运行微调时报错“torch.cuda.OutOfMemoryError: CUDA out of memory. ”

这是我的训练参数

TrainingArguments( _n_gpu=1, accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True}, adafactor=False, adam_beta1=0.9, adam_beta2=0.95, adam_epsilon=1e-08, auto_find_batch_size=False, bf16=True, bf16_full_eval=False, cache_dir=None, data_seed=None, dataloader_drop_last=False, dataloader_num_workers=0, dataloader_persistent_workers=False, dataloader_pin_memory=True, dataloader_prefetch_factor=None, ddp_backend=None, ddp_broadcast_buffers=None, ddp_bucket_cap_mb=None, ddp_find_unused_parameters=None, ddp_timeout=1800, debug=[], deepspeed=./finetune/ds_config_zero2.json, disable_tqdm=False, dispatch_batches=None, do_eval=False, do_predict=False, do_train=False, eval_accumulation_steps=None, eval_delay=0, eval_steps=None, evaluation_strategy=no, fp16=False, fp16_backend=auto, fp16_full_eval=False, fp16_opt_level=O1, fsdp=[], fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}, fsdp_min_num_params=0, fsdp_transformer_layer_cls_to_wrap=None, full_determinism=False, gradient_accumulation_steps=1, gradient_checkpointing=True, gradient_checkpointing_kwargs=None, greater_is_better=None, group_by_length=False, half_precision_backend=auto, hub_always_push=False, hub_model_id=None, hub_private_repo=False, hub_strategy=every_save, hub_token=, ignore_data_skip=False, include_inputs_for_metrics=False, include_num_input_tokens_seen=False, include_tokens_per_second=False, jit_mode_eval=False, label_names=None, label_smoothing_factor=0.0, learning_rate=5e-06, length_column_name=length, load_best_model_at_end=False, local_rank=1, log_level=passive, log_level_replica=warning, log_on_each_node=False, logging_dir=./outputs/runs/Apr29_11-42-07_anming, logging_first_step=False, logging_nan_inf_filter=True, logging_steps=1.0, logging_strategy=steps, lr_scheduler_kwargs={}, lr_scheduler_type=cosine, max_grad_norm=1.0, max_steps=-1, metric_for_best_model=None, model_max_length=4096, mp_parameters=, neftune_noise_alpha=None, no_cuda=False, num_train_epochs=3.0, optim=adamw_torch, optim_args=None, output_dir=./outputs/, overwrite_output_dir=False, past_index=-1, per_device_eval_batch_size=8, per_device_train_batch_size=16, prediction_loss_only=False, push_to_hub=False, push_to_hub_model_id=None, push_to_hub_organization=None, push_to_hub_token=, ray_scope=last, remove_unused_columns=True, report_to=['tensorboard'], resume_from_checkpoint=None, run_name=./outputs/, save_on_each_node=False, save_only_model=False, save_safetensors=True, save_steps=200, save_strategy=steps, save_total_limit=None, seed=1024, skip_memory_metrics=True, split_batches=None, tf32=True, torch_compile=False, torch_compile_backend=None, torch_compile_mode=None, torchdynamo=None, tpu_metrics_debug=False, tpu_num_cores=None, use_cpu=False, use_ipex=False, use_legacy_prediction_loop=False, use_mps_device=False, warmup_ratio=0.01, warmup_steps=0, weight_decay=0.1, )TrainingArguments( _n_gpu=1, accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True}, adafactor=False, adam_beta1=0.9, adam_beta2=0.95, adam_epsilon=1e-08, auto_find_batch_size=False, bf16=True, bf16_full_eval=False, cache_dir=None, data_seed=None, dataloader_drop_last=False, dataloader_num_workers=0, dataloader_persistent_workers=False, dataloader_pin_memory=True, dataloader_prefetch_factor=None, ddp_backend=None, ddp_broadcast_buffers=None, ddp_bucket_cap_mb=None, ddp_find_unused_parameters=None, ddp_timeout=1800, debug=[], deepspeed=./finetune/ds_config_zero2.json, disable_tqdm=False, dispatch_batches=None, do_eval=False, do_predict=False, do_train=False, eval_accumulation_steps=None, eval_delay=0, eval_steps=None, evaluation_strategy=no, fp16=False, fp16_backend=auto, fp16_full_eval=False, fp16_opt_level=O1, fsdp=[], fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}, fsdp_min_num_params=0, fsdp_transformer_layer_cls_to_wrap=None, full_determinism=False, gradient_accumulation_steps=1, gradient_checkpointing=True, gradient_checkpointing_kwargs=None, greater_is_better=None, group_by_length=False, half_precision_backend=auto, hub_always_push=False, hub_model_id=None, hub_private_repo=False, hub_strategy=every_save, hub_token=, ignore_data_skip=False, include_inputs_for_metrics=False, include_num_input_tokens_seen=False, include_tokens_per_second=False, jit_mode_eval=False, label_names=None, label_smoothing_factor=0.0, learning_rate=5e-06, length_column_name=length, load_best_model_at_end=False, local_rank=0, log_level=passive, log_level_replica=warning, log_on_each_node=False, logging_dir=./outputs/runs/Apr29_11-42-07_anming, logging_first_step=False, logging_nan_inf_filter=True, logging_steps=1.0, logging_strategy=steps, lr_scheduler_kwargs={}, lr_scheduler_type=cosine, max_grad_norm=1.0, max_steps=-1, metric_for_best_model=None, model_max_length=4096, mp_parameters=, neftune_noise_alpha=None, no_cuda=False, num_train_epochs=3.0, optim=adamw_torch, optim_args=None, output_dir=./outputs/, overwrite_output_dir=False, past_index=-1, per_device_eval_batch_size=8, per_device_train_batch_size=16, prediction_loss_only=False, push_to_hub=False, push_to_hub_model_id=None, push_to_hub_organization=None, push_to_hub_token=, ray_scope=last, remove_unused_columns=True, report_to=['tensorboard'], resume_from_checkpoint=None, run_name=./outputs/, save_on_each_node=False, save_only_model=False, save_safetensors=True, save_steps=200, save_strategy=steps, save_total_limit=None, seed=1024, skip_memory_metrics=True, split_batches=None, tf32=True, torch_compile=False, torch_compile_backend=None, torch_compile_mode=None, torchdynamo=None, tpu_metrics_debug=False, tpu_num_cores=None, use_cpu=False, use_ipex=False, use_legacy_prediction_loop=False, use_mps_device=False, warmup_ratio=0.01, warmup_steps=0, weight_decay=0.1, )

Qihoo360 / 360zhinao

运行微调时报错“torch.cuda.OutOfMemoryError: CUDA out of memory. ” #11