      "logits/chosen": -0.46614986658096313,
+      "logits/rejected": -0.6599275469779968,
+      "logps/chosen": -5.982501983642578,
+      "logps/rejected": -67.7139892578125,
+      "loss": 0.5879941582679749,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.4303022027015686,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.16055577993392944,
+      "rewards/margins": 5.052016258239746,
+      "rewards/rejected": -4.89146089553833,
+      "step": 35,
+      "train_speed(iter/s)": 0.010749
+    },
+    {
+      "epoch": 0.20756756756756756,
+      "grad_norm": 1.990370273590088,
+      "learning_rate": 0.00019979703135946278,
+      "logits/chosen": -0.4655105471611023,
+      "logits/rejected": -0.6668959856033325,
+      "logps/chosen": -4.062808036804199,
+      "logps/rejected": -72.87210845947266,
+      "loss": 0.4453062415122986,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.25426411628723145,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.09591292589902878,
+      "rewards/margins": 5.528655052185059,
+      "rewards/rejected": -5.432741641998291,
+      "step": 36,
+      "train_speed(iter/s)": 0.010749
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 2.0748369693756104,
+      "learning_rate": 0.00019975442539816012,
+      "logits/chosen": -0.4139191210269928,
+      "logits/rejected": -0.7536109685897827,
+      "logps/chosen": -5.956673622131348,
+      "logps/rejected": -66.99488830566406,
+      "loss": 0.5514004826545715,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3969358503818512,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.33245787024497986,
+      "rewards/margins": 5.0754570960998535,
+      "rewards/rejected": -4.742999076843262,
+      "step": 37,
+      "train_speed(iter/s)": 0.01075
+    },
+    {
+      "epoch": 0.2190990990990991,
+      "grad_norm": 1.6668081283569336,
+      "learning_rate": 0.0001997077686759643,
+      "logits/chosen": -0.42306363582611084,
+      "logits/rejected": -0.6915761232376099,
+      "logps/chosen": -3.814265727996826,
+      "logps/rejected": -82.30941772460938,
+      "loss": 0.3710038363933563,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.22380009293556213,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.06087161600589752,
+      "rewards/margins": 6.076653957366943,
+      "rewards/rejected": -6.015782356262207,
+      "step": 38,
+      "train_speed(iter/s)": 0.01075
+    },
+    {
+      "epoch": 0.22486486486486487,
+      "grad_norm": 13.827637672424316,
+      "learning_rate": 0.00019965706308748028,
+      "logits/chosen": -0.49769100546836853,
+      "logits/rejected": -0.7370949983596802,
+      "logps/chosen": -3.956310272216797,
+      "logps/rejected": -79.41323852539062,
+      "loss": 0.6633445620536804,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3870386779308319,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -0.13494591414928436,
+      "rewards/margins": 6.147372722625732,
+      "rewards/rejected": -6.282319068908691,
+      "step": 39,
+      "train_speed(iter/s)": 0.010751
+    },
+    {
+      "epoch": 0.23063063063063063,
+      "grad_norm": 2.19854474067688,
+      "learning_rate": 0.0001996023106917267,
+      "logits/chosen": -0.44343504309654236,
+      "logits/rejected": -0.6730125546455383,
+      "logps/chosen": -2.9238576889038086,
+      "logps/rejected": -77.40494537353516,
+      "loss": 0.3453696668148041,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.19069285690784454,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.14595907926559448,
+      "rewards/margins": 6.002254486083984,
+      "rewards/rejected": -5.856295108795166,
+      "step": 40,
+      "train_speed(iter/s)": 0.010752
+    },
+    {
+      "epoch": 0.2363963963963964,
+      "grad_norm": 6.0794596672058105,
+      "learning_rate": 0.00019954351371205203,
+      "logits/chosen": -0.47579658031463623,
+      "logits/rejected": -0.6826990842819214,
+      "logps/chosen": -5.335310935974121,
+      "logps/rejected": -75.78384399414062,
+      "loss": 0.575509250164032,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.4477939307689667,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.2103371024131775,
+      "rewards/margins": 6.012211322784424,
+      "rewards/rejected": -5.801875114440918,
+      "step": 41,
+      "train_speed(iter/s)": 0.010754
+    },
+    {
+      "epoch": 0.24216216216216216,
+      "grad_norm": 1.4152264595031738,
+      "learning_rate": 0.00019948067453604473,
+      "logits/chosen": -0.49620914459228516,
+      "logits/rejected": -0.7446714043617249,
+      "logps/chosen": -3.284031867980957,
+      "logps/rejected": -77.9183349609375,
+      "loss": 0.29828324913978577,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.21160344779491425,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.18678154051303864,
+      "rewards/margins": 6.247434616088867,
+      "rewards/rejected": -6.060652732849121,
+      "step": 42,
+      "train_speed(iter/s)": 0.010755
+    },
+    {
+      "epoch": 0.24792792792792792,
+      "grad_norm": 2.9591963291168213,
+      "learning_rate": 0.00019941379571543596,
+      "logits/chosen": -0.4777877628803253,
+      "logits/rejected": -0.6974171996116638,
+      "logps/chosen": -2.9466118812561035,
+      "logps/rejected": -66.2719497680664,
+      "loss": 0.35521066188812256,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.22793087363243103,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.27246373891830444,
+      "rewards/margins": 5.403186798095703,
+      "rewards/rejected": -5.130722999572754,
+      "step": 43,
+      "train_speed(iter/s)": 0.010756
+    },
+    {
+      "epoch": 0.2536936936936937,
+      "grad_norm": 4.199843883514404,
+      "learning_rate": 0.00019934287996599607,
+      "logits/chosen": -0.4372207522392273,
+      "logits/rejected": -0.700534462928772,
+      "logps/chosen": -3.31451416015625,
+      "logps/rejected": -70.0909194946289,
+      "loss": 0.3793654143810272,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.20433279871940613,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.059201404452323914,
+      "rewards/margins": 4.912087917327881,
+      "rewards/rejected": -4.852885723114014,
+      "step": 44,
+      "train_speed(iter/s)": 0.010756
+    },
+    {
+      "epoch": 0.2594594594594595,
+      "grad_norm": 0.7397028803825378,
+      "learning_rate": 0.00019926793016742435,
+      "logits/chosen": -0.42062634229660034,
+      "logits/rejected": -0.7593384981155396,
+      "logps/chosen": -1.404799222946167,
+      "logps/rejected": -102.4892349243164,
+      "loss": 0.1313902586698532,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.09521690011024475,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.12229674309492111,
+      "rewards/margins": 8.041703224182129,
+      "rewards/rejected": -7.919405937194824,
+      "step": 45,
+      "train_speed(iter/s)": 0.010754
+    },
+    {
+      "epoch": 0.26522522522522524,
+      "grad_norm": 5.14201545715332,
+      "learning_rate": 0.00019918894936323195,
+      "logits/chosen": -0.3798273205757141,
+      "logits/rejected": -0.5865642428398132,
+      "logps/chosen": -6.911247730255127,
+      "logps/rejected": -72.06941223144531,
+      "loss": 0.49299517273902893,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.27924880385398865,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.1307646483182907,
+      "rewards/margins": 5.647462368011475,
+      "rewards/rejected": -5.516696929931641,
+      "step": 46,
+      "train_speed(iter/s)": 0.010753
+    },
+    {
+      "epoch": 0.270990990990991,
+      "grad_norm": 1.6439752578735352,
+      "learning_rate": 0.00019910594076061853,
+      "logits/chosen": -0.4431299567222595,
+      "logits/rejected": -0.7268660664558411,
+      "logps/chosen": -2.0275888442993164,
+      "logps/rejected": -81.84529876708984,
+      "loss": 0.23740296065807343,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.14832313358783722,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.06923090666532516,
+      "rewards/margins": 6.1783928871154785,
+      "rewards/rejected": -6.109161376953125,
+      "step": 47,
+      "train_speed(iter/s)": 0.010753
+    },
+    {
+      "epoch": 0.27675675675675676,
+      "grad_norm": 6.499570846557617,
+      "learning_rate": 0.0001990189077303418,
+      "logits/chosen": -0.45142635703086853,
+      "logits/rejected": -0.7249264121055603,
+      "logps/chosen": -5.794652938842773,
+      "logps/rejected": -78.53363800048828,
+      "loss": 0.46158841252326965,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3398270010948181,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.24555836617946625,
+      "rewards/margins": 6.308343887329102,
+      "rewards/rejected": -6.062786102294922,
+      "step": 48,
+      "train_speed(iter/s)": 0.010753
+    },
+    {
+      "epoch": 0.2825225225225225,
+      "grad_norm": 5.58511209487915,
+      "learning_rate": 0.00019892785380658078,
+      "logits/chosen": -0.4099079370498657,
+      "logits/rejected": -0.6898340582847595,
+      "logps/chosen": -7.449429035186768,
+      "logps/rejected": -92.18098449707031,
+      "loss": 0.5815075039863586,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.41358357667922974,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.053265977650880814,
+      "rewards/margins": 7.150655269622803,
+      "rewards/rejected": -7.203920841217041,
+      "step": 49,
+      "train_speed(iter/s)": 0.010752
+    },
+    {
+      "epoch": 0.2882882882882883,
+      "grad_norm": 1.5411535501480103,
+      "learning_rate": 0.00019883278268679216,
+      "logits/chosen": -0.4670988917350769,
+      "logits/rejected": -0.6586328148841858,
+      "logps/chosen": -6.984646320343018,
+      "logps/rejected": -89.9208755493164,
+      "loss": 0.45735102891921997,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.28862154483795166,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.1682412177324295,
+      "rewards/margins": 7.457679748535156,
+      "rewards/rejected": -7.289438247680664,
+      "step": 50,
+      "train_speed(iter/s)": 0.01075
+    },
+    {
+      "epoch": 0.29405405405405405,
+      "grad_norm": 6.264907360076904,
+      "learning_rate": 0.00019873369823156036,
+      "logits/chosen": -0.35443419218063354,
+      "logits/rejected": -0.7265148758888245,
+      "logps/chosen": -5.11212158203125,
+      "logps/rejected": -97.37500762939453,
+      "loss": 0.648152232170105,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.5091428756713867,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.03551238775253296,
+      "rewards/margins": 7.55380392074585,
+      "rewards/rejected": -7.589317321777344,
+      "step": 51,
+      "train_speed(iter/s)": 0.010707
+    },
+    {
+      "epoch": 0.2998198198198198,
+      "grad_norm": 2.1848013401031494,
+      "learning_rate": 0.00019863060446444054,
+      "logits/chosen": -0.5179776549339294,
+      "logits/rejected": -0.7255269289016724,
+      "logps/chosen": -6.782817363739014,
+      "logps/rejected": -86.8519515991211,
+      "loss": 0.5702924728393555,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3926149308681488,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.05346503108739853,
+      "rewards/margins": 7.021766185760498,
+      "rewards/rejected": -6.968300819396973,
+      "step": 52,
+      "train_speed(iter/s)": 0.010708
+    },
+    {
+      "epoch": 0.3055855855855856,
+      "grad_norm": 3.3478569984436035,
+      "learning_rate": 0.00019852350557179538,
+      "logits/chosen": -0.5434097051620483,
+      "logits/rejected": -0.8009167313575745,
+      "logps/chosen": -4.053906440734863,
+      "logps/rejected": -105.9520263671875,
+      "loss": 0.4421282112598419,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3492012917995453,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.1857108175754547,
+      "rewards/margins": 9.035101890563965,
+      "rewards/rejected": -8.849390983581543,
+      "step": 53,
+      "train_speed(iter/s)": 0.010709
+    },
+    {
+      "epoch": 0.31135135135135134,
+      "grad_norm": 0.8761724829673767,
+      "learning_rate": 0.00019841240590262493,
+      "logits/chosen": -0.45293307304382324,
+      "logits/rejected": -0.7608242034912109,
+      "logps/chosen": -0.6350191831588745,
+      "logps/rejected": -91.82192993164062,
+      "loss": 0.11042662709951401,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.06284812837839127,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.1965743899345398,
+      "rewards/margins": 7.520632743835449,
+      "rewards/rejected": -7.324058532714844,
+      "step": 54,
+      "train_speed(iter/s)": 0.01071
+    },
+    {
+      "epoch": 0.3171171171171171,
+      "grad_norm": 5.344722747802734,
+      "learning_rate": 0.0001982973099683902,
+      "logits/chosen": -0.5431379675865173,
+      "logits/rejected": -0.8100754022598267,
+      "logps/chosen": -1.98147714138031,
+      "logps/rejected": -91.00532531738281,
+      "loss": 0.3004758059978485,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.1778518110513687,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.1774783581495285,
+      "rewards/margins": 7.744819641113281,
+      "rewards/rejected": -7.567340850830078,
+      "step": 55,
+      "train_speed(iter/s)": 0.010711
+    },
+    {
+      "epoch": 0.32288288288288286,
+      "grad_norm": 5.424385070800781,
+      "learning_rate": 0.00019817822244282973,
+      "logits/chosen": -0.48769453167915344,
+      "logits/rejected": -0.7255769371986389,
+      "logps/chosen": -3.6506757736206055,
+      "logps/rejected": -84.81509399414062,
+      "loss": 0.4457959830760956,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.23988068103790283,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.11002250760793686,
+      "rewards/margins": 6.652472019195557,
+      "rewards/rejected": -6.542449951171875,
+      "step": 56,
+      "train_speed(iter/s)": 0.010711
+    },
+    {
+      "epoch": 0.3286486486486486,
+      "grad_norm": 2.3812549114227295,
+      "learning_rate": 0.00019805514816177006,
+      "logits/chosen": -0.556337833404541,
+      "logits/rejected": -0.7923259735107422,
+      "logps/chosen": -5.455251693725586,
+      "logps/rejected": -89.06196594238281,
+      "loss": 0.4395422339439392,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.29659417271614075,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.01958906650543213,
+      "rewards/margins": 6.816771984100342,
+      "rewards/rejected": -6.836360931396484,
+      "step": 57,
+      "train_speed(iter/s)": 0.01071
+    },
+    {
+      "epoch": 0.33441441441441444,
+      "grad_norm": 8.72442626953125,
+      "learning_rate": 0.00019792809212292912,
+      "logits/chosen": -0.6302639842033386,
+      "logits/rejected": -0.9231972098350525,
+      "logps/chosen": -5.753312110900879,
+      "logps/rejected": -103.71651458740234,
+      "loss": 1.1447618007659912,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.6733683943748474,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -0.26626041531562805,
+      "rewards/margins": 8.245807647705078,
+      "rewards/rejected": -8.512067794799805,
+      "step": 58,
+      "train_speed(iter/s)": 0.010711
+    },
+    {
+      "epoch": 0.3401801801801802,
+      "grad_norm": 2.6023576259613037,
+      "learning_rate": 0.00019779705948571346,
+      "logits/chosen": -0.5867213010787964,
+      "logits/rejected": -0.8538931608200073,
+      "logps/chosen": -3.069394111633301,
+      "logps/rejected": -96.5278091430664,
+      "loss": 0.3167837858200073,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.20217999815940857,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.10776210576295853,
+      "rewards/margins": 8.32030963897705,
+      "rewards/rejected": -8.21254825592041,
+      "step": 59,
+      "train_speed(iter/s)": 0.010712
+    },
+    {
+      "epoch": 0.34594594594594597,
+      "grad_norm": 2.678529977798462,
+      "learning_rate": 0.00019766205557100868,
+      "logits/chosen": -0.6138476133346558,
+      "logits/rejected": -0.7842435836791992,
+      "logps/chosen": -6.119746208190918,
+      "logps/rejected": -88.4565658569336,
+      "loss": 0.44386592507362366,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2780015468597412,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.06730165332555771,
+      "rewards/margins": 7.516484260559082,
+      "rewards/rejected": -7.449182987213135,
+      "step": 60,
+      "train_speed(iter/s)": 0.010713
+    },
+    {
+      "epoch": 0.3517117117117117,
+      "grad_norm": 2.223165512084961,
+      "learning_rate": 0.00019752308586096326,
+      "logits/chosen": -0.5360291004180908,
+      "logits/rejected": -0.894256591796875,
+      "logps/chosen": -1.3611795902252197,
+      "logps/rejected": -122.1118392944336,
+      "loss": 0.13098187744617462,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.07775433361530304,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.12457617372274399,
+      "rewards/margins": 10.008496284484863,
+      "rewards/rejected": -9.883918762207031,
+      "step": 61,
+      "train_speed(iter/s)": 0.010713
+    },
+    {
+      "epoch": 0.3574774774774775,
+      "grad_norm": 10.891358375549316,
+      "learning_rate": 0.0001973801559987661,
+      "logits/chosen": -0.7173175811767578,
+      "logits/rejected": -0.8792683482170105,
+      "logps/chosen": -12.6211519241333,
+      "logps/rejected": -73.51658630371094,
+      "loss": 1.0753252506256104,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.9100903868675232,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.495037317276001,
+      "rewards/margins": 5.504097938537598,
+      "rewards/rejected": -5.9991350173950195,
+      "step": 62,
+      "train_speed(iter/s)": 0.010715
+    },
+    {
+      "epoch": 0.36324324324324325,
+      "grad_norm": 26.432544708251953,
+      "learning_rate": 0.00019723327178841743,
+      "logits/chosen": -0.6629639267921448,
+      "logits/rejected": -0.8958379030227661,
+      "logps/chosen": -5.678650856018066,
+      "logps/rejected": -81.07159423828125,
+      "loss": 0.6158818602561951,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.4310333728790283,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.02597002126276493,
+      "rewards/margins": 6.523518085479736,
+      "rewards/rejected": -6.497547149658203,
+      "step": 63,
+      "train_speed(iter/s)": 0.010716
+    },
+    {
+      "epoch": 0.369009009009009,
+      "grad_norm": 4.643962860107422,
+      "learning_rate": 0.0001970824391944927,
+      "logits/chosen": -0.6061327457427979,
+      "logits/rejected": -0.8394734859466553,
+      "logps/chosen": -6.52767276763916,
+      "logps/rejected": -74.48380279541016,
+      "loss": 0.5795359015464783,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.414874792098999,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.0007130205631256104,
+      "rewards/margins": 5.6362080574035645,
+      "rewards/rejected": -5.635494709014893,
+      "step": 64,
+      "train_speed(iter/s)": 0.010716
+    },
+    {
+      "epoch": 0.3747747747747748,
+      "grad_norm": 1.8694281578063965,
+      "learning_rate": 0.00019692766434190105,
+      "logits/chosen": -0.6322288513183594,
+      "logits/rejected": -0.8566944003105164,
+      "logps/chosen": -1.9187999963760376,
+      "logps/rejected": -56.726871490478516,
+      "loss": 0.26561087369918823,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.12671466171741486,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.17268410325050354,
+      "rewards/margins": 4.417227268218994,
+      "rewards/rejected": -4.244543075561523,
+      "step": 65,
+      "train_speed(iter/s)": 0.010717
+    },
+    {
+      "epoch": 0.38054054054054054,
+      "grad_norm": 2.3087944984436035,
+      "learning_rate": 0.00019676895351563588,
+      "logits/chosen": -0.6048943996429443,
+      "logits/rejected": -0.8088940382003784,
+      "logps/chosen": -4.430207252502441,
+      "logps/rejected": -64.05681610107422,
+      "loss": 0.33754763007164,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.22767014801502228,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.025299668312072754,
+      "rewards/margins": 4.6934709548950195,
+      "rewards/rejected": -4.668170928955078,
+      "step": 66,
+      "train_speed(iter/s)": 0.010717
+    },
+    {
+      "epoch": 0.3863063063063063,
+      "grad_norm": 2.2111432552337646,
+      "learning_rate": 0.00019660631316052021,
+      "logits/chosen": -0.623546838760376,
+      "logits/rejected": -0.811290979385376,
+      "logps/chosen": -6.471018314361572,
+      "logps/rejected": -57.678314208984375,
+      "loss": 0.4503980875015259,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.34320953488349915,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -0.06837636232376099,
+      "rewards/margins": 4.10473108291626,
+      "rewards/rejected": -4.173107624053955,
+      "step": 67,
+      "train_speed(iter/s)": 0.010718
+    },
+    {
+      "epoch": 0.39207207207207206,
+      "grad_norm": 1.7001421451568604,
+      "learning_rate": 0.00019643974988094458,
+      "logits/chosen": -0.5895470380783081,
+      "logits/rejected": -0.7888103723526001,
+      "logps/chosen": -5.163002014160156,
+      "logps/rejected": -45.541908264160156,
+      "loss": 0.42964497208595276,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3103106617927551,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.3456801772117615,
+      "rewards/margins": 3.432896614074707,
+      "rewards/rejected": -3.08721661567688,
+      "step": 68,
+      "train_speed(iter/s)": 0.01072
+    },
+    {
+      "epoch": 0.3978378378378378,
+      "grad_norm": 3.909360885620117,
+      "learning_rate": 0.00019626927044059914,
+      "logits/chosen": -0.6134467124938965,
+      "logits/rejected": -0.7660009860992432,
+      "logps/chosen": -5.3520307540893555,
+      "logps/rejected": -49.259185791015625,
+      "loss": 0.5181165337562561,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.30421963334083557,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.014049924910068512,
+      "rewards/margins": 3.3693478107452393,
+      "rewards/rejected": -3.355297565460205,
+      "step": 69,
+      "train_speed(iter/s)": 0.01072
+    },
+    {
+      "epoch": 0.4036036036036036,
+      "grad_norm": 2.67299485206604,
+      "learning_rate": 0.00019609488176219886,
+      "logits/chosen": -0.6107767820358276,
+      "logits/rejected": -0.7868641018867493,
+      "logps/chosen": -3.616837501525879,
+      "logps/rejected": -50.11117172241211,
+      "loss": 0.3146559000015259,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.21267317235469818,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.13358786702156067,
+      "rewards/margins": 3.8761374950408936,
+      "rewards/rejected": -3.742549419403076,
+      "step": 70,
+      "train_speed(iter/s)": 0.010721
+    },
+    {
+      "epoch": 0.40936936936936935,
+      "grad_norm": 7.013823509216309,
+      "learning_rate": 0.00019591659092720227,
+      "logits/chosen": -0.5896518230438232,
+      "logits/rejected": -0.858326256275177,
+      "logps/chosen": -2.8418002128601074,
+      "logps/rejected": -64.546630859375,
+      "loss": 0.37414392828941345,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.25448697805404663,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.024522263556718826,
+      "rewards/margins": 4.37087345123291,
+      "rewards/rejected": -4.346351623535156,
+      "step": 71,
+      "train_speed(iter/s)": 0.010722
+    },
+    {
+      "epoch": 0.4151351351351351,
+      "grad_norm": 2.0888102054595947,
+      "learning_rate": 0.00019573440517552427,
+      "logits/chosen": -0.5669695138931274,
+      "logits/rejected": -0.7572087049484253,
+      "logps/chosen": -2.418893337249756,
+      "logps/rejected": -50.858699798583984,
+      "loss": 0.3724361062049866,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.20290419459342957,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.16552519798278809,
+      "rewards/margins": 3.5489120483398438,
+      "rewards/rejected": -3.3833868503570557,
+      "step": 72,
+      "train_speed(iter/s)": 0.010723
+    },
+    {
+      "epoch": 0.4209009009009009,
+      "grad_norm": 1.9631896018981934,
+      "learning_rate": 0.00019554833190524182,
+      "logits/chosen": -0.575507402420044,
+      "logits/rejected": -0.7697166204452515,
+      "logps/chosen": -3.5829074382781982,
+      "logps/rejected": -52.79640197753906,
+      "loss": 0.400629460811615,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2976799011230469,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.17720367014408112,
+      "rewards/margins": 4.019702434539795,
+      "rewards/rejected": -3.842499017715454,
+      "step": 73,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 3.227459192276001,
+      "learning_rate": 0.00019535837867229363,
+      "logits/chosen": -0.6866841316223145,
+      "logits/rejected": -0.8467439413070679,
+      "logps/chosen": -1.4898688793182373,
+      "logps/rejected": -52.35246276855469,
+      "loss": 0.2691449224948883,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.16896064579486847,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.20350010693073273,
+      "rewards/margins": 4.1118974685668945,
+      "rewards/rejected": -3.9083974361419678,
+      "step": 74,
+      "train_speed(iter/s)": 0.010726
+    },
+    {
+      "epoch": 0.43243243243243246,
+      "grad_norm": 2.0599851608276367,
+      "learning_rate": 0.0001951645531901734,
+      "logits/chosen": -0.6125411987304688,
+      "logits/rejected": -0.8178025484085083,
+      "logps/chosen": -5.753840923309326,
+      "logps/rejected": -60.490638732910156,
+      "loss": 0.4720686376094818,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2955226004123688,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": 0.02442253567278385,
+      "rewards/margins": 4.439631462097168,
+      "rewards/rejected": -4.41520881652832,
+      "step": 75,
+      "train_speed(iter/s)": 0.010726
+    },
+    {
+      "epoch": 0.4381981981981982,
+      "grad_norm": 1.8315515518188477,
+      "learning_rate": 0.00019496686332961646,
+      "logits/chosen": -0.5511488914489746,
+      "logits/rejected": -0.8103834390640259,
+      "logps/chosen": -2.737548828125,
+      "logps/rejected": -72.79937744140625,
+      "loss": 0.2598808705806732,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.20735348761081696,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.20416659116744995,
+      "rewards/margins": 5.639776706695557,
+      "rewards/rejected": -5.435609817504883,
+      "step": 76,
+      "train_speed(iter/s)": 0.010726
+    },
+    {
+      "epoch": 0.443963963963964,
+      "grad_norm": 4.658299922943115,
+      "learning_rate": 0.00019476531711828027,
+      "logits/chosen": -0.5621539354324341,
+      "logits/rejected": -0.817601203918457,
+      "logps/chosen": -2.554128646850586,
+      "logps/rejected": -75.40179443359375,
+      "loss": 0.29839468002319336,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.21096321940422058,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.11831088364124298,
+      "rewards/margins": 5.797797679901123,
+      "rewards/rejected": -5.679486274719238,
+      "step": 77,
+      "train_speed(iter/s)": 0.010727
+    },
+    {
+      "epoch": 0.44972972972972974,
+      "grad_norm": 7.5920844078063965,
+      "learning_rate": 0.00019455992274041835,
+      "logits/chosen": -0.5446800589561462,
+      "logits/rejected": -0.7866684794425964,
+      "logps/chosen": -5.180241584777832,
+      "logps/rejected": -80.02595520019531,
+      "loss": 0.4992501735687256,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.4071677029132843,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.006200561299920082,
+      "rewards/margins": 6.354435920715332,
+      "rewards/rejected": -6.348235607147217,
+      "step": 78,
+      "train_speed(iter/s)": 0.010728
+    },
+    {
+      "epoch": 0.4554954954954955,
+      "grad_norm": 2.045180082321167,
+      "learning_rate": 0.00019435068853654807,
+      "logits/chosen": -0.5892366170883179,
+      "logits/rejected": -0.782485842704773,
+      "logps/chosen": -7.082699775695801,
+      "logps/rejected": -76.8997573852539,
+      "loss": 0.4684464931488037,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.38624832034111023,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.03203499689698219,
+      "rewards/margins": 5.84527587890625,
+      "rewards/rejected": -5.813241004943848,
+      "step": 79,
+      "train_speed(iter/s)": 0.010728
+    },
+    {
+      "epoch": 0.46126126126126127,
+      "grad_norm": 6.412923812866211,
+      "learning_rate": 0.00019413762300311182,
+      "logits/chosen": -0.6524518728256226,
+      "logits/rejected": -0.8889943957328796,
+      "logps/chosen": -6.470341682434082,
+      "logps/rejected": -86.74302673339844,
+      "loss": 0.7839629650115967,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.6858316659927368,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.10852594673633575,
+      "rewards/margins": 6.819857120513916,
+      "rewards/rejected": -6.928382873535156,
+      "step": 80,
+      "train_speed(iter/s)": 0.010729
+    },
+    {
+      "epoch": 0.46702702702702703,
+      "grad_norm": 2.171276807785034,
+      "learning_rate": 0.00019392073479213213,
+      "logits/chosen": -0.5607702136039734,
+      "logits/rejected": -0.7560180425643921,
+      "logps/chosen": -6.120340347290039,
+      "logps/rejected": -81.5516128540039,
+      "loss": 0.575468897819519,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3079308271408081,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.05391833186149597,
+      "rewards/margins": 5.951426029205322,
+      "rewards/rejected": -6.005344390869141,
+      "step": 81,
+      "train_speed(iter/s)": 0.010729
+    },
+    {
+      "epoch": 0.4727927927927928,
+      "grad_norm": 1.5074808597564697,
+      "learning_rate": 0.0001937000327108601,
+      "logits/chosen": -0.5817635655403137,
+      "logits/rejected": -0.8122580051422119,
+      "logps/chosen": -3.3293583393096924,
+      "logps/rejected": -76.75574493408203,
+      "loss": 0.27197888493537903,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.1920165866613388,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.04472556710243225,
+      "rewards/margins": 6.159332275390625,
+      "rewards/rejected": -6.114606857299805,
+      "step": 82,
+      "train_speed(iter/s)": 0.01073
+    },
+    {
+      "epoch": 0.47855855855855856,
+      "grad_norm": 8.971928596496582,
+      "learning_rate": 0.0001934755257214181,
+      "logits/chosen": -0.5557482838630676,
+      "logits/rejected": -0.791240930557251,
+      "logps/chosen": -5.363508701324463,
+      "logps/rejected": -71.66944122314453,
+      "loss": 0.8509294986724854,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.7254281640052795,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.1657000482082367,
+      "rewards/margins": 5.284771919250488,
+      "rewards/rejected": -5.450472354888916,
+      "step": 83,
+      "train_speed(iter/s)": 0.010731
+    },
+    {
+      "epoch": 0.4843243243243243,
+      "grad_norm": 3.7317609786987305,
+      "learning_rate": 0.00019324722294043558,
+      "logits/chosen": -0.49631816148757935,
+      "logits/rejected": -0.7238998413085938,
+      "logps/chosen": -5.0319366455078125,
+      "logps/rejected": -55.3072395324707,
+      "loss": 0.4805355668067932,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.30268925428390503,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.16919195652008057,
+      "rewards/margins": 4.183167934417725,
+      "rewards/rejected": -4.013975620269775,
+      "step": 84,
+      "train_speed(iter/s)": 0.010732
+    },
+    {
+      "epoch": 0.4900900900900901,
+      "grad_norm": 7.251592636108398,
+      "learning_rate": 0.00019301513363867895,
+      "logits/chosen": -0.4797047972679138,
+      "logits/rejected": -0.699076771736145,
+      "logps/chosen": -4.068089485168457,
+      "logps/rejected": -62.654762268066406,
+      "loss": 0.41215279698371887,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3088461756706238,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.13345223665237427,
+      "rewards/margins": 4.483903884887695,
+      "rewards/rejected": -4.350452423095703,
+      "step": 85,
+      "train_speed(iter/s)": 0.010732
+    },
+    {
+      "epoch": 0.49585585585585584,
+      "grad_norm": 1.6465262174606323,
+      "learning_rate": 0.0001927792672406751,
+      "logits/chosen": -0.4875030219554901,
+      "logits/rejected": -0.7307705879211426,
+      "logps/chosen": -2.494859218597412,
+      "logps/rejected": -57.72221374511719,
+      "loss": 0.31443703174591064,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.1957024335861206,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.1970575600862503,
+      "rewards/margins": 4.039019584655762,
+      "rewards/rejected": -3.8419618606567383,
+      "step": 86,
+      "train_speed(iter/s)": 0.010732
+    },
+    {
+      "epoch": 0.5016216216216216,
+      "grad_norm": 2.362053394317627,
+      "learning_rate": 0.00019253963332432878,
+      "logits/chosen": -0.6053374409675598,
+      "logits/rejected": -0.7439757585525513,
+      "logps/chosen": -5.3451008796691895,
+      "logps/rejected": -47.488853454589844,
+      "loss": 0.4348059296607971,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.23809021711349487,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.019803959876298904,
+      "rewards/margins": 3.427215099334717,
+      "rewards/rejected": -3.447019338607788,
+      "step": 87,
+      "train_speed(iter/s)": 0.010733
+    },
+    {
+      "epoch": 0.5073873873873874,
+      "grad_norm": 2.1728515625,
+      "learning_rate": 0.00019229624162053356,
+      "logits/chosen": -0.5406431555747986,
+      "logits/rejected": -0.739841103553772,
+      "logps/chosen": -2.6260576248168945,
+      "logps/rejected": -56.7901496887207,
+      "loss": 0.4087463915348053,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2899452745914459,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.0745973214507103,
+      "rewards/margins": 4.15200662612915,
+      "rewards/rejected": -4.077408790588379,
+      "step": 88,
+      "train_speed(iter/s)": 0.010733
+    },
+    {
+      "epoch": 0.5131531531531531,
+      "grad_norm": 1.563063144683838,
+      "learning_rate": 0.00019204910201277672,
+      "logits/chosen": -0.5593241453170776,
+      "logits/rejected": -0.7506377100944519,
+      "logps/chosen": -3.1961171627044678,
+      "logps/rejected": -63.025291442871094,
+      "loss": 0.20274411141872406,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.15050008893013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.13150721788406372,
+      "rewards/margins": 4.959373950958252,
+      "rewards/rejected": -4.827866554260254,
+      "step": 89,
+      "train_speed(iter/s)": 0.010734
+    },
+    {
+      "epoch": 0.518918918918919,
+      "grad_norm": 1.8057938814163208,
+      "learning_rate": 0.0001917982245367379,
+      "logits/chosen": -0.5387026071548462,
+      "logits/rejected": -0.779965341091156,
+      "logps/chosen": -5.658284664154053,
+      "logps/rejected": -61.30659484863281,
+      "loss": 0.5316486358642578,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.4009394347667694,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.2519141137599945,
+      "rewards/margins": 4.251716613769531,
+      "rewards/rejected": -3.9998018741607666,
+      "step": 90,
+      "train_speed(iter/s)": 0.010734
+    },
+    {
+      "epoch": 0.5246846846846847,
+      "grad_norm": 1.6844978332519531,
+      "learning_rate": 0.00019154361937988163,
+      "logits/chosen": -0.5544890761375427,
+      "logits/rejected": -0.7064109444618225,
+      "logps/chosen": -6.5527496337890625,
+      "logps/rejected": -65.5311050415039,
+      "loss": 0.38739749789237976,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2913617193698883,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.1942709982395172,
+      "rewards/margins": 5.296841621398926,
+      "rewards/rejected": -5.102570533752441,
+      "step": 91,
+      "train_speed(iter/s)": 0.010734
+    },
+    {
+      "epoch": 0.5304504504504505,
+      "grad_norm": 3.110102653503418,
+      "learning_rate": 0.00019128529688104364,
+      "logits/chosen": -0.5477601885795593,
+      "logits/rejected": -0.7957394123077393,
+      "logps/chosen": -2.064303398132324,
+      "logps/rejected": -68.15472412109375,
+      "loss": 0.3176850974559784,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.21008610725402832,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.18838940560817719,
+      "rewards/margins": 5.544022560119629,
+      "rewards/rejected": -5.355632781982422,
+      "step": 92,
+      "train_speed(iter/s)": 0.010735
+    },
+    {
+      "epoch": 0.5362162162162162,
+      "grad_norm": 3.8029165267944336,
+      "learning_rate": 0.00019102326753001086,
+      "logits/chosen": -0.5357910394668579,
+      "logits/rejected": -0.7163031101226807,
+      "logps/chosen": -4.390456676483154,
+      "logps/rejected": -55.20818328857422,
+      "loss": 0.4618917405605316,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.33474794030189514,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.36951008439064026,
+      "rewards/margins": 4.2895612716674805,
+      "rewards/rejected": -3.920051336288452,
+      "step": 93,
+      "train_speed(iter/s)": 0.010735
+    },
+    {
+      "epoch": 0.541981981981982,
+      "grad_norm": 1.8191365003585815,
+      "learning_rate": 0.00019075754196709572,
+      "logits/chosen": -0.5523058176040649,
+      "logits/rejected": -0.7712856531143188,
+      "logps/chosen": -2.1930363178253174,
+      "logps/rejected": -66.22615814208984,
+      "loss": 0.28284692764282227,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2161579132080078,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.14300797879695892,
+      "rewards/margins": 5.302109718322754,
+      "rewards/rejected": -5.159101486206055,
+      "step": 94,
+      "train_speed(iter/s)": 0.010735
+    },
+    {
+      "epoch": 0.5477477477477477,
+      "grad_norm": 2.65755558013916,
+      "learning_rate": 0.00019048813098270387,
+      "logits/chosen": -0.49344098567962646,
+      "logits/rejected": -0.7781980633735657,
+      "logps/chosen": -3.545104742050171,
+      "logps/rejected": -78.23759460449219,
+      "loss": 0.3401697278022766,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3010118901729584,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.07626594603061676,
+      "rewards/margins": 5.846086025238037,
+      "rewards/rejected": -5.769820213317871,
+      "step": 95,
+      "train_speed(iter/s)": 0.010736
+    },
+    {
+      "epoch": 0.5535135135135135,
+      "grad_norm": 1.6140971183776855,
+      "learning_rate": 0.0001902150455168962,
+      "logits/chosen": -0.4496663808822632,
+      "logits/rejected": -0.7431093454360962,
+      "logps/chosen": -2.8349695205688477,
+      "logps/rejected": -72.62895965576172,
+      "loss": 0.3297802209854126,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2532678544521332,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.149441659450531,
+      "rewards/margins": 5.512978553771973,
+      "rewards/rejected": -5.363536357879639,
+      "step": 96,
+      "train_speed(iter/s)": 0.010735
+    },
+    {
+      "epoch": 0.5592792792792792,
+      "grad_norm": 2.243736743927002,
+      "learning_rate": 0.0001899382966589443,
+      "logits/chosen": -0.44952693581581116,
+      "logits/rejected": -0.7028639316558838,
+      "logps/chosen": -5.277416229248047,
+      "logps/rejected": -69.73493957519531,
+      "loss": 0.41486623883247375,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.25434425473213196,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.13436982035636902,
+      "rewards/margins": 4.8991923332214355,
+      "rewards/rejected": -4.764822006225586,
+      "step": 97,
+      "train_speed(iter/s)": 0.010735
+    },
+    {
+      "epoch": 0.565045045045045,
+      "grad_norm": 4.0099101066589355,
+      "learning_rate": 0.0001896578956468805,
+      "logits/chosen": -0.5480562448501587,
+      "logits/rejected": -0.7537853717803955,
+      "logps/chosen": -4.075699806213379,
+      "logps/rejected": -61.6773681640625,
+      "loss": 0.34289711713790894,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.24206680059432983,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.09123305231332779,
+      "rewards/margins": 4.509951114654541,
+      "rewards/rejected": -4.418718338012695,
+      "step": 98,
+      "train_speed(iter/s)": 0.010736
+    },
+    {
+      "epoch": 0.5708108108108109,
+      "grad_norm": 12.38167667388916,
+      "learning_rate": 0.00018937385386704126,
+      "logits/chosen": -0.6038029789924622,
+      "logits/rejected": -0.7428975105285645,
+      "logps/chosen": -6.938250541687012,
+      "logps/rejected": -58.5086555480957,
+      "loss": 0.4984804391860962,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.37261444330215454,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.10895882546901703,
+      "rewards/margins": 4.678299903869629,
+      "rewards/rejected": -4.569341659545898,
+      "step": 99,
+      "train_speed(iter/s)": 0.010736
+    },
+    {
+      "epoch": 0.5765765765765766,
+      "grad_norm": 4.255463123321533,
+      "learning_rate": 0.00018908618285360494,
+      "logits/chosen": -0.5780411958694458,
+      "logits/rejected": -0.7493711709976196,
+      "logps/chosen": -6.087734222412109,
+      "logps/rejected": -68.05068969726562,
+      "loss": 0.5377764701843262,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.4144035577774048,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.07401561737060547,
+      "rewards/margins": 5.022759437561035,
+      "rewards/rejected": -4.94874382019043,
+      "step": 100,
+      "train_speed(iter/s)": 0.010737
+    },
+    {
+      "epoch": 0.5823423423423424,
+      "grad_norm": 2.052121162414551,
+      "learning_rate": 0.00018879489428812334,
+      "logits/chosen": -0.5155804753303528,
+      "logits/rejected": -0.7454657554626465,
+      "logps/chosen": -4.093721389770508,
+      "logps/rejected": -73.28987121582031,
+      "loss": 0.4639092683792114,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3733065724372864,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.16241005063056946,
+      "rewards/margins": 5.644482135772705,
+      "rewards/rejected": -5.482071876525879,
+      "step": 101,
+      "train_speed(iter/s)": 0.010718
+    },
+    {
+      "epoch": 0.5881081081081081,
+      "grad_norm": 2.1512622833251953,
+      "learning_rate": 0.0001884999999990475,
+      "logits/chosen": -0.5633401274681091,
+      "logits/rejected": -0.7381998896598816,
+      "logps/chosen": -6.3428497314453125,
+      "logps/rejected": -52.1317138671875,
+      "loss": 0.4867536425590515,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.34330999851226807,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.33921706676483154,
+      "rewards/margins": 4.20556116104126,
+      "rewards/rejected": -3.8663439750671387,
+      "step": 102,
+      "train_speed(iter/s)": 0.010718
+    },
+    {
+      "epoch": 0.5938738738738739,
+      "grad_norm": 3.3250627517700195,
+      "learning_rate": 0.00018820151196124717,
+      "logits/chosen": -0.5157067179679871,
+      "logits/rejected": -0.8171908855438232,
+      "logps/chosen": -3.4731268882751465,
+      "logps/rejected": -87.99000549316406,
+      "loss": 0.5975285768508911,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.46800604462623596,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.06358014047145844,
+      "rewards/margins": 6.483148097991943,
+      "rewards/rejected": -6.419568061828613,
+      "step": 103,
+      "train_speed(iter/s)": 0.010718
+    },
+    {
+      "epoch": 0.5996396396396396,
+      "grad_norm": 3.476541757583618,
+      "learning_rate": 0.0001878994422955246,
+      "logits/chosen": -0.5131168365478516,
+      "logits/rejected": -0.8059231042861938,
+      "logps/chosen": -3.1174309253692627,
+      "logps/rejected": -86.71981811523438,
+      "loss": 0.3160158693790436,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.25080835819244385,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.28578388690948486,
+      "rewards/margins": 6.907861232757568,
+      "rewards/rejected": -6.622077941894531,
+      "step": 104,
+      "train_speed(iter/s)": 0.010718
+    },
+    {
+      "epoch": 0.6054054054054054,
+      "grad_norm": 5.32436990737915,
+      "learning_rate": 0.00018759380326812257,
+      "logits/chosen": -0.6340402960777283,
+      "logits/rejected": -0.867938756942749,
+      "logps/chosen": -2.323990821838379,
+      "logps/rejected": -71.03258514404297,
+      "loss": 0.31046488881111145,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.21423426270484924,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -0.011083951219916344,
+      "rewards/margins": 6.008091926574707,
+      "rewards/rejected": -6.0191755294799805,
+      "step": 105,
+      "train_speed(iter/s)": 0.010719
+    },
+    {
+      "epoch": 0.6111711711711711,
+      "grad_norm": 3.4041733741760254,
+      "learning_rate": 0.0001872846072902259,
+      "logits/chosen": -0.589099645614624,
+      "logits/rejected": -0.8120458126068115,
+      "logps/chosen": -1.5791796445846558,
+      "logps/rejected": -82.34098052978516,
+      "loss": 0.2268940657377243,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.14617523550987244,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.2277921438217163,
+      "rewards/margins": 6.835321426391602,
+      "rewards/rejected": -6.607529640197754,
+      "step": 106,
+      "train_speed(iter/s)": 0.010719
+    },
+    {
+      "epoch": 0.616936936936937,
+      "grad_norm": 1.9216675758361816,
+      "learning_rate": 0.00018697186691745782,
+      "logits/chosen": -0.599471926689148,
+      "logits/rejected": -0.8271054029464722,
+      "logps/chosen": -4.24713134765625,
+      "logps/rejected": -84.08229064941406,
+      "loss": 0.3841734826564789,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.31657740473747253,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.022511344403028488,
+      "rewards/margins": 6.769836902618408,
+      "rewards/rejected": -6.747325897216797,
+      "step": 107,
+      "train_speed(iter/s)": 0.01072
+    },
+    {
+      "epoch": 0.6227027027027027,
+      "grad_norm": 4.744035720825195,
+      "learning_rate": 0.0001866555948493699,
+      "logits/chosen": -0.5849882960319519,
+      "logits/rejected": -0.8727071285247803,
+      "logps/chosen": -2.891716718673706,
+      "logps/rejected": -92.83207702636719,
+      "loss": 0.31115370988845825,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.22799423336982727,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.19745981693267822,
+      "rewards/margins": 7.553801536560059,
+      "rewards/rejected": -7.35634183883667,
+      "step": 108,
+      "train_speed(iter/s)": 0.01072
+    },
+    {
+      "epoch": 0.6284684684684685,
+      "grad_norm": 2.028069019317627,
+      "learning_rate": 0.00018633580392892648,
+      "logits/chosen": -0.5856168270111084,
+      "logits/rejected": -0.8363710641860962,
+      "logps/chosen": -5.515890121459961,
+      "logps/rejected": -83.518310546875,
+      "loss": 0.40675532817840576,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.33311402797698975,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.19537094235420227,
+      "rewards/margins": 6.882720947265625,
+      "rewards/rejected": -6.687350273132324,
+      "step": 109,
+      "train_speed(iter/s)": 0.01072
+    },
+    {
+      "epoch": 0.6342342342342342,
+      "grad_norm": 1.8457213640213013,
+      "learning_rate": 0.00018601250714198302,
+      "logits/chosen": -0.6694027185440063,
+      "logits/rejected": -0.8495978713035583,
+      "logps/chosen": -6.527522087097168,
+      "logps/rejected": -69.24601745605469,
+      "loss": 0.4539996385574341,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3218171000480652,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.16864001750946045,
+      "rewards/margins": 5.4212541580200195,
+      "rewards/rejected": -5.252613544464111,
+      "step": 110,
+      "train_speed(iter/s)": 0.01072
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.4716756343841553,
+      "learning_rate": 0.00018568571761675893,
+      "logits/chosen": -0.5902320742607117,
+      "logits/rejected": -0.8840252757072449,
+      "logps/chosen": -5.05216121673584,
+      "logps/rejected": -96.3882064819336,
+      "loss": 0.3378267288208008,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2499268501996994,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.21043285727500916,
+      "rewards/margins": 7.197650909423828,
+      "rewards/rejected": -6.987217903137207,
+      "step": 111,
+      "train_speed(iter/s)": 0.01072
+    },
+    {
+      "epoch": 0.6457657657657657,
+      "grad_norm": 0.7805799245834351,
+      "learning_rate": 0.00018535544862330436,
+      "logits/chosen": -0.6686023473739624,
+      "logits/rejected": -0.986074686050415,
+      "logps/chosen": -2.231717824935913,
+      "logps/rejected": -96.76434326171875,
+      "loss": 0.16498756408691406,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.14222995936870575,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.14315873384475708,
+      "rewards/margins": 7.930060386657715,
+      "rewards/rejected": -7.786901473999023,
+      "step": 112,
+      "train_speed(iter/s)": 0.01072
+    },
+    {
+      "epoch": 0.6515315315315315,
+      "grad_norm": 2.115647792816162,
+      "learning_rate": 0.00018502171357296144,
+      "logits/chosen": -0.7186330556869507,
+      "logits/rejected": -0.9198458194732666,
+      "logps/chosen": -6.138432502746582,
+      "logps/rejected": -76.40232849121094,
+      "loss": 0.42215695977211,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.34653419256210327,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.3074129521846771,
+      "rewards/margins": 6.320471286773682,
+      "rewards/rejected": -6.013057708740234,
+      "step": 113,
+      "train_speed(iter/s)": 0.010721
+    },
+    {
+      "epoch": 0.6572972972972972,
+      "grad_norm": 8.20527458190918,
+      "learning_rate": 0.00018468452601781948,
+      "logits/chosen": -0.6733801364898682,
+      "logits/rejected": -0.9577184319496155,
+      "logps/chosen": -5.523393154144287,
+      "logps/rejected": -94.21669006347656,
+      "loss": 0.4980089068412781,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.37918829917907715,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.07570207118988037,
+      "rewards/margins": 7.661881446838379,
+      "rewards/rejected": -7.586179256439209,
+      "step": 114,
+      "train_speed(iter/s)": 0.010721
+    },
+    {
+      "epoch": 0.6630630630630631,
+      "grad_norm": 1.4824076890945435,
+      "learning_rate": 0.00018434389965016495,
+      "logits/chosen": -0.6685976386070251,
+      "logits/rejected": -0.9235626459121704,
+      "logps/chosen": -4.8032307624816895,
+      "logps/rejected": -89.84734344482422,
+      "loss": 0.3988059163093567,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3272039592266083,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.1318884938955307,
+      "rewards/margins": 7.2766876220703125,
+      "rewards/rejected": -7.14479923248291,
+      "step": 115,
+      "train_speed(iter/s)": 0.010721
+    },
+    {
+      "epoch": 0.6688288288288289,
+      "grad_norm": 6.457195281982422,
+      "learning_rate": 0.00018399984830192522,
+      "logits/chosen": -0.6563739776611328,
+      "logits/rejected": -0.9446012377738953,
+      "logps/chosen": -5.333343505859375,
+      "logps/rejected": -96.29754638671875,
+      "loss": 0.6648478507995605,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.5736638307571411,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.1766398400068283,
+      "rewards/margins": 7.466494560241699,
+      "rewards/rejected": -7.289853572845459,
+      "step": 116,
+      "train_speed(iter/s)": 0.010721
+    },
+    {
+      "epoch": 0.6745945945945946,
+      "grad_norm": 0.6062999367713928,
+      "learning_rate": 0.00018365238594410696,
+      "logits/chosen": -0.6826354265213013,
+      "logits/rejected": -0.9526041150093079,
+      "logps/chosen": -2.1560752391815186,
+      "logps/rejected": -97.08880615234375,
+      "loss": 0.15202483534812927,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.1097334772348404,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.08674149215221405,
+      "rewards/margins": 8.09538745880127,
+      "rewards/rejected": -8.008646011352539,
+      "step": 117,
+      "train_speed(iter/s)": 0.010721
+    },
+    {
+      "epoch": 0.6803603603603604,
+      "grad_norm": 4.866965293884277,
+      "learning_rate": 0.00018330152668622892,
+      "logits/chosen": -0.6270961761474609,
+      "logits/rejected": -0.8659616708755493,
+      "logps/chosen": -6.8050456047058105,
+      "logps/rejected": -94.24461364746094,
+      "loss": 0.5779068470001221,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.4634403884410858,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.08972588181495667,
+      "rewards/margins": 7.540095806121826,
+      "rewards/rejected": -7.450369834899902,
+      "step": 118,
+      "train_speed(iter/s)": 0.010721
+    },
+    {
+      "epoch": 0.6861261261261261,
+      "grad_norm": 1.9984689950942993,
+      "learning_rate": 0.00018294728477574886,
+      "logits/chosen": -0.6620824933052063,
+      "logits/rejected": -0.9279792308807373,
+      "logps/chosen": -3.62056303024292,
+      "logps/rejected": -88.79190826416016,
+      "loss": 0.5508708357810974,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.44049572944641113,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.04914311319589615,
+      "rewards/margins": 7.090651512145996,
+      "rewards/rejected": -7.041508197784424,
+      "step": 119,
+      "train_speed(iter/s)": 0.010721
+    },
+    {
+      "epoch": 0.6918918918918919,
+      "grad_norm": 2.634037971496582,
+      "learning_rate": 0.000182589674597485,
+      "logits/chosen": -0.6129943132400513,
+      "logits/rejected": -0.891259491443634,
+      "logps/chosen": -4.1637163162231445,
+      "logps/rejected": -89.52510833740234,
+      "loss": 0.5133303999900818,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.35086461901664734,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.015482299029827118,
+      "rewards/margins": 7.1094279289245605,
+      "rewards/rejected": -7.0939459800720215,
+      "step": 120,
+      "train_speed(iter/s)": 0.010722
+    },
+    {
+      "epoch": 0.6976576576576576,
+      "grad_norm": 2.7602248191833496,
+      "learning_rate": 0.00018222871067303192,
+      "logits/chosen": -0.644217848777771,
+      "logits/rejected": -0.8804385662078857,
+      "logps/chosen": -4.665724277496338,
+      "logps/rejected": -75.18159484863281,
+      "loss": 0.4733228385448456,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3284577429294586,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.11940830945968628,
+      "rewards/margins": 6.475360870361328,
+      "rewards/rejected": -6.355952262878418,
+      "step": 121,
+      "train_speed(iter/s)": 0.010722
+    },
+    {
+      "epoch": 0.7034234234234235,
+      "grad_norm": 8.460821151733398,
+      "learning_rate": 0.0001818644076601709,
+      "logits/chosen": -0.6995394825935364,
+      "logits/rejected": -0.9071613550186157,
+      "logps/chosen": -8.95744800567627,
+      "logps/rejected": -77.84011840820312,
+      "loss": 0.6963841915130615,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.5519821047782898,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.2606813609600067,
+      "rewards/margins": 6.603857040405273,
+      "rewards/rejected": -6.343176364898682,
+      "step": 122,
+      "train_speed(iter/s)": 0.010722
+    },
+    {
+      "epoch": 0.7091891891891892,
+      "grad_norm": 3.587855100631714,
+      "learning_rate": 0.00018149678035227473,
+      "logits/chosen": -0.6442317962646484,
+      "logits/rejected": -0.8487176895141602,
+      "logps/chosen": -6.872650623321533,
+      "logps/rejected": -82.89649963378906,
+      "loss": 0.4728536605834961,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3608010411262512,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.009775836020708084,
+      "rewards/margins": 6.596733093261719,
+      "rewards/rejected": -6.586956977844238,
+      "step": 123,
+      "train_speed(iter/s)": 0.010723
+    },
+    {
+      "epoch": 0.714954954954955,
+      "grad_norm": 1.7244524955749512,
+      "learning_rate": 0.00018112584367770685,
+      "logits/chosen": -0.6685165762901306,
+      "logits/rejected": -0.8828461766242981,
+      "logps/chosen": -5.397648334503174,
+      "logps/rejected": -85.60098266601562,
+      "loss": 0.389848530292511,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.32140427827835083,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.12320394814014435,
+      "rewards/margins": 7.233897686004639,
+      "rewards/rejected": -7.11069393157959,
+      "step": 124,
+      "train_speed(iter/s)": 0.010723
+    },
+    {
+      "epoch": 0.7207207207207207,
+      "grad_norm": 0.7455915808677673,
+      "learning_rate": 0.00018075161269921527,
+      "logits/chosen": -0.5614687204360962,
+      "logits/rejected": -0.8907942771911621,
+      "logps/chosen": -1.7438545227050781,
+      "logps/rejected": -92.23735809326172,
+      "loss": 0.12226351350545883,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.10527773201465607,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.15749390423297882,
+      "rewards/margins": 7.594954967498779,
+      "rewards/rejected": -7.437460422515869,
+      "step": 125,
+      "train_speed(iter/s)": 0.010723
+    },
+    {
+      "epoch": 0.7264864864864865,
+      "grad_norm": 1.0758198499679565,
+      "learning_rate": 0.000180374102613321,
+      "logits/chosen": -0.6551331877708435,
+      "logits/rejected": -0.8264045119285583,
+      "logps/chosen": -4.308568000793457,
+      "logps/rejected": -80.06883239746094,
+      "loss": 0.36240947246551514,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.24730846285820007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.10061657428741455,
+      "rewards/margins": 6.429587364196777,
+      "rewards/rejected": -6.328970909118652,
+      "step": 126,
+      "train_speed(iter/s)": 0.010723
+    },
+    {
+      "epoch": 0.7322522522522522,
+      "grad_norm": 1.0857983827590942,
+      "learning_rate": 0.0001799933287497007,
+      "logits/chosen": -0.6380553841590881,
+      "logits/rejected": -0.8699716925621033,
+      "logps/chosen": -1.9937247037887573,
+      "logps/rejected": -87.90292358398438,
+      "loss": 0.18343912065029144,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.12965701520442963,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.09683795273303986,
+      "rewards/margins": 7.11926794052124,
+      "rewards/rejected": -7.022429943084717,
+      "step": 127,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.738018018018018,
+      "grad_norm": 9.539369583129883,
+      "learning_rate": 0.00017960930657056438,
+      "logits/chosen": -0.6350418925285339,
+      "logits/rejected": -0.8755402565002441,
+      "logps/chosen": -2.645111560821533,
+      "logps/rejected": -82.07209777832031,
+      "loss": 0.23406416177749634,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.17941246926784515,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.07386763393878937,
+      "rewards/margins": 6.727149486541748,
+      "rewards/rejected": -6.653281211853027,
+      "step": 128,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.7437837837837837,
+      "grad_norm": 10.807381629943848,
+      "learning_rate": 0.00017922205167002754,
+      "logits/chosen": -0.6329447031021118,
+      "logits/rejected": -0.8397620320320129,
+      "logps/chosen": -4.6306023597717285,
+      "logps/rejected": -67.41867065429688,
+      "loss": 0.5849911570549011,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.41021403670310974,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.06239153444766998,
+      "rewards/margins": 5.311315059661865,
+      "rewards/rejected": -5.248923301696777,
+      "step": 129,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.7495495495495496,
+      "grad_norm": 1.8455349206924438,
+      "learning_rate": 0.00017883157977347774,
+      "logits/chosen": -0.5277632474899292,
+      "logits/rejected": -0.785979151725769,
+      "logps/chosen": -4.069696426391602,
+      "logps/rejected": -82.98876953125,
+      "loss": 0.3647957146167755,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.24460868537425995,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.12994280457496643,
+      "rewards/margins": 6.137853622436523,
+      "rewards/rejected": -6.00791072845459,
+      "step": 130,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.7553153153153153,
+      "grad_norm": 1.5157482624053955,
+      "learning_rate": 0.00017843790673693625,
+      "logits/chosen": -0.533730685710907,
+      "logits/rejected": -0.8200556635856628,
+      "logps/chosen": -4.038061141967773,
+      "logps/rejected": -82.76172637939453,
+      "loss": 0.2888023555278778,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2086402326822281,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.2728549540042877,
+      "rewards/margins": 6.415480613708496,
+      "rewards/rejected": -6.142626762390137,
+      "step": 131,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.7610810810810811,
+      "grad_norm": 1.936601996421814,
+      "learning_rate": 0.00017804104854641408,
+      "logits/chosen": -0.6249516606330872,
+      "logits/rejected": -0.8721716403961182,
+      "logps/chosen": -3.304114818572998,
+      "logps/rejected": -60.27552795410156,
+      "loss": 0.415643572807312,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.28065580129623413,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.2913169264793396,
+      "rewards/margins": 4.785688400268555,
+      "rewards/rejected": -4.49437141418457,
+      "step": 132,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.7668468468468469,
+      "grad_norm": 3.6359870433807373,
+      "learning_rate": 0.0001776410213172628,
+      "logits/chosen": -0.5627670288085938,
+      "logits/rejected": -0.8085115551948547,
+      "logps/chosen": -1.7719781398773193,
+      "logps/rejected": -72.61830139160156,
+      "loss": 0.2871711552143097,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.1900225728750229,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.089139424264431,
+      "rewards/margins": 5.8200178146362305,
+      "rewards/rejected": -5.730878829956055,
+      "step": 133,
+      "train_speed(iter/s)": 0.010725
+    },
+    {
+      "epoch": 0.7726126126126126,
+      "grad_norm": 1.4323631525039673,
+      "learning_rate": 0.00017723784129352018,
+      "logits/chosen": -0.6231422424316406,
+      "logits/rejected": -0.8504309058189392,
+      "logps/chosen": -1.730849027633667,
+      "logps/rejected": -68.78044128417969,
+      "loss": 0.2242366373538971,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.13950544595718384,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.18260075151920319,
+      "rewards/margins": 5.708943843841553,
+      "rewards/rejected": -5.526342391967773,
+      "step": 134,
+      "train_speed(iter/s)": 0.010725
+    },
+    {
+      "epoch": 0.7783783783783784,
+      "grad_norm": 1.6163887977600098,
+      "learning_rate": 0.00017683152484725066,
+      "logits/chosen": -0.6224597692489624,
+      "logits/rejected": -0.7799749374389648,
+      "logps/chosen": -4.717617988586426,
+      "logps/rejected": -61.31199264526367,
+      "loss": 0.4397742450237274,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3140389323234558,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.25990185141563416,
+      "rewards/margins": 4.902276039123535,
+      "rewards/rejected": -4.642374515533447,
+      "step": 135,
+      "train_speed(iter/s)": 0.010726
+    },
+    {
+      "epoch": 0.7841441441441441,
+      "grad_norm": 1.2988686561584473,
+      "learning_rate": 0.00017642208847788032,
+      "logits/chosen": -0.4455900192260742,
+      "logits/rejected": -0.8272579312324524,
+      "logps/chosen": -3.7604730129241943,
+      "logps/rejected": -91.75283813476562,
+      "loss": 0.3069111108779907,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.257954478263855,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.17708764970302582,
+      "rewards/margins": 7.115184783935547,
+      "rewards/rejected": -6.93809700012207,
+      "step": 136,
+      "train_speed(iter/s)": 0.010725
+    },
+    {
+      "epoch": 0.78990990990991,
+      "grad_norm": 3.960479974746704,
+      "learning_rate": 0.00017600954881152693,
+      "logits/chosen": -0.480897456407547,
+      "logits/rejected": -0.806812584400177,
+      "logps/chosen": -2.339344024658203,
+      "logps/rejected": -88.25721740722656,
+      "loss": 0.25455033779144287,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.1620534360408783,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.10610076785087585,
+      "rewards/margins": 6.920160293579102,
+      "rewards/rejected": -6.814059734344482,
+      "step": 137,
+      "train_speed(iter/s)": 0.010725
+    },
+    {
+      "epoch": 0.7956756756756757,
+      "grad_norm": 9.401166915893555,
+      "learning_rate": 0.00017559392260032506,
+      "logits/chosen": -0.4589274525642395,
+      "logits/rejected": -0.7634912729263306,
+      "logps/chosen": -5.8359503746032715,
+      "logps/rejected": -96.87335205078125,
+      "loss": 0.5270398855209351,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.43480122089385986,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -0.14989905059337616,
+      "rewards/margins": 7.5454607009887695,
+      "rewards/rejected": -7.69536018371582,
+      "step": 138,
+      "train_speed(iter/s)": 0.010725
+    },
+    {
+      "epoch": 0.8014414414414415,
+      "grad_norm": 4.686446666717529,
+      "learning_rate": 0.00017517522672174548,
+      "logits/chosen": -0.5010412931442261,
+      "logits/rejected": -0.8063571453094482,
+      "logps/chosen": -2.5103189945220947,
+      "logps/rejected": -96.94129943847656,
+      "loss": 0.3778824508190155,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.252981573343277,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.10596343874931335,
+      "rewards/margins": 7.605740070343018,
+      "rewards/rejected": -7.499776840209961,
+      "step": 139,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.8072072072072072,
+      "grad_norm": 1.6708407402038574,
+      "learning_rate": 0.00017475347817790996,
+      "logits/chosen": -0.4751094877719879,
+      "logits/rejected": -0.8238545060157776,
+      "logps/chosen": -3.2878451347351074,
+      "logps/rejected": -86.26036071777344,
+      "loss": 0.24279947578907013,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.16919009387493134,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.07166648656129837,
+      "rewards/margins": 6.587783336639404,
+      "rewards/rejected": -6.516116142272949,
+      "step": 140,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.812972972972973,
+      "grad_norm": 1.4335037469863892,
+      "learning_rate": 0.00017432869409490093,
+      "logits/chosen": -0.5489400625228882,
+      "logits/rejected": -0.7731753587722778,
+      "logps/chosen": -5.348051071166992,
+      "logps/rejected": -70.64154815673828,
+      "loss": 0.36098673939704895,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2713705897331238,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.23904970288276672,
+      "rewards/margins": 5.921088695526123,
+      "rewards/rejected": -5.682039737701416,
+      "step": 141,
+      "train_speed(iter/s)": 0.010725
+    },
+    {
+      "epoch": 0.8187387387387387,
+      "grad_norm": 2.012854814529419,
+      "learning_rate": 0.00017390089172206592,
+      "logits/chosen": -0.467896968126297,
+      "logits/rejected": -0.793892502784729,
+      "logps/chosen": -1.7580054998397827,
+      "logps/rejected": -85.476318359375,
+      "loss": 0.23942089080810547,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.17797623574733734,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.09021879732608795,
+      "rewards/margins": 6.8085222244262695,
+      "rewards/rejected": -6.718303680419922,
+      "step": 142,
+      "train_speed(iter/s)": 0.010725
+    },
+    {
+      "epoch": 0.8245045045045045,
+      "grad_norm": 5.257446765899658,
+      "learning_rate": 0.00017347008843131712,
+      "logits/chosen": -0.4858510196208954,
+      "logits/rejected": -0.7565414309501648,
+      "logps/chosen": -3.7341561317443848,
+      "logps/rejected": -75.44990539550781,
+      "loss": 0.43588101863861084,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3324872553348541,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": 0.1589755266904831,
+      "rewards/margins": 5.903942584991455,
+      "rewards/rejected": -5.744967460632324,
+      "step": 143,
+      "train_speed(iter/s)": 0.010725
+    },
+    {
+      "epoch": 0.8302702702702702,
+      "grad_norm": 1.9966129064559937,
+      "learning_rate": 0.00017303630171642607,
+      "logits/chosen": -0.4013749659061432,
+      "logits/rejected": -0.7098768353462219,
+      "logps/chosen": -2.0640015602111816,
+      "logps/rejected": -80.94207763671875,
+      "loss": 0.2669733166694641,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2289275974035263,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.20508992671966553,
+      "rewards/margins": 5.947912216186523,
+      "rewards/rejected": -5.742822170257568,
+      "step": 144,
+      "train_speed(iter/s)": 0.010725
+    },
+    {
+      "epoch": 0.836036036036036,
+      "grad_norm": 4.144430637359619,
+      "learning_rate": 0.0001725995491923131,
+      "logits/chosen": -0.34412139654159546,
+      "logits/rejected": -0.7268174290657043,
+      "logps/chosen": -2.0844953060150146,
+      "logps/rejected": -86.29147338867188,
+      "loss": 0.2860412001609802,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.21140313148498535,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.1814691722393036,
+      "rewards/margins": 6.325296401977539,
+      "rewards/rejected": -6.143826961517334,
+      "step": 145,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.8418018018018018,
+      "grad_norm": 6.901271343231201,
+      "learning_rate": 0.0001721598485943322,
+      "logits/chosen": -0.4293047785758972,
+      "logits/rejected": -0.6931589841842651,
+      "logps/chosen": -3.5573177337646484,
+      "logps/rejected": -80.00125885009766,
+      "loss": 0.402279257774353,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.30397528409957886,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -0.027988389134407043,
+      "rewards/margins": 6.108896732330322,
+      "rewards/rejected": -6.136884689331055,
+      "step": 146,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.8475675675675676,
+      "grad_norm": 1.3138624429702759,
+      "learning_rate": 0.00017171721777755074,
+      "logits/chosen": -0.441506952047348,
+      "logits/rejected": -0.6183236241340637,
+      "logps/chosen": -5.245753288269043,
+      "logps/rejected": -61.3995246887207,
+      "loss": 0.5000537633895874,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.3299512565135956,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.1415078043937683,
+      "rewards/margins": 4.831845760345459,
+      "rewards/rejected": -4.690337181091309,
+      "step": 147,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.8533333333333334,
+      "grad_norm": 2.7394113540649414,
+      "learning_rate": 0.00017127167471602447,
+      "logits/chosen": -0.44564759731292725,
+      "logits/rejected": -0.7324830889701843,
+      "logps/chosen": -2.2436249256134033,
+      "logps/rejected": -73.5162124633789,
+      "loss": 0.20336386561393738,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.11223578453063965,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.17428822815418243,
+      "rewards/margins": 5.678185939788818,
+      "rewards/rejected": -5.503897666931152,
+      "step": 148,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.8590990990990991,
+      "grad_norm": 1.488072395324707,
+      "learning_rate": 0.00017082323750206761,
+      "logits/chosen": -0.44784775376319885,
+      "logits/rejected": -0.6909451484680176,
+      "logps/chosen": -3.5523109436035156,
+      "logps/rejected": -70.1702651977539,
+      "loss": 0.31512442231178284,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2102314829826355,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": 0.11678414791822433,
+      "rewards/margins": 5.312015533447266,
+      "rewards/rejected": -5.1952314376831055,
+      "step": 149,
+      "train_speed(iter/s)": 0.010724
+    },
+    {
+      "epoch": 0.8648648648648649,
+      "grad_norm": 2.6578073501586914,
+      "learning_rate": 0.00017037192434551823,
+      "logits/chosen": -0.39348986744880676,
+      "logits/rejected": -0.690170407295227,
+      "logps/chosen": -3.4318065643310547,
+      "logps/rejected": -76.70863342285156,
+      "loss": 0.40308576822280884,
+      "memory(GiB)": 41.24,
+      "nll_loss": 0.2762233316898346,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": 0.1851682811975479,
+      "rewards/margins": 5.912222385406494,
+      "rewards/rejected": -5.727054119110107,
+      "step": 150,
+      "train_speed(iter/s)": 0.010724
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 519,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "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": 8.203837372786278e+16,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d6e7aa38d7b9243e5aa6f35f81c7761702583d1c76484ff540cfefab9c93572
+size 8632

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)