scripts/v1_5/eval/llava_raw_200k/mmbench_cn.sh

#!/bin/bash

SPLIT="mmbench_dev_cn_20231003"

python -m llava.eval.model_vqa_mmbench \
    --model-path /mnt/petrelfs/zhuchenglin/LLaVA/checkpoints_ft/llava_raw_200k/llava-v1.5-13b-pretrain \
    --model-base lmsys/vicuna-13b-v1.5 \
    --question-file ./playground/data/eval/mmbench_cn/$SPLIT.tsv \
    --answers-file ./playground/data/eval/mmbench_cn/answers/$SPLIT/llava-v1.5-13b_llava_raw_200k.jsonl \
    --lang cn \
    --single-pred-prompt \
    --temperature 0 \
    --conv-mode vicuna_v1

mkdir -p playground/data/eval/mmbench_cn/answers_upload/${SPLIT}_llava_raw_200k

python scripts/convert_mmbench_for_submission.py \
    --annotation-file ./playground/data/eval/mmbench_cn/$SPLIT.tsv \
    --result-dir ./playground/data/eval/mmbench_cn/answers/$SPLIT \
    --upload-dir ./playground/data/eval/mmbench_cn/answers_upload/${SPLIT}_llava_raw_200k \
    --experiment llava-v1.5-13b_llava_raw_200k