Upload folder using huggingface_hub

README.md

@@ -1,3 +1,202 @@
----
-license: apache-2.0
----
+---
+base_model: llava-hf/LLaVA-NeXT-Video-7B-hf
+library_name: peft
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.14.0

adapter_config.json

@@ -0,0 +1,29 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "/home/tzz/.cache/modelscope/hub/models/swift/LLaVA-NeXT-Video-7B-hf",
+  "bias": "none",
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": [],
+  "peft_type": "LORA",
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": "(?!^(multi_modal_projector))^(language_model)(?!.*(lora_A|lora_B|base_layer|emb|wte|shared|lm_head|output|score|v_head|classifier)).*",
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4402f2f8e389e6ae59c09a456ea4678952568fbff790eaba43ff1c6be7a97825
+size 639699488

additional_config.json

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

args.json

@@ -0,0 +1,356 @@
+{
+  "model": "swift/LLaVA-NeXT-Video-7B-hf",
+  "model_type": "llava_next_video_hf",
+  "model_revision": null,
+  "task_type": "causal_lm",
+  "torch_dtype": "bfloat16",
+  "attn_impl": null,
+  "num_labels": null,
+  "rope_scaling": null,
+  "device_map": null,
+  "max_memory": {},
+  "local_repo_path": null,
+  "template": "llava_next_video_hf",
+  "system": null,
+  "max_length": 4096,
+  "truncation_strategy": "delete",
+  "max_pixels": null,
+  "tools_prompt": "react_en",
+  "norm_bbox": null,
+  "padding_side": "right",
+  "loss_scale": "default",
+  "sequence_parallel_size": 1,
+  "use_chat_template": true,
+  "template_backend": "swift",
+  "dataset": [
+    "/data1/tzz/VQA/train_lsvq_swift_3000.json"
+  ],
+  "val_dataset": [],
+  "split_dataset_ratio": 0.01,
+  "data_seed": 42,
+  "dataset_num_proc": 1,
+  "streaming": false,
+  "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": 0.0,
+  "top_k": null,
+  "top_p": null,
+  "repetition_penalty": null,
+  "num_beams": 1,
+  "stream": false,
+  "stop_words": [],
+  "logprobs": false,
+  "top_logprobs": null,
+  "ckpt_dir": null,
+  "load_dataset_config": null,
+  "lora_modules": [],
+  "tuner_backend": "peft",
+  "train_type": "lora",
+  "adapters": [],
+  "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": "/data1/tzz/VQA/ckpt/llava_next_video/v2-20250226-080739",
+  "overwrite_output_dir": false,
+  "do_train": false,
+  "do_eval": false,
+  "do_predict": false,
+  "eval_strategy": "epoch",
+  "prediction_loss_only": false,
+  "per_device_train_batch_size": 1,
+  "per_device_eval_batch_size": 1,
+  "per_gpu_train_batch_size": null,
+  "per_gpu_eval_batch_size": null,
+  "gradient_accumulation_steps": 16,
+  "eval_accumulation_steps": null,
+  "eval_delay": 0,
+  "torch_empty_cache_steps": null,
+  "learning_rate": 1e-05,
+  "weight_decay": 0.1,
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.999,
+  "adam_epsilon": 1e-08,
+  "max_grad_norm": 1.0,
+  "num_train_epochs": 1.0,
+  "max_steps": -1,
+  "lr_scheduler_type": "cosine",
+  "lr_scheduler_kwargs": null,
+  "warmup_ratio": 0.05,
+  "warmup_steps": 0,
+  "log_level": "passive",
+  "log_level_replica": "warning",
+  "log_on_each_node": true,
+  "logging_dir": "/data1/tzz/VQA/ckpt/llava_next_video/v2-20250226-080739/runs",
+  "logging_strategy": "steps",
+  "logging_first_step": true,
+  "logging_steps": 1,
+  "logging_nan_inf_filter": true,
+  "save_strategy": "epoch",
+  "save_steps": 500,
+  "save_total_limit": 1,
+  "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": null,
+  "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,
+  "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": 3,
+      "offload_optimizer": {
+        "device": "none",
+        "pin_memory": true
+      },
+      "offload_param": {
+        "device": "none",
+        "pin_memory": true
+      },
+      "overlap_comm": true,
+      "contiguous_gradients": true,
+      "sub_group_size": 1000000000.0,
+      "reduce_bucket_size": "auto",
+      "stage3_prefetch_bucket_size": "auto",
+      "stage3_param_persistence_threshold": "auto",
+      "stage3_max_live_parameters": 1000000000.0,
+      "stage3_max_reuse_distance": 1000000000.0,
+      "stage3_gather_16bit_weights_on_model_save": 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": [
+    "wandb"
+  ],
+  "ddp_find_unused_parameters": null,
+  "ddp_bucket_cap_mb": null,
+  "ddp_broadcast_buffers": null,
+  "dataloader_pin_memory": true,
+  "dataloader_persistent_workers": false,
+  "skip_memory_metrics": true,
+  "use_legacy_prediction_loop": false,
+  "push_to_hub": false,
+  "resume_from_checkpoint": null,
+  "hub_model_id": null,
+  "hub_strategy": "every_save",
+  "hub_private_repo": null,
+  "hub_always_push": false,
+  "gradient_checkpointing": true,
+  "gradient_checkpointing_kwargs": null,
+  "include_inputs_for_metrics": false,
+  "include_for_metrics": [],
+  "eval_do_concat_batches": true,
+  "fp16_backend": "auto",
+  "evaluation_strategy": "epoch",
+  "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,
+  "dispatch_batches": null,
+  "split_batches": 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,
+  "freeze_parameters": [
+    "vision_tower",
+    "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": 128,
+  "lora_alpha": 256,
+  "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,
+  "use_liger": false,
+  "model_layer_cls_name": null,
+  "metric_warmup_step": 0,
+  "fsdp_num": 1,
+  "acc_steps": 1,
+  "swanlab_token": null,
+  "swanlab_project": null,
+  "swanlab_workspace": null,
+  "swanlab_exp_name": null,
+  "swanlab_mode": "cloud",
+  "add_version": true,
+  "resume_only_model": false,
+  "check_model": true,
+  "create_checkpoint_symlink": false,
+  "packing": false,
+  "lazy_tokenize": true,
+  "external_plugins": [],
+  "loss_type": null,
+  "optimizer": null,
+  "metric": null,
+  "acc_strategy": "token",
+  "rank": 0,
+  "global_world_size": 1,
+  "local_world_size": 1,
+  "model_suffix": "LLaVA-NeXT-Video-7B-hf",
+  "model_info": "ModelInfo(model_type='llava_next_video_hf', model_dir='/home/tzz/.cache/modelscope/hub/models/swift/LLaVA-NeXT-Video-7B-hf', torch_dtype=torch.bfloat16, max_model_len=4096, quant_method=None, quant_bits=None, rope_scaling={'factor': 2.5, 'type': 'linear'}, config=None, task_type='causal_lm', num_labels=None)",
+  "model_meta": "ModelMeta(model_type='llava_next_video_hf', model_groups=[ModelGroup(models=[Model(ms_model_id='swift/LLaVA-NeXT-Video-7B-DPO-hf', hf_model_id='llava-hf/LLaVA-NeXT-Video-7B-DPO-hf', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='swift/LLaVA-NeXT-Video-7B-32K-hf', hf_model_id='llava-hf/LLaVA-NeXT-Video-7B-32K-hf', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='swift/LLaVA-NeXT-Video-7B-hf', hf_model_id='llava-hf/LLaVA-NeXT-Video-7B-hf', model_path=None, ms_revision=None, hf_revision=None)], ignore_patterns=None, requires=None, tags=[])], template='llava_next_video_hf', get_function=<function get_model_tokenizer_llava_next_video at 0x7f31a2703880>, model_arch='llava_next_video_hf', architectures=['LlavaNextVideoForConditionalGeneration'], additional_saved_files=[], torch_dtype=None, is_multimodal=True, is_reward=False, task_type=None, ignore_patterns=['*.zip', '*.gguf', '*.pth', '*.pt', 'consolidated*', 'onnx/*', '*.safetensors.md', '*.msgpack', '*.onnx', '*.ot', '*.h5', '*.bin', '*.safetensors'], requires=['transformers>=4.42', 'av'], tags=[])",
+  "model_dir": "/home/tzz/.cache/modelscope/hub/models/swift/LLaVA-NeXT-Video-7B-hf",
+  "hub": "<class 'swift.hub.hub.MSHub'>",
+  "training_args": "Seq2SeqTrainingArguments(output_dir='/data1/tzz/VQA/ckpt/llava_next_video/v2-20250226-080739', overwrite_output_dir=False, do_train=False, do_eval=True, do_predict=False, eval_strategy=<IntervalStrategy.EPOCH: 'epoch'>, prediction_loss_only=False, per_device_train_batch_size=1, per_device_eval_batch_size=1, per_gpu_train_batch_size=None, per_gpu_eval_batch_size=None, gradient_accumulation_steps=16, eval_accumulation_steps=None, eval_delay=0, torch_empty_cache_steps=None, learning_rate=1e-05, weight_decay=0.1, adam_beta1=0.9, adam_beta2=0.999, adam_epsilon=1e-08, max_grad_norm=1.0, num_train_epochs=1.0, max_steps=-1, lr_scheduler_type=<SchedulerType.COSINE: 'cosine'>, lr_scheduler_kwargs=None, warmup_ratio=0.05, warmup_steps=0, log_level='passive', log_level_replica='warning', log_on_each_node=True, logging_dir='/data1/tzz/VQA/ckpt/llava_next_video/v2-20250226-080739/runs', logging_strategy=<IntervalStrategy.STEPS: 'steps'>, logging_first_step=True, logging_steps=1, logging_nan_inf_filter=True, save_strategy=<SaveStrategy.EPOCH: 'epoch'>, save_steps=500, save_total_limit=1, 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=None, dataloader_num_workers=4, dataloader_prefetch_factor=None, past_index=-1, run_name='/data1/tzz/VQA/ckpt/llava_next_video/v2-20250226-080739', 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}, 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': 3, 'offload_optimizer': {'device': 'none', 'pin_memory': True}, 'offload_param': {'device': 'none', 'pin_memory': True}, 'overlap_comm': True, 'contiguous_gradients': True, 'sub_group_size': 1000000000.0, 'reduce_bucket_size': 'auto', 'stage3_prefetch_bucket_size': 'auto', 'stage3_param_persistence_threshold': 'auto', 'stage3_max_live_parameters': 1000000000.0, 'stage3_max_reuse_distance': 1000000000.0, 'stage3_gather_16bit_weights_on_model_save': 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=['wandb'], ddp_find_unused_parameters=None, ddp_bucket_cap_mb=None, ddp_broadcast_buffers=None, dataloader_pin_memory=True, dataloader_persistent_workers=False, skip_memory_metrics=True, use_legacy_prediction_loop=False, push_to_hub=False, resume_from_checkpoint=None, hub_model_id=None, hub_strategy=<HubStrategy.EVERY_SAVE: 'every_save'>, hub_token=None, hub_private_repo=None, hub_always_push=False, gradient_checkpointing=True, gradient_checkpointing_kwargs=None, include_inputs_for_metrics=False, include_for_metrics=[], eval_do_concat_batches=True, fp16_backend='auto', evaluation_strategy='epoch', 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, dispatch_batches=None, split_batches=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, sortish_sampler=False, predict_with_generate=False, generation_max_length=None, generation_num_beams=None, generation_config=None, acc_strategy='token', sequence_parallel_size=1, check_model=True, train_sampler_random=True, is_encoder_decoder=False, metric_warmup_step=0, train_dataset_sample=-1, fsdp_num=1, acc_steps=1, train_type='lora', optimizer=None, local_repo_path=None, galore_config=None)"
+}

global_step185/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb142298d09b7266f4b455468c7c704fb3afa56f2afdc8e0e2378594f054e37c
+size 3837792176

global_step185/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:991e9c2223ae1c3d7654420f5be184b06db39d4b78c74586bcda26d94db1c516
+size 528622

latest

@@ -0,0 +1 @@
+global_step185

rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14b9c6bfdcac96caec5b34f932265a4ed33bc329eb6a8061c7a515c092f92e52
+size 14244

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3a2e1772b9728f7cdbde917d5843c9d3b3305888987ed6e39b76c36ce92fef7
+size 1064

trainer_state.json

@@ -0,0 +1,1714 @@
+{
+  "best_metric": 0.12199707,
+  "best_model_checkpoint": "/data1/tzz/VQA/ckpt/llava_next_video/v2-20250226-080739/checkpoint-185",
+  "epoch": 0.9966329966329966,
+  "eval_steps": 500,
+  "global_step": 185,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0053872053872053875,
+      "grad_norm": 14.159545000064247,
+      "learning_rate": 1.0000000000000002e-06,
+      "loss": 31.90625,
+      "memory(GiB)": 22.53,
+      "step": 1,
+      "train_speed(iter/s)": 0.022985
+    },
+    {
+      "epoch": 0.010774410774410775,
+      "grad_norm": 14.616283963493206,
+      "learning_rate": 2.0000000000000003e-06,
+      "loss": 31.5234375,
+      "memory(GiB)": 22.53,
+      "step": 2,
+      "train_speed(iter/s)": 0.028711
+    },
+    {
+      "epoch": 0.01616161616161616,
+      "grad_norm": 13.121864716464238,
+      "learning_rate": 3e-06,
+      "loss": 33.6796875,
+      "memory(GiB)": 22.53,
+      "step": 3,
+      "train_speed(iter/s)": 0.031289
+    },
+    {
+      "epoch": 0.02154882154882155,
+      "grad_norm": 11.258740067609244,
+      "learning_rate": 4.000000000000001e-06,
+      "loss": 31.8203125,
+      "memory(GiB)": 22.53,
+      "step": 4,
+      "train_speed(iter/s)": 0.032739
+    },
+    {
+      "epoch": 0.026936026936026935,
+      "grad_norm": 13.170936715126654,
+      "learning_rate": 5e-06,
+      "loss": 29.2109375,
+      "memory(GiB)": 22.55,
+      "step": 5,
+      "train_speed(iter/s)": 0.033213
+    },
+    {
+      "epoch": 0.03232323232323232,
+      "grad_norm": 14.330929445232412,
+      "learning_rate": 6e-06,
+      "loss": 28.078125,
+      "memory(GiB)": 22.55,
+      "step": 6,
+      "train_speed(iter/s)": 0.033152
+    },
+    {
+      "epoch": 0.03771043771043771,
+      "grad_norm": 13.548506738998086,
+      "learning_rate": 7e-06,
+      "loss": 27.5078125,
+      "memory(GiB)": 22.55,
+      "step": 7,
+      "train_speed(iter/s)": 0.033486
+    },
+    {
+      "epoch": 0.0430976430976431,
+      "grad_norm": 8.666929263748118,
+      "learning_rate": 8.000000000000001e-06,
+      "loss": 24.9609375,
+      "memory(GiB)": 22.56,
+      "step": 8,
+      "train_speed(iter/s)": 0.033686
+    },
+    {
+      "epoch": 0.048484848484848485,
+      "grad_norm": 11.066925048714,
+      "learning_rate": 9e-06,
+      "loss": 19.890625,
+      "memory(GiB)": 22.57,
+      "step": 9,
+      "train_speed(iter/s)": 0.0337
+    },
+    {
+      "epoch": 0.05387205387205387,
+      "grad_norm": 8.973276554829988,
+      "learning_rate": 1e-05,
+      "loss": 14.328125,
+      "memory(GiB)": 22.58,
+      "step": 10,
+      "train_speed(iter/s)": 0.03368
+    },
+    {
+      "epoch": 0.05925925925925926,
+      "grad_norm": 4.92025256078084,
+      "learning_rate": 9.999194339645292e-06,
+      "loss": 11.275390625,
+      "memory(GiB)": 22.58,
+      "step": 11,
+      "train_speed(iter/s)": 0.033773
+    },
+    {
+      "epoch": 0.06464646464646465,
+      "grad_norm": 2.5239985209180706,
+      "learning_rate": 9.996777618216608e-06,
+      "loss": 9.6875,
+      "memory(GiB)": 22.58,
+      "step": 12,
+      "train_speed(iter/s)": 0.034222
+    },
+    {
+      "epoch": 0.07003367003367003,
+      "grad_norm": 3.3965201854332046,
+      "learning_rate": 9.992750614536606e-06,
+      "loss": 7.869140625,
+      "memory(GiB)": 22.58,
+      "step": 13,
+      "train_speed(iter/s)": 0.034559
+    },
+    {
+      "epoch": 0.07542087542087542,
+      "grad_norm": 3.83999730322345,
+      "learning_rate": 9.987114626364172e-06,
+      "loss": 7.22265625,
+      "memory(GiB)": 22.58,
+      "step": 14,
+      "train_speed(iter/s)": 0.034521
+    },
+    {
+      "epoch": 0.08080808080808081,
+      "grad_norm": 4.824626769554233,
+      "learning_rate": 9.979871469976197e-06,
+      "loss": 7.0576171875,
+      "memory(GiB)": 22.58,
+      "step": 15,
+      "train_speed(iter/s)": 0.03441
+    },
+    {
+      "epoch": 0.0861952861952862,
+      "grad_norm": 3.043760096951554,
+      "learning_rate": 9.971023479582258e-06,
+      "loss": 5.4990234375,
+      "memory(GiB)": 22.58,
+      "step": 16,
+      "train_speed(iter/s)": 0.034356
+    },
+    {
+      "epoch": 0.09158249158249158,
+      "grad_norm": 1.3971212531371173,
+      "learning_rate": 9.960573506572391e-06,
+      "loss": 4.044921875,
+      "memory(GiB)": 22.58,
+      "step": 17,
+      "train_speed(iter/s)": 0.03432
+    },
+    {
+      "epoch": 0.09696969696969697,
+      "grad_norm": 1.5367962214559587,
+      "learning_rate": 9.948524918598175e-06,
+      "loss": 3.44189453125,
+      "memory(GiB)": 22.58,
+      "step": 18,
+      "train_speed(iter/s)": 0.034225
+    },
+    {
+      "epoch": 0.10235690235690235,
+      "grad_norm": 1.2329087385122603,
+      "learning_rate": 9.934881598487478e-06,
+      "loss": 3.4072265625,
+      "memory(GiB)": 22.58,
+      "step": 19,
+      "train_speed(iter/s)": 0.034123
+    },
+    {
+      "epoch": 0.10774410774410774,
+      "grad_norm": 0.8648810367159049,
+      "learning_rate": 9.91964794299315e-06,
+      "loss": 3.0048828125,
+      "memory(GiB)": 22.58,
+      "step": 20,
+      "train_speed(iter/s)": 0.03406
+    },
+    {
+      "epoch": 0.11313131313131314,
+      "grad_norm": 1.1333084548737522,
+      "learning_rate": 9.902828861376101e-06,
+      "loss": 2.973876953125,
+      "memory(GiB)": 22.58,
+      "step": 21,
+      "train_speed(iter/s)": 0.03407
+    },
+    {
+      "epoch": 0.11851851851851852,
+      "grad_norm": 1.67328747436259,
+      "learning_rate": 9.884429773823238e-06,
+      "loss": 2.460693359375,
+      "memory(GiB)": 22.58,
+      "step": 22,
+      "train_speed(iter/s)": 0.033985
+    },
+    {
+      "epoch": 0.12390572390572391,
+      "grad_norm": 0.8370283899907709,
+      "learning_rate": 9.864456609700726e-06,
+      "loss": 2.162109375,
+      "memory(GiB)": 22.58,
+      "step": 23,
+      "train_speed(iter/s)": 0.033817
+    },
+    {
+      "epoch": 0.1292929292929293,
+      "grad_norm": 0.7984037408374535,
+      "learning_rate": 9.842915805643156e-06,
+      "loss": 2.711669921875,
+      "memory(GiB)": 22.58,
+      "step": 24,
+      "train_speed(iter/s)": 0.033701
+    },
+    {
+      "epoch": 0.13468013468013468,
+      "grad_norm": 0.5877571918682093,
+      "learning_rate": 9.819814303479268e-06,
+      "loss": 1.707275390625,
+      "memory(GiB)": 22.58,
+      "step": 25,
+      "train_speed(iter/s)": 0.033872
+    },
+    {
+      "epoch": 0.14006734006734006,
+      "grad_norm": 1.4800629465642858,
+      "learning_rate": 9.79515954799483e-06,
+      "loss": 2.813720703125,
+      "memory(GiB)": 22.58,
+      "step": 26,
+      "train_speed(iter/s)": 0.034031
+    },
+    {
+      "epoch": 0.14545454545454545,
+      "grad_norm": 2.1222533390916443,
+      "learning_rate": 9.768959484533461e-06,
+      "loss": 3.59912109375,
+      "memory(GiB)": 22.58,
+      "step": 27,
+      "train_speed(iter/s)": 0.034169
+    },
+    {
+      "epoch": 0.15084175084175083,
+      "grad_norm": 0.8369490081884605,
+      "learning_rate": 9.741222556436132e-06,
+      "loss": 1.89404296875,
+      "memory(GiB)": 22.58,
+      "step": 28,
+      "train_speed(iter/s)": 0.034295
+    },
+    {
+      "epoch": 0.15622895622895622,
+      "grad_norm": 0.5854633514891076,
+      "learning_rate": 9.711957702320176e-06,
+      "loss": 1.986328125,
+      "memory(GiB)": 22.58,
+      "step": 29,
+      "train_speed(iter/s)": 0.034448
+    },
+    {
+      "epoch": 0.16161616161616163,
+      "grad_norm": 0.35782476089852655,
+      "learning_rate": 9.681174353198687e-06,
+      "loss": 2.087890625,
+      "memory(GiB)": 22.58,
+      "step": 30,
+      "train_speed(iter/s)": 0.034568
+    },
+    {
+      "epoch": 0.16700336700336701,
+      "grad_norm": 0.7861618699933016,
+      "learning_rate": 9.648882429441258e-06,
+      "loss": 2.669921875,
+      "memory(GiB)": 22.58,
+      "step": 31,
+      "train_speed(iter/s)": 0.034675
+    },
+    {
+      "epoch": 0.1723905723905724,
+      "grad_norm": 0.536791680824106,
+      "learning_rate": 9.615092337576987e-06,
+      "loss": 2.203125,
+      "memory(GiB)": 22.58,
+      "step": 32,
+      "train_speed(iter/s)": 0.034758
+    },
+    {
+      "epoch": 0.17777777777777778,
+      "grad_norm": 1.3726808261834198,
+      "learning_rate": 9.579814966940833e-06,
+      "loss": 2.114501953125,
+      "memory(GiB)": 22.58,
+      "step": 33,
+      "train_speed(iter/s)": 0.034839
+    },
+    {
+      "epoch": 0.18316498316498317,
+      "grad_norm": 0.8535138723050261,
+      "learning_rate": 9.543061686164374e-06,
+      "loss": 2.1591796875,
+      "memory(GiB)": 22.58,
+      "step": 34,
+      "train_speed(iter/s)": 0.034969
+    },
+    {
+      "epoch": 0.18855218855218855,
+      "grad_norm": 0.6726334477065563,
+      "learning_rate": 9.504844339512096e-06,
+      "loss": 2.35791015625,
+      "memory(GiB)": 22.58,
+      "step": 35,
+      "train_speed(iter/s)": 0.035076
+    },
+    {
+      "epoch": 0.19393939393939394,
+      "grad_norm": 0.7227226956981251,
+      "learning_rate": 9.465175243064428e-06,
+      "loss": 2.400390625,
+      "memory(GiB)": 22.58,
+      "step": 36,
+      "train_speed(iter/s)": 0.035195
+    },
+    {
+      "epoch": 0.19932659932659932,
+      "grad_norm": 0.7075241914063357,
+      "learning_rate": 9.424067180748692e-06,
+      "loss": 1.476318359375,
+      "memory(GiB)": 22.58,
+      "step": 37,
+      "train_speed(iter/s)": 0.035278
+    },
+    {
+      "epoch": 0.2047138047138047,
+      "grad_norm": 0.8285808812880359,
+      "learning_rate": 9.381533400219319e-06,
+      "loss": 2.50634765625,
+      "memory(GiB)": 22.58,
+      "step": 38,
+      "train_speed(iter/s)": 0.035354
+    },
+    {
+      "epoch": 0.2101010101010101,
+      "grad_norm": 0.747109858212397,
+      "learning_rate": 9.337587608588588e-06,
+      "loss": 2.397216796875,
+      "memory(GiB)": 22.58,
+      "step": 39,
+      "train_speed(iter/s)": 0.035434
+    },
+    {
+      "epoch": 0.21548821548821548,
+      "grad_norm": 0.8997236382866319,
+      "learning_rate": 9.292243968009332e-06,
+      "loss": 2.3466796875,
+      "memory(GiB)": 22.58,
+      "step": 40,
+      "train_speed(iter/s)": 0.035447
+    },
+    {
+      "epoch": 0.22087542087542086,
+      "grad_norm": 0.3854506877674985,
+      "learning_rate": 9.24551709111097e-06,
+      "loss": 1.607421875,
+      "memory(GiB)": 22.58,
+      "step": 41,
+      "train_speed(iter/s)": 0.035398
+    },
+    {
+      "epoch": 0.22626262626262628,
+      "grad_norm": 0.4259732475000951,
+      "learning_rate": 9.197422036290386e-06,
+      "loss": 1.921630859375,
+      "memory(GiB)": 22.58,
+      "step": 42,
+      "train_speed(iter/s)": 0.035349
+    },
+    {
+      "epoch": 0.23164983164983166,
+      "grad_norm": 0.46150408574103824,
+      "learning_rate": 9.147974302859158e-06,
+      "loss": 1.41650390625,
+      "memory(GiB)": 22.58,
+      "step": 43,
+      "train_speed(iter/s)": 0.035321
+    },
+    {
+      "epoch": 0.23703703703703705,
+      "grad_norm": 0.5918291232050616,
+      "learning_rate": 9.09718982604866e-06,
+      "loss": 1.58154296875,
+      "memory(GiB)": 22.58,
+      "step": 44,
+      "train_speed(iter/s)": 0.03529
+    },
+    {
+      "epoch": 0.24242424242424243,
+      "grad_norm": 1.1984794966626473,
+      "learning_rate": 9.045084971874738e-06,
+      "loss": 2.67236328125,
+      "memory(GiB)": 22.58,
+      "step": 45,
+      "train_speed(iter/s)": 0.035244
+    },
+    {
+      "epoch": 0.24781144781144782,
+      "grad_norm": 0.7304425352094286,
+      "learning_rate": 8.991676531863507e-06,
+      "loss": 1.993408203125,
+      "memory(GiB)": 22.58,
+      "step": 46,
+      "train_speed(iter/s)": 0.0352
+    },
+    {
+      "epoch": 0.2531986531986532,
+      "grad_norm": 0.8247667804924503,
+      "learning_rate": 8.936981717640061e-06,
+      "loss": 2.8740234375,
+      "memory(GiB)": 22.58,
+      "step": 47,
+      "train_speed(iter/s)": 0.035111
+    },
+    {
+      "epoch": 0.2585858585858586,
+      "grad_norm": 1.072788633508109,
+      "learning_rate": 8.881018155381766e-06,
+      "loss": 1.845458984375,
+      "memory(GiB)": 22.58,
+      "step": 48,
+      "train_speed(iter/s)": 0.035139
+    },
+    {
+      "epoch": 0.26397306397306397,
+      "grad_norm": 0.6949566674892941,
+      "learning_rate": 8.823803880137993e-06,
+      "loss": 2.345458984375,
+      "memory(GiB)": 22.58,
+      "step": 49,
+      "train_speed(iter/s)": 0.035224
+    },
+    {
+      "epoch": 0.26936026936026936,
+      "grad_norm": 0.3214051528089464,
+      "learning_rate": 8.765357330018056e-06,
+      "loss": 1.640869140625,
+      "memory(GiB)": 22.58,
+      "step": 50,
+      "train_speed(iter/s)": 0.035311
+    },
+    {
+      "epoch": 0.27474747474747474,
+      "grad_norm": 0.8127331172569063,
+      "learning_rate": 8.705697340249275e-06,
+      "loss": 2.334716796875,
+      "memory(GiB)": 22.58,
+      "step": 51,
+      "train_speed(iter/s)": 0.035368
+    },
+    {
+      "epoch": 0.2801346801346801,
+      "grad_norm": 0.6993353179443554,
+      "learning_rate": 8.644843137107058e-06,
+      "loss": 2.2666015625,
+      "memory(GiB)": 22.58,
+      "step": 52,
+      "train_speed(iter/s)": 0.03541
+    },
+    {
+      "epoch": 0.2855218855218855,
+      "grad_norm": 0.7930646229400613,
+      "learning_rate": 8.582814331718961e-06,
+      "loss": 1.73876953125,
+      "memory(GiB)": 22.58,
+      "step": 53,
+      "train_speed(iter/s)": 0.035443
+    },
+    {
+      "epoch": 0.2909090909090909,
+      "grad_norm": 0.47348696234661886,
+      "learning_rate": 8.519630913744726e-06,
+      "loss": 1.8544921875,
+      "memory(GiB)": 22.58,
+      "step": 54,
+      "train_speed(iter/s)": 0.035485
+    },
+    {
+      "epoch": 0.2962962962962963,
+      "grad_norm": 0.5105789152298116,
+      "learning_rate": 8.455313244934324e-06,
+      "loss": 2.10107421875,
+      "memory(GiB)": 22.58,
+      "step": 55,
+      "train_speed(iter/s)": 0.03552
+    },
+    {
+      "epoch": 0.30168350168350166,
+      "grad_norm": 0.48874730617457113,
+      "learning_rate": 8.389882052566106e-06,
+      "loss": 2.19189453125,
+      "memory(GiB)": 22.58,
+      "step": 56,
+      "train_speed(iter/s)": 0.035547
+    },
+    {
+      "epoch": 0.30707070707070705,
+      "grad_norm": 0.7017590448005361,
+      "learning_rate": 8.32335842276713e-06,
+      "loss": 1.605224609375,
+      "memory(GiB)": 22.58,
+      "step": 57,
+      "train_speed(iter/s)": 0.035484
+    },
+    {
+      "epoch": 0.31245791245791243,
+      "grad_norm": 0.7736924894631574,
+      "learning_rate": 8.255763793717868e-06,
+      "loss": 2.123779296875,
+      "memory(GiB)": 22.58,
+      "step": 58,
+      "train_speed(iter/s)": 0.035432
+    },
+    {
+      "epoch": 0.3178451178451178,
+      "grad_norm": 0.6091631207035194,
+      "learning_rate": 8.18711994874345e-06,
+      "loss": 1.8798828125,
+      "memory(GiB)": 22.58,
+      "step": 59,
+      "train_speed(iter/s)": 0.035351
+    },
+    {
+      "epoch": 0.32323232323232326,
+      "grad_norm": 0.6745360872937951,
+      "learning_rate": 8.117449009293668e-06,
+      "loss": 2.36767578125,
+      "memory(GiB)": 22.58,
+      "step": 60,
+      "train_speed(iter/s)": 0.035291
+    },
+    {
+      "epoch": 0.32861952861952864,
+      "grad_norm": 1.1170607516843722,
+      "learning_rate": 8.046773427814043e-06,
+      "loss": 2.153076171875,
+      "memory(GiB)": 22.58,
+      "step": 61,
+      "train_speed(iter/s)": 0.035255
+    },
+    {
+      "epoch": 0.33400673400673403,
+      "grad_norm": 0.42517306211931166,
+      "learning_rate": 7.975115980510187e-06,
+      "loss": 1.717041015625,
+      "memory(GiB)": 22.58,
+      "step": 62,
+      "train_speed(iter/s)": 0.035224
+    },
+    {
+      "epoch": 0.3393939393939394,
+      "grad_norm": 0.8043024113222557,
+      "learning_rate": 7.902499760007867e-06,
+      "loss": 1.85888671875,
+      "memory(GiB)": 22.58,
+      "step": 63,
+      "train_speed(iter/s)": 0.035142
+    },
+    {
+      "epoch": 0.3447811447811448,
+      "grad_norm": 0.9761638945939747,
+      "learning_rate": 7.828948167911073e-06,
+      "loss": 1.906005859375,
+      "memory(GiB)": 22.58,
+      "step": 64,
+      "train_speed(iter/s)": 0.035063
+    },
+    {
+      "epoch": 0.3501683501683502,
+      "grad_norm": 0.4137734068293326,
+      "learning_rate": 7.754484907260513e-06,
+      "loss": 2.05712890625,
+      "memory(GiB)": 22.58,
+      "step": 65,
+      "train_speed(iter/s)": 0.034992
+    },
+    {
+      "epoch": 0.35555555555555557,
+      "grad_norm": 0.6313489954771672,
+      "learning_rate": 7.679133974894984e-06,
+      "loss": 1.56591796875,
+      "memory(GiB)": 22.58,
+      "step": 66,
+      "train_speed(iter/s)": 0.035062
+    },
+    {
+      "epoch": 0.36094276094276095,
+      "grad_norm": 0.7916770866661113,
+      "learning_rate": 7.602919653718044e-06,
+      "loss": 1.32373046875,
+      "memory(GiB)": 22.58,
+      "step": 67,
+      "train_speed(iter/s)": 0.035123
+    },
+    {
+      "epoch": 0.36632996632996634,
+      "grad_norm": 0.7005145101509135,
+      "learning_rate": 7.5258665048725065e-06,
+      "loss": 1.677490234375,
+      "memory(GiB)": 22.58,
+      "step": 68,
+      "train_speed(iter/s)": 0.035192
+    },
+    {
+      "epoch": 0.3717171717171717,
+      "grad_norm": 0.5600472715983401,
+      "learning_rate": 7.447999359825263e-06,
+      "loss": 1.8934326171875,
+      "memory(GiB)": 22.58,
+      "step": 69,
+      "train_speed(iter/s)": 0.035242
+    },
+    {
+      "epoch": 0.3771043771043771,
+      "grad_norm": 0.7799156688047453,
+      "learning_rate": 7.369343312364994e-06,
+      "loss": 1.737060546875,
+      "memory(GiB)": 22.58,
+      "step": 70,
+      "train_speed(iter/s)": 0.035303
+    },
+    {
+      "epoch": 0.3824915824915825,
+      "grad_norm": 1.0088361337375438,
+      "learning_rate": 7.289923710515338e-06,
+      "loss": 2.55859375,
+      "memory(GiB)": 22.58,
+      "step": 71,
+      "train_speed(iter/s)": 0.035339
+    },
+    {
+      "epoch": 0.3878787878787879,
+      "grad_norm": 0.7778606766770365,
+      "learning_rate": 7.2097661483661355e-06,
+      "loss": 2.3927001953125,
+      "memory(GiB)": 22.58,
+      "step": 72,
+      "train_speed(iter/s)": 0.035398
+    },
+    {
+      "epoch": 0.39326599326599326,
+      "grad_norm": 0.7503526567701239,
+      "learning_rate": 7.128896457825364e-06,
+      "loss": 2.4095458984375,
+      "memory(GiB)": 22.58,
+      "step": 73,
+      "train_speed(iter/s)": 0.035435
+    },
+    {
+      "epoch": 0.39865319865319865,
+      "grad_norm": 0.9293852718192778,
+      "learning_rate": 7.047340700294454e-06,
+      "loss": 2.0943603515625,
+      "memory(GiB)": 22.58,
+      "step": 74,
+      "train_speed(iter/s)": 0.035473
+    },
+    {
+      "epoch": 0.40404040404040403,
+      "grad_norm": 1.2981158494810365,
+      "learning_rate": 6.965125158269619e-06,
+      "loss": 2.36279296875,
+      "memory(GiB)": 22.58,
+      "step": 75,
+      "train_speed(iter/s)": 0.035498
+    },
+    {
+      "epoch": 0.4094276094276094,
+      "grad_norm": 0.5915357318010657,
+      "learning_rate": 6.88227632687196e-06,
+      "loss": 1.13037109375,
+      "memory(GiB)": 22.58,
+      "step": 76,
+      "train_speed(iter/s)": 0.035521
+    },
+    {
+      "epoch": 0.4148148148148148,
+      "grad_norm": 0.8289109263502568,
+      "learning_rate": 6.798820905309036e-06,
+      "loss": 2.245849609375,
+      "memory(GiB)": 22.58,
+      "step": 77,
+      "train_speed(iter/s)": 0.035549
+    },
+    {
+      "epoch": 0.4202020202020202,
+      "grad_norm": 0.7332772758108902,
+      "learning_rate": 6.714785788270658e-06,
+      "loss": 1.794189453125,
+      "memory(GiB)": 22.58,
+      "step": 78,
+      "train_speed(iter/s)": 0.035574
+    },
+    {
+      "epoch": 0.4255892255892256,
+      "grad_norm": 0.8695389561000924,
+      "learning_rate": 6.63019805726171e-06,
+      "loss": 2.107177734375,
+      "memory(GiB)": 22.58,
+      "step": 79,
+      "train_speed(iter/s)": 0.035564
+    },
+    {
+      "epoch": 0.43097643097643096,
+      "grad_norm": 1.0578963540355828,
+      "learning_rate": 6.545084971874738e-06,
+      "loss": 2.2099609375,
+      "memory(GiB)": 22.58,
+      "step": 80,
+      "train_speed(iter/s)": 0.035518
+    },
+    {
+      "epoch": 0.43636363636363634,
+      "grad_norm": 0.5355473518839581,
+      "learning_rate": 6.459473961005168e-06,
+      "loss": 1.679931640625,
+      "memory(GiB)": 22.58,
+      "step": 81,
+      "train_speed(iter/s)": 0.035449
+    },
+    {
+      "epoch": 0.4417508417508417,
+      "grad_norm": 0.47562295475695077,
+      "learning_rate": 6.373392614011952e-06,
+      "loss": 1.548828125,
+      "memory(GiB)": 22.58,
+      "step": 82,
+      "train_speed(iter/s)": 0.03541
+    },
+    {
+      "epoch": 0.4471380471380471,
+      "grad_norm": 1.1873250939202482,
+      "learning_rate": 6.286868671826513e-06,
+      "loss": 2.3310546875,
+      "memory(GiB)": 22.58,
+      "step": 83,
+      "train_speed(iter/s)": 0.035383
+    },
+    {
+      "epoch": 0.45252525252525255,
+      "grad_norm": 0.6325848523967413,
+      "learning_rate": 6.19993001801283e-06,
+      "loss": 1.63232421875,
+      "memory(GiB)": 22.58,
+      "step": 84,
+      "train_speed(iter/s)": 0.035357
+    },
+    {
+      "epoch": 0.45791245791245794,
+      "grad_norm": 0.6180246232374331,
+      "learning_rate": 6.112604669781572e-06,
+      "loss": 2.5283203125,
+      "memory(GiB)": 22.58,
+      "step": 85,
+      "train_speed(iter/s)": 0.035328
+    },
+    {
+      "epoch": 0.4632996632996633,
+      "grad_norm": 0.9254342636136799,
+      "learning_rate": 6.024920768961153e-06,
+      "loss": 2.09814453125,
+      "memory(GiB)": 22.58,
+      "step": 86,
+      "train_speed(iter/s)": 0.03531
+    },
+    {
+      "epoch": 0.4686868686868687,
+      "grad_norm": 1.0220943585915119,
+      "learning_rate": 5.936906572928625e-06,
+      "loss": 1.8603515625,
+      "memory(GiB)": 22.58,
+      "step": 87,
+      "train_speed(iter/s)": 0.035243
+    },
+    {
+      "epoch": 0.4740740740740741,
+      "grad_norm": 0.547874150160307,
+      "learning_rate": 5.848590445503345e-06,
+      "loss": 2.2890625,
+      "memory(GiB)": 22.58,
+      "step": 88,
+      "train_speed(iter/s)": 0.03516
+    },
+    {
+      "epoch": 0.4794612794612795,
+      "grad_norm": 0.7203446700675221,
+      "learning_rate": 5.760000847806337e-06,
+      "loss": 1.68115234375,
+      "memory(GiB)": 22.58,
+      "step": 89,
+      "train_speed(iter/s)": 0.035117
+    },
+    {
+      "epoch": 0.48484848484848486,
+      "grad_norm": 0.7628245708662847,
+      "learning_rate": 5.671166329088278e-06,
+      "loss": 2.126953125,
+      "memory(GiB)": 22.58,
+      "step": 90,
+      "train_speed(iter/s)": 0.035147
+    },
+    {
+      "epoch": 0.49023569023569025,
+      "grad_norm": 0.8089999734614459,
+      "learning_rate": 5.582115517529114e-06,
+      "loss": 1.948486328125,
+      "memory(GiB)": 22.58,
+      "step": 91,
+      "train_speed(iter/s)": 0.035179
+    },
+    {
+      "epoch": 0.49562289562289563,
+      "grad_norm": 0.5039876551970663,
+      "learning_rate": 5.4928771110122185e-06,
+      "loss": 1.849853515625,
+      "memory(GiB)": 22.58,
+      "step": 92,
+      "train_speed(iter/s)": 0.035212
+    },
+    {
+      "epoch": 0.501010101010101,
+      "grad_norm": 0.9008917409254343,
+      "learning_rate": 5.403479867876087e-06,
+      "loss": 2.642578125,
+      "memory(GiB)": 22.58,
+      "step": 93,
+      "train_speed(iter/s)": 0.035235
+    },
+    {
+      "epoch": 0.5063973063973064,
+      "grad_norm": 1.1384096826151604,
+      "learning_rate": 5.3139525976465675e-06,
+      "loss": 2.49365234375,
+      "memory(GiB)": 22.58,
+      "step": 94,
+      "train_speed(iter/s)": 0.035265
+    },
+    {
+      "epoch": 0.5117845117845118,
+      "grad_norm": 0.7491826485818727,
+      "learning_rate": 5.224324151752575e-06,
+      "loss": 1.88037109375,
+      "memory(GiB)": 22.58,
+      "step": 95,
+      "train_speed(iter/s)": 0.035291
+    },
+    {
+      "epoch": 0.5171717171717172,
+      "grad_norm": 0.6169314437426718,
+      "learning_rate": 5.134623414228315e-06,
+      "loss": 1.485595703125,
+      "memory(GiB)": 22.58,
+      "step": 96,
+      "train_speed(iter/s)": 0.035322
+    },
+    {
+      "epoch": 0.5225589225589226,
+      "grad_norm": 0.7458411085328407,
+      "learning_rate": 5.04487929240499e-06,
+      "loss": 2.030517578125,
+      "memory(GiB)": 22.58,
+      "step": 97,
+      "train_speed(iter/s)": 0.035358
+    },
+    {
+      "epoch": 0.5279461279461279,
+      "grad_norm": 0.36969067992245414,
+      "learning_rate": 4.955120707595011e-06,
+      "loss": 1.82421875,
+      "memory(GiB)": 22.58,
+      "step": 98,
+      "train_speed(iter/s)": 0.03539
+    },
+    {
+      "epoch": 0.5333333333333333,
+      "grad_norm": 0.7184526746731991,
+      "learning_rate": 4.865376585771687e-06,
+      "loss": 2.1650390625,
+      "memory(GiB)": 22.58,
+      "step": 99,
+      "train_speed(iter/s)": 0.035417
+    },
+    {
+      "epoch": 0.5387205387205387,
+      "grad_norm": 0.5860047275017632,
+      "learning_rate": 4.775675848247427e-06,
+      "loss": 2.016845703125,
+      "memory(GiB)": 22.58,
+      "step": 100,
+      "train_speed(iter/s)": 0.035451
+    },
+    {
+      "epoch": 0.5441077441077441,
+      "grad_norm": 0.6740666234718802,
+      "learning_rate": 4.686047402353433e-06,
+      "loss": 1.481689453125,
+      "memory(GiB)": 22.58,
+      "step": 101,
+      "train_speed(iter/s)": 0.035484
+    },
+    {
+      "epoch": 0.5494949494949495,
+      "grad_norm": 0.5962985498733315,
+      "learning_rate": 4.596520132123915e-06,
+      "loss": 2.225341796875,
+      "memory(GiB)": 22.58,
+      "step": 102,
+      "train_speed(iter/s)": 0.035522
+    },
+    {
+      "epoch": 0.5548821548821549,
+      "grad_norm": 0.6185754487719404,
+      "learning_rate": 4.507122888987782e-06,
+      "loss": 2.630615234375,
+      "memory(GiB)": 22.58,
+      "step": 103,
+      "train_speed(iter/s)": 0.035566
+    },
+    {
+      "epoch": 0.5602693602693603,
+      "grad_norm": 0.8891703200104817,
+      "learning_rate": 4.417884482470887e-06,
+      "loss": 1.98291015625,
+      "memory(GiB)": 22.58,
+      "step": 104,
+      "train_speed(iter/s)": 0.03558
+    },
+    {
+      "epoch": 0.5656565656565656,
+      "grad_norm": 0.5620520767612842,
+      "learning_rate": 4.3288336709117246e-06,
+      "loss": 1.933349609375,
+      "memory(GiB)": 22.58,
+      "step": 105,
+      "train_speed(iter/s)": 0.035549
+    },
+    {
+      "epoch": 0.571043771043771,
+      "grad_norm": 1.3690550098042635,
+      "learning_rate": 4.239999152193664e-06,
+      "loss": 2.217529296875,
+      "memory(GiB)": 22.58,
+      "step": 106,
+      "train_speed(iter/s)": 0.035527
+    },
+    {
+      "epoch": 0.5764309764309764,
+      "grad_norm": 0.4160377433886458,
+      "learning_rate": 4.1514095544966556e-06,
+      "loss": 1.737060546875,
+      "memory(GiB)": 22.58,
+      "step": 107,
+      "train_speed(iter/s)": 0.03551
+    },
+    {
+      "epoch": 0.5818181818181818,
+      "grad_norm": 0.8209806760015574,
+      "learning_rate": 4.063093427071376e-06,
+      "loss": 2.782470703125,
+      "memory(GiB)": 22.58,
+      "step": 108,
+      "train_speed(iter/s)": 0.035486
+    },
+    {
+      "epoch": 0.5872053872053872,
+      "grad_norm": 0.726795857048424,
+      "learning_rate": 3.975079231038848e-06,
+      "loss": 2.009521484375,
+      "memory(GiB)": 22.58,
+      "step": 109,
+      "train_speed(iter/s)": 0.035449
+    },
+    {
+      "epoch": 0.5925925925925926,
+      "grad_norm": 1.2624010183388914,
+      "learning_rate": 3.887395330218429e-06,
+      "loss": 2.59814453125,
+      "memory(GiB)": 22.58,
+      "step": 110,
+      "train_speed(iter/s)": 0.035431
+    },
+    {
+      "epoch": 0.597979797979798,
+      "grad_norm": 0.7513165711048129,
+      "learning_rate": 3.8000699819871704e-06,
+      "loss": 1.6396484375,
+      "memory(GiB)": 22.58,
+      "step": 111,
+      "train_speed(iter/s)": 0.035402
+    },
+    {
+      "epoch": 0.6033670033670033,
+      "grad_norm": 0.4587115862936887,
+      "learning_rate": 3.7131313281734895e-06,
+      "loss": 2.044189453125,
+      "memory(GiB)": 22.58,
+      "step": 112,
+      "train_speed(iter/s)": 0.035373
+    },
+    {
+      "epoch": 0.6087542087542087,
+      "grad_norm": 0.41256540620865373,
+      "learning_rate": 3.62660738598805e-06,
+      "loss": 1.9287109375,
+      "memory(GiB)": 22.58,
+      "step": 113,
+      "train_speed(iter/s)": 0.03534
+    },
+    {
+      "epoch": 0.6141414141414141,
+      "grad_norm": 0.4286929355926436,
+      "learning_rate": 3.540526038994834e-06,
+      "loss": 1.646728515625,
+      "memory(GiB)": 22.58,
+      "step": 114,
+      "train_speed(iter/s)": 0.035359
+    },
+    {
+      "epoch": 0.6195286195286195,
+      "grad_norm": 0.8246295061207459,
+      "learning_rate": 3.4549150281252635e-06,
+      "loss": 1.7587890625,
+      "memory(GiB)": 22.58,
+      "step": 115,
+      "train_speed(iter/s)": 0.035387
+    },
+    {
+      "epoch": 0.6249158249158249,
+      "grad_norm": 0.653674454928138,
+      "learning_rate": 3.3698019427382912e-06,
+      "loss": 1.9765625,
+      "memory(GiB)": 22.58,
+      "step": 116,
+      "train_speed(iter/s)": 0.035417
+    },
+    {
+      "epoch": 0.6303030303030303,
+      "grad_norm": 0.6402748838297282,
+      "learning_rate": 3.2852142117293435e-06,
+      "loss": 1.94970703125,
+      "memory(GiB)": 22.58,
+      "step": 117,
+      "train_speed(iter/s)": 0.035431
+    },
+    {
+      "epoch": 0.6356902356902356,
+      "grad_norm": 0.5582058394376362,
+      "learning_rate": 3.2011790946909673e-06,
+      "loss": 1.9755859375,
+      "memory(GiB)": 22.58,
+      "step": 118,
+      "train_speed(iter/s)": 0.03546
+    },
+    {
+      "epoch": 0.641077441077441,
+      "grad_norm": 0.8447371297083311,
+      "learning_rate": 3.11772367312804e-06,
+      "loss": 1.784423828125,
+      "memory(GiB)": 22.58,
+      "step": 119,
+      "train_speed(iter/s)": 0.035493
+    },
+    {
+      "epoch": 0.6464646464646465,
+      "grad_norm": 0.7640836687261319,
+      "learning_rate": 3.0348748417303826e-06,
+      "loss": 1.76171875,
+      "memory(GiB)": 22.58,
+      "step": 120,
+      "train_speed(iter/s)": 0.035513
+    },
+    {
+      "epoch": 0.6518518518518519,
+      "grad_norm": 0.6689239585125656,
+      "learning_rate": 2.9526592997055488e-06,
+      "loss": 2.076904296875,
+      "memory(GiB)": 22.58,
+      "step": 121,
+      "train_speed(iter/s)": 0.03554
+    },
+    {
+      "epoch": 0.6572390572390573,
+      "grad_norm": 0.8205443169011045,
+      "learning_rate": 2.871103542174637e-06,
+      "loss": 2.4423828125,
+      "memory(GiB)": 22.58,
+      "step": 122,
+      "train_speed(iter/s)": 0.035564
+    },
+    {
+      "epoch": 0.6626262626262627,
+      "grad_norm": 0.3861380215034983,
+      "learning_rate": 2.790233851633868e-06,
+      "loss": 1.405517578125,
+      "memory(GiB)": 22.58,
+      "step": 123,
+      "train_speed(iter/s)": 0.035589
+    },
+    {
+      "epoch": 0.6680134680134681,
+      "grad_norm": 0.9319720706049784,
+      "learning_rate": 2.7100762894846633e-06,
+      "loss": 1.884033203125,
+      "memory(GiB)": 22.58,
+      "step": 124,
+      "train_speed(iter/s)": 0.035611
+    },
+    {
+      "epoch": 0.6734006734006734,
+      "grad_norm": 0.4894495365923113,
+      "learning_rate": 2.6306566876350072e-06,
+      "loss": 1.992431640625,
+      "memory(GiB)": 22.58,
+      "step": 125,
+      "train_speed(iter/s)": 0.035618
+    },
+    {
+      "epoch": 0.6787878787878788,
+      "grad_norm": 0.5156966779296556,
+      "learning_rate": 2.55200064017474e-06,
+      "loss": 1.7987060546875,
+      "memory(GiB)": 22.58,
+      "step": 126,
+      "train_speed(iter/s)": 0.035593
+    },
+    {
+      "epoch": 0.6841750841750842,
+      "grad_norm": 0.39627149470201456,
+      "learning_rate": 2.4741334951274948e-06,
+      "loss": 1.779541015625,
+      "memory(GiB)": 22.58,
+      "step": 127,
+      "train_speed(iter/s)": 0.035563
+    },
+    {
+      "epoch": 0.6895622895622896,
+      "grad_norm": 0.7990132228587018,
+      "learning_rate": 2.3970803462819586e-06,
+      "loss": 2.385498046875,
+      "memory(GiB)": 22.58,
+      "step": 128,
+      "train_speed(iter/s)": 0.035533
+    },
+    {
+      "epoch": 0.694949494949495,
+      "grad_norm": 0.542336867995926,
+      "learning_rate": 2.320866025105016e-06,
+      "loss": 1.775390625,
+      "memory(GiB)": 22.58,
+      "step": 129,
+      "train_speed(iter/s)": 0.035487
+    },
+    {
+      "epoch": 0.7003367003367004,
+      "grad_norm": 0.40553603638944413,
+      "learning_rate": 2.245515092739488e-06,
+      "loss": 1.65771484375,
+      "memory(GiB)": 22.58,
+      "step": 130,
+      "train_speed(iter/s)": 0.035457
+    },
+    {
+      "epoch": 0.7057239057239058,
+      "grad_norm": 0.5705311307759141,
+      "learning_rate": 2.171051832088928e-06,
+      "loss": 1.392578125,
+      "memory(GiB)": 22.58,
+      "step": 131,
+      "train_speed(iter/s)": 0.035439
+    },
+    {
+      "epoch": 0.7111111111111111,
+      "grad_norm": 0.5637194621292295,
+      "learning_rate": 2.097500239992132e-06,
+      "loss": 1.808349609375,
+      "memory(GiB)": 22.58,
+      "step": 132,
+      "train_speed(iter/s)": 0.035412
+    },
+    {
+      "epoch": 0.7164983164983165,
+      "grad_norm": 1.0166298249729564,
+      "learning_rate": 2.0248840194898155e-06,
+      "loss": 1.88232421875,
+      "memory(GiB)": 22.58,
+      "step": 133,
+      "train_speed(iter/s)": 0.035367
+    },
+    {
+      "epoch": 0.7218855218855219,
+      "grad_norm": 0.365517442677317,
+      "learning_rate": 1.95322657218596e-06,
+      "loss": 1.8359375,
+      "memory(GiB)": 22.58,
+      "step": 134,
+      "train_speed(iter/s)": 0.035304
+    },
+    {
+      "epoch": 0.7272727272727273,
+      "grad_norm": 0.5937921378630181,
+      "learning_rate": 1.8825509907063328e-06,
+      "loss": 2.16162109375,
+      "memory(GiB)": 22.58,
+      "step": 135,
+      "train_speed(iter/s)": 0.035244
+    },
+    {
+      "epoch": 0.7326599326599327,
+      "grad_norm": 0.5630691840328598,
+      "learning_rate": 1.8128800512565514e-06,
+      "loss": 1.953369140625,
+      "memory(GiB)": 22.58,
+      "step": 136,
+      "train_speed(iter/s)": 0.035272
+    },
+    {
+      "epoch": 0.7380471380471381,
+      "grad_norm": 0.9036946278139879,
+      "learning_rate": 1.7442362062821323e-06,
+      "loss": 3.1923828125,
+      "memory(GiB)": 22.58,
+      "step": 137,
+      "train_speed(iter/s)": 0.035287
+    },
+    {
+      "epoch": 0.7434343434343434,
+      "grad_norm": 0.5335511498935785,
+      "learning_rate": 1.6766415772328732e-06,
+      "loss": 1.705322265625,
+      "memory(GiB)": 22.58,
+      "step": 138,
+      "train_speed(iter/s)": 0.035295
+    },
+    {
+      "epoch": 0.7488215488215488,
+      "grad_norm": 0.8149099249815346,
+      "learning_rate": 1.610117947433897e-06,
+      "loss": 2.81689453125,
+      "memory(GiB)": 22.58,
+      "step": 139,
+      "train_speed(iter/s)": 0.035308
+    },
+    {
+      "epoch": 0.7542087542087542,
+      "grad_norm": 0.5287002241309334,
+      "learning_rate": 1.544686755065677e-06,
+      "loss": 1.266357421875,
+      "memory(GiB)": 22.58,
+      "step": 140,
+      "train_speed(iter/s)": 0.035318
+    },
+    {
+      "epoch": 0.7595959595959596,
+      "grad_norm": 0.6139302197140588,
+      "learning_rate": 1.4803690862552755e-06,
+      "loss": 1.817626953125,
+      "memory(GiB)": 22.58,
+      "step": 141,
+      "train_speed(iter/s)": 0.035343
+    },
+    {
+      "epoch": 0.764983164983165,
+      "grad_norm": 0.6333656991964685,
+      "learning_rate": 1.4171856682810386e-06,
+      "loss": 2.101806640625,
+      "memory(GiB)": 22.58,
+      "step": 142,
+      "train_speed(iter/s)": 0.035364
+    },
+    {
+      "epoch": 0.7703703703703704,
+      "grad_norm": 0.8829740683592863,
+      "learning_rate": 1.3551568628929434e-06,
+      "loss": 2.508056640625,
+      "memory(GiB)": 22.58,
+      "step": 143,
+      "train_speed(iter/s)": 0.03539
+    },
+    {
+      "epoch": 0.7757575757575758,
+      "grad_norm": 0.5801508492146695,
+      "learning_rate": 1.2943026597507268e-06,
+      "loss": 1.6142578125,
+      "memory(GiB)": 22.58,
+      "step": 144,
+      "train_speed(iter/s)": 0.035413
+    },
+    {
+      "epoch": 0.7811447811447811,
+      "grad_norm": 0.48056036748223746,
+      "learning_rate": 1.234642669981946e-06,
+      "loss": 1.942138671875,
+      "memory(GiB)": 22.58,
+      "step": 145,
+      "train_speed(iter/s)": 0.035431
+    },
+    {
+      "epoch": 0.7865319865319865,
+      "grad_norm": 0.5473637984491948,
+      "learning_rate": 1.1761961198620081e-06,
+      "loss": 1.748779296875,
+      "memory(GiB)": 22.58,
+      "step": 146,
+      "train_speed(iter/s)": 0.035455
+    },
+    {
+      "epoch": 0.7919191919191919,
+      "grad_norm": 0.7226102542834439,
+      "learning_rate": 1.118981844618236e-06,
+      "loss": 1.657470703125,
+      "memory(GiB)": 22.58,
+      "step": 147,
+      "train_speed(iter/s)": 0.035472
+    },
+    {
+      "epoch": 0.7973063973063973,
+      "grad_norm": 0.677002948688539,
+      "learning_rate": 1.06301828235994e-06,
+      "loss": 1.730224609375,
+      "memory(GiB)": 22.58,
+      "step": 148,
+      "train_speed(iter/s)": 0.035492
+    },
+    {
+      "epoch": 0.8026936026936027,
+      "grad_norm": 0.4690206204454014,
+      "learning_rate": 1.0083234681364934e-06,
+      "loss": 1.97509765625,
+      "memory(GiB)": 22.58,
+      "step": 149,
+      "train_speed(iter/s)": 0.035513
+    },
+    {
+      "epoch": 0.8080808080808081,
+      "grad_norm": 0.38431237166068455,
+      "learning_rate": 9.549150281252633e-07,
+      "loss": 1.977783203125,
+      "memory(GiB)": 22.58,
+      "step": 150,
+      "train_speed(iter/s)": 0.035533
+    },
+    {
+      "epoch": 0.8134680134680135,
+      "grad_norm": 1.4318443328161967,
+      "learning_rate": 9.028101739513406e-07,
+      "loss": 2.696533203125,
+      "memory(GiB)": 22.58,
+      "step": 151,
+      "train_speed(iter/s)": 0.035549
+    },
+    {
+      "epoch": 0.8188552188552188,
+      "grad_norm": 0.39825000243591335,
+      "learning_rate": 8.520256971408453e-07,
+      "loss": 1.52294921875,
+      "memory(GiB)": 22.58,
+      "step": 152,
+      "train_speed(iter/s)": 0.035566
+    },
+    {
+      "epoch": 0.8242424242424242,
+      "grad_norm": 0.403223921534723,
+      "learning_rate": 8.025779637096138e-07,
+      "loss": 2.0869140625,
+      "memory(GiB)": 22.58,
+      "step": 153,
+      "train_speed(iter/s)": 0.035581
+    },
+    {
+      "epoch": 0.8296296296296296,
+      "grad_norm": 0.39408518518211616,
+      "learning_rate": 7.544829088890326e-07,
+      "loss": 2.085693359375,
+      "memory(GiB)": 22.58,
+      "step": 154,
+      "train_speed(iter/s)": 0.035601
+    },
+    {
+      "epoch": 0.835016835016835,
+      "grad_norm": 0.6580639598152973,
+      "learning_rate": 7.077560319906696e-07,
+      "loss": 1.58740234375,
+      "memory(GiB)": 22.58,
+      "step": 155,
+      "train_speed(iter/s)": 0.035601
+    },
+    {
+      "epoch": 0.8404040404040404,
+      "grad_norm": 0.5455643216936216,
+      "learning_rate": 6.624123914114122e-07,
+      "loss": 1.76953125,
+      "memory(GiB)": 22.58,
+      "step": 156,
+      "train_speed(iter/s)": 0.035584
+    },
+    {
+      "epoch": 0.8457912457912458,
+      "grad_norm": 0.9580661740362665,
+      "learning_rate": 6.184665997806832e-07,
+      "loss": 2.3505859375,
+      "memory(GiB)": 22.58,
+      "step": 157,
+      "train_speed(iter/s)": 0.035562
+    },
+    {
+      "epoch": 0.8511784511784511,
+      "grad_norm": 0.49273093322057226,
+      "learning_rate": 5.759328192513075e-07,
+      "loss": 1.632080078125,
+      "memory(GiB)": 22.58,
+      "step": 158,
+      "train_speed(iter/s)": 0.035543
+    },
+    {
+      "epoch": 0.8565656565656565,
+      "grad_norm": 0.5074137587596991,
+      "learning_rate": 5.348247569355736e-07,
+      "loss": 1.71240234375,
+      "memory(GiB)": 22.58,
+      "step": 159,
+      "train_speed(iter/s)": 0.03552
+    },
+    {
+      "epoch": 0.8619528619528619,
+      "grad_norm": 0.7185716029221749,
+      "learning_rate": 4.951556604879049e-07,
+      "loss": 2.36669921875,
+      "memory(GiB)": 22.58,
+      "step": 160,
+      "train_speed(iter/s)": 0.035503
+    },
+    {
+      "epoch": 0.8673400673400673,
+      "grad_norm": 0.7811542866299452,
+      "learning_rate": 4.569383138356276e-07,
+      "loss": 1.678955078125,
+      "memory(GiB)": 22.58,
+      "step": 161,
+      "train_speed(iter/s)": 0.035485
+    },
+    {
+      "epoch": 0.8727272727272727,
+      "grad_norm": 0.5128778192942757,
+      "learning_rate": 4.201850330591678e-07,
+      "loss": 2.072998046875,
+      "memory(GiB)": 22.58,
+      "step": 162,
+      "train_speed(iter/s)": 0.035459
+    },
+    {
+      "epoch": 0.8781144781144781,
+      "grad_norm": 0.6851552480944826,
+      "learning_rate": 3.8490766242301356e-07,
+      "loss": 1.55322265625,
+      "memory(GiB)": 22.58,
+      "step": 163,
+      "train_speed(iter/s)": 0.035422
+    },
+    {
+      "epoch": 0.8835016835016835,
+      "grad_norm": 1.0656634793505568,
+      "learning_rate": 3.511175705587433e-07,
+      "loss": 2.09228515625,
+      "memory(GiB)": 22.58,
+      "step": 164,
+      "train_speed(iter/s)": 0.035412
+    },
+    {
+      "epoch": 0.8888888888888888,
+      "grad_norm": 0.4704867924767551,
+      "learning_rate": 3.18825646801314e-07,
+      "loss": 2.178955078125,
+      "memory(GiB)": 22.58,
+      "step": 165,
+      "train_speed(iter/s)": 0.03542
+    },
+    {
+      "epoch": 0.8942760942760942,
+      "grad_norm": 0.3438531193817133,
+      "learning_rate": 2.8804229767982637e-07,
+      "loss": 1.828125,
+      "memory(GiB)": 22.58,
+      "step": 166,
+      "train_speed(iter/s)": 0.035441
+    },
+    {
+      "epoch": 0.8996632996632996,
+      "grad_norm": 0.9072486327466182,
+      "learning_rate": 2.587774435638679e-07,
+      "loss": 1.902099609375,
+      "memory(GiB)": 22.58,
+      "step": 167,
+      "train_speed(iter/s)": 0.035458
+    },
+    {
+      "epoch": 0.9050505050505051,
+      "grad_norm": 0.40209833194248146,
+      "learning_rate": 2.3104051546654016e-07,
+      "loss": 1.72314453125,
+      "memory(GiB)": 22.58,
+      "step": 168,
+      "train_speed(iter/s)": 0.035472
+    },
+    {
+      "epoch": 0.9104377104377105,
+      "grad_norm": 0.6534758670706157,
+      "learning_rate": 2.0484045200517222e-07,
+      "loss": 1.73095703125,
+      "memory(GiB)": 22.58,
+      "step": 169,
+      "train_speed(iter/s)": 0.03548
+    },
+    {
+      "epoch": 0.9158249158249159,
+      "grad_norm": 0.36229213242531244,
+      "learning_rate": 1.801856965207338e-07,
+      "loss": 1.954345703125,
+      "memory(GiB)": 22.58,
+      "step": 170,
+      "train_speed(iter/s)": 0.035499
+    },
+    {
+      "epoch": 0.9212121212121213,
+      "grad_norm": 0.41462023840060064,
+      "learning_rate": 1.5708419435684463e-07,
+      "loss": 1.726318359375,
+      "memory(GiB)": 22.58,
+      "step": 171,
+      "train_speed(iter/s)": 0.035512
+    },
+    {
+      "epoch": 0.9265993265993266,
+      "grad_norm": 0.793282464162017,
+      "learning_rate": 1.3554339029927532e-07,
+      "loss": 2.07861328125,
+      "memory(GiB)": 22.58,
+      "step": 172,
+      "train_speed(iter/s)": 0.035526
+    },
+    {
+      "epoch": 0.931986531986532,
+      "grad_norm": 0.4924403822397691,
+      "learning_rate": 1.1557022617676217e-07,
+      "loss": 1.400634765625,
+      "memory(GiB)": 22.58,
+      "step": 173,
+      "train_speed(iter/s)": 0.035541
+    },
+    {
+      "epoch": 0.9373737373737374,
+      "grad_norm": 0.41980069690346106,
+      "learning_rate": 9.717113862389993e-08,
+      "loss": 2.12158203125,
+      "memory(GiB)": 22.58,
+      "step": 174,
+      "train_speed(iter/s)": 0.03556
+    },
+    {
+      "epoch": 0.9427609427609428,
+      "grad_norm": 0.8809220146060189,
+      "learning_rate": 8.035205700685167e-08,
+      "loss": 2.621826171875,
+      "memory(GiB)": 22.58,
+      "step": 175,
+      "train_speed(iter/s)": 0.035577
+    },
+    {
+      "epoch": 0.9481481481481482,
+      "grad_norm": 0.6908254679787823,
+      "learning_rate": 6.511840151252169e-08,
+      "loss": 1.813232421875,
+      "memory(GiB)": 22.58,
+      "step": 176,
+      "train_speed(iter/s)": 0.035597
+    },
+    {
+      "epoch": 0.9535353535353536,
+      "grad_norm": 0.49484208186969647,
+      "learning_rate": 5.1475081401825553e-08,
+      "loss": 1.9814453125,
+      "memory(GiB)": 22.58,
+      "step": 177,
+      "train_speed(iter/s)": 0.035578
+    },
+    {
+      "epoch": 0.958922558922559,
+      "grad_norm": 0.6989450753180266,
+      "learning_rate": 3.9426493427611177e-08,
+      "loss": 1.78466796875,
+      "memory(GiB)": 22.58,
+      "step": 178,
+      "train_speed(iter/s)": 0.035563
+    },
+    {
+      "epoch": 0.9643097643097643,
+      "grad_norm": 0.5543481036485521,
+      "learning_rate": 2.8976520417742794e-08,
+      "loss": 1.727783203125,
+      "memory(GiB)": 22.58,
+      "step": 179,
+      "train_speed(iter/s)": 0.035552
+    },
+    {
+      "epoch": 0.9696969696969697,
+      "grad_norm": 0.5545843045026326,
+      "learning_rate": 2.012853002380466e-08,
+      "loss": 1.75634765625,
+      "memory(GiB)": 22.58,
+      "step": 180,
+      "train_speed(iter/s)": 0.035543
+    },
+    {
+      "epoch": 0.9750841750841751,
+      "grad_norm": 0.3433152184276571,
+      "learning_rate": 1.2885373635829756e-08,
+      "loss": 1.64208984375,
+      "memory(GiB)": 22.58,
+      "step": 181,
+      "train_speed(iter/s)": 0.035533
+    },
+    {
+      "epoch": 0.9804713804713805,
+      "grad_norm": 0.9002229182397717,
+      "learning_rate": 7.249385463395375e-09,
+      "loss": 2.177490234375,
+      "memory(GiB)": 22.58,
+      "step": 182,
+      "train_speed(iter/s)": 0.035517
+    },
+    {
+      "epoch": 0.9858585858585859,
+      "grad_norm": 0.5840020558119475,
+      "learning_rate": 3.2223817833931803e-09,
+      "loss": 1.4775390625,
+      "memory(GiB)": 22.58,
+      "step": 183,
+      "train_speed(iter/s)": 0.035499
+    },
+    {
+      "epoch": 0.9912457912457913,
+      "grad_norm": 0.31651969118225726,
+      "learning_rate": 8.056603547090813e-10,
+      "loss": 1.804931640625,
+      "memory(GiB)": 22.58,
+      "step": 184,
+      "train_speed(iter/s)": 0.035513
+    },
+    {
+      "epoch": 0.9966329966329966,
+      "grad_norm": 0.5699524292753597,
+      "learning_rate": 0.0,
+      "loss": 1.653076171875,
+      "memory(GiB)": 22.58,
+      "step": 185,
+      "train_speed(iter/s)": 0.035529
+    },
+    {
+      "epoch": 0.9966329966329966,
+      "eval_loss": 0.12199707329273224,
+      "eval_runtime": 16.4404,
+      "eval_samples_per_second": 1.825,
+      "eval_steps_per_second": 1.825,
+      "step": 185
+    },
+    {
+      "epoch": 0.9966329966329966,
+      "eval_loss": 0.12199707329273224,
+      "eval_runtime": 18.3596,
+      "eval_samples_per_second": 1.634,
+      "eval_steps_per_second": 1.634,
+      "step": 185
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 185,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 664501364736.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:549e622b17a582e38ce2f4e01eadb3074fb9d02d89f1d257ec75ba841132e4e4
+size 8120

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)