MiniCPM-V/eval_mm/vqaeval/shell/run_inference.sh

export CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
python -m torch.distributed.launch \
    --nproc_per_node=${NPROC_PER_NODE:-8} \
    --nnodes=${WORLD_SIZE:-1} \
    --node_rank=${RANK:-0} \
    --master_addr=${MASTER_ADDR:-127.0.0.1} \
    --master_port=${MASTER_PORT:-12345} \
    ./eval.py \
    --model_name minicpmv26 \
    --model_path \
    --generate_method interleave \
    --eval_textVQA \
    --eval_docVQA \
    --answer_path ./answers \
    --batchsize 1