Skip to content

Latest commit

 

History

History
45 lines (38 loc) · 1.24 KB

README.md

File metadata and controls

45 lines (38 loc) · 1.24 KB
Raw

Command

For end-to-end models, run the following commands for corresponding models:

  • CogVLM
python run_eval_cogvlm.py --from_pretrained cogvlm-chat  --version chat --english --bf16 \
    --query PATH/TO/C-VQA-Real_questions.csv \
    --type cogvlm-chat
  • LAVIS
python run_eval_lavis.py \
    --model-name blip2_t5 \
    --model-type pretrain_flant5xxl \
    --query PATH/TO/C-VQA-Real_questions.csv \
    --type blip2

You can also use -model-name blip2_vicuna_instruct --model-type vicuna7b, -model-name blip2_vicuna_instruct --model-type vicuna13b, -model-name blip2_t5_instruct --model-type flant5xxl for other models.

  • LLaVA
python run_eval_llava.py \
    --query PATH/TO/C-VQA-Real_questions.csv \
    --model-path liuhaotian/llava-v1.5-7b \
    --type llava_v15_7b

You can also use --model-path liuhaotian/llava-v1.5-13b, --model-path PATH/TO/LLaVA-7B-v0, --model-path PATH/TO/LLaVA-7B-v1-1 for other models.

  • MiniGPT-v2
python run_eval_minigpt4.py \
    --query PATH/TO/C-VQA-Real_questions.csv \
    --cfg_path eval_configs/minigptv2_eval.yaml \
    --type minigptv2
  • Qwen-VL
python run_eval_qwen.py \
    --query PATH/TO/C-VQA-Real_questions.csv \
    --type qwen