Spaces:
Running
on
A10G
Running
on
A10G
CKPT_NAME="Video-LLaVA-7B" | |
CKPT="checkpoints/${CKPT_NAME}" | |
EVAL="eval" | |
python3 -m llava.eval.model_vqa \ | |
--model-path ${CKPT} \ | |
--question-file ${EVAL}/mm-vet/llava-mm-vet.jsonl \ | |
--image-folder ${EVAL}/mm-vet/images \ | |
--answers-file ${EVAL}/mm-vet/answers/${CKPT_NAME}.jsonl \ | |
--temperature 0 \ | |
--conv-mode vicuna_v1 | |
mkdir -p ${EVAL}/mm-vet/results | |
python3 scripts/convert_mmvet_for_eval.py \ | |
--src ${EVAL}/mm-vet/answers/${CKPT_NAME}.jsonl \ | |
--dst ${EVAL}/mm-vet/results/${CKPT_NAME}.json | |
python3 scripts/eval_gpt_mmvet.py \ | |
--mmvet_path ${EVAL}/mm-vet \ | |
--ckpt_name ${CKPT_NAME} \ | |
--result_path ${EVAL}/mm-vet/results | |