VILA/scripts/v1_5/eval/docvqa.sh

#!/bin/bash
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
gpu_list="${CUDA_VISIBLE_DEVICES:-0}"
IFS=',' read -ra GPULIST <<< "$gpu_list"

CHUNKS=${#GPULIST[@]}

MODEL_PATH=$1
CKPT=$2
CONV_MODE=vicuna_v1
if [ "$#" -ge 3 ]; then
    CONV_MODE="$3"
fi


for IDX in $(seq 0 $((CHUNKS-1))); do
    CUDA_VISIBLE_DEVICES=${GPULIST[$IDX]} python -m llava.eval.evaluate_vqa \
        --model-path $MODEL_PATH \
        --generation-config '{"max_new_tokens": 100}' \
        --dataset docvqa_test \
        --image-folder ./playground/data/eval/docvqa \
        --data-path ./playground/data/eval/docvqa/test.jsonl \
        --answers-file ./playground/data/eval/docvqa/answers/$CKPT/${CHUNKS}_${IDX}.jsonl \
        --num-chunks $CHUNKS \
        --chunk-idx $IDX \
        --conv-mode $CONV_MODE
done
wait