MLLM Finetuning Demo

MLLM-Finetuning-Demo

Instal Llama-Factory

git clone https://github.com/hiyouga/LLaMA-Factory.git
cd LLaMA-Factory
pip install -e .[torch,metrics]
cd .. # 回到项目根目录

Pra-pelatihan

Penyelarasan fitur di LLAVA, Freeze language_model dan vision_tower , dan hanya fine-tune multi_modal_projector .

CUDA_VISIBLE_DEVICES=0 llamafactory-cli train config/llava_pt.yaml

Penyesuaian yang baik

CUDA_VISIBLE_DEVICES=0 llamafactory-cli train config/llava_lora_sft.yaml

Obrolan web

CUDA_VISIBLE_DEVICES=0 llamafactory-cli webchat 
--model_name_or_path llava-hf/llava-1.5-7b-hf 
--adapter_name_or_path saves/llava1_5-7b/lora/sft 
--template llava

Unggah Dataset ke Huggingface

Harap ganti kunci Anda sendiri di upload_dataset.py .

python3 upload_dataset.py

Ekspor dan unggah model ke Huggingface

Harap ganti export_hub_model_id dan hf_hub_token Anda sendiri di config/llava_lora_sft_export.yaml .

CUDA_VISIBLE_DEVICES=0 llamafactory-cli export config/llava_lora_sft_export.yaml