Skip to content

Qwen2.5-VL全参数训练视觉层和projector,lora微调语言模型的模型训练与导出方式? #7353

Description

@echochoc

两个问题:

  1. 我使用如下训练参数微调Qwen2.5-VL-3B-Instruct,希望全参数训练视觉层和projector,lora微调语言模型,是否正确
    llamafactory-cli train \ --stage sft \ --do_train True \ --model_name_or_path /path/to/Qwen2.5-VL-3B-Instruct \ --preprocessing_num_workers 16 \ --finetuning_type lora \ --template qwen2_vl \ --flash_attn auto \ --dataset_dir data \ --dataset my_dataset \ --cutoff_len 2048 \ --learning_rate 5e-05 \ --num_train_epochs 20.0 \ --max_samples 100000 \ --per_device_train_batch_size 2 \ --gradient_accumulation_steps 8 \ --lr_scheduler_type cosine \ --max_grad_norm 1.0 \ --logging_steps 5 \ --save_steps 100 \ --warmup_steps 0 \ --packing False \ --report_to none \ --output_dir /path/to/output/dir \ --bf16 True \ --plot_loss True \ --trust_remote_code True \ --ddp_timeout 180000000 \ --include_num_input_tokens_seen True \ --optim adamw_torch \ --lora_rank 8 \ --lora_alpha 16 \ --lora_dropout 0.05 \ --lora_target all \ --no_freeze_vision_tower \ --no_freeze_multi_modal_projector

  2. 训练好的模型如何进行参数合并?

Metadata

Metadata

Assignees

No one assigned

    Labels

    duplicateThis issue or pull request already exists

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions