inference_utils.sh

# Run this script from the root directory as
# bash scripts/inference_utils.sh

export HF_HOME=../cache/
 model_name="llava-hf/llava-v1.6-mistral-7b-hf"
 model_name="Lin-Chen/ShareGPT4V-7B"
model_name="renjiepi/G-LLaVA-7B"
 model_name="llava-hf/llava-1.5-13b-hf"
 model_name="llava-hf/llava-v1.6-vicuna-13b-hf"
 model_name="Lin-Chen/ShareGPT4V-13B"
model_name="Qwen/Qwen2-VL-2B-Instruct"

# This script aims to carry out manual distributed data inference
# using multi GPUs by splitting the data manually across GPUs
# based on operator preference.
 export CUDA_VISIBLE_DEVICES=4
 python scripts/inference_utils.py \
 -dp "./datastore/" \
 -mn "$model_name" \
 -drs 0 \
 -dre 30 \
 --mode "inference"

#export CUDA_VISIBLE_DEVICES=5
# python scripts/inference_utils.py \
# -dp "./datastore/" \
# -mn "$model_name" \
# -drs 30 \
# -dre 60 \
# --mode "inference"

# export CUDA_VISIBLE_DEVICES=6
# python scripts/inference_utils.py \
# -dp "./datastore/" \
# -mn "$model_name" \
# -drs 60 \
# -dre 90 \
# --mode "inference"

# export CUDA_VISIBLE_DEVICES=7
# python scripts/inference_utils.py \
# -dp "./datastore/" \
# -mn "$model_name" \
# -drs 90 \
# -dre 131 \
# --mode "inference"