scripts/bash/misc/gen_dpo.sh

#!/bin/bash

proj_root="/mnt/HithinkOmniSSD/user_workspace/liukai4/projects/JavisDiT-dev"
data_root="/mnt/HithinkOmniSSD/user_workspace/liukai4/datasets/JavisDiT/train"

model_path="${proj_root}/exps/audio_video/dualffn_audio_nolora_video_fulllora_r64a16_data330k/epoch001-global_step13000"

resolution=480p # 240p
num_frames=81  # 5s
aspect_ratio="9:16"

# 1-sample: 480p*5k=22h, 360p*6k=12.5h, 240p*8.2k=7h
data_mode=TAVGBench_train_fps16_dpo_${resolution}_6k_ths

prompt_path="${data_root}/video/${data_mode}.csv"

cfg_file="${proj_root}/configs/wan2.1/inference/ths/sample.py"
save_dir="/mnt/workspace/inference/dop_gen/${data_mode}"
log_path="/mnt/workspace/log/${data_mode}.txt"

rm -rf ${save_dir}

torchrun --standalone --nproc_per_node 1 \
    scripts/inference.py \
    ${cfg_file} \
    --layernorm-kernel false \
    --resolution ${resolution} --num-frames ${num_frames} --aspect-ratio ${aspect_ratio} \
    --prompt-path ${prompt_path} --model-path ${model_path} --num-sample 3 \
    --save-dir ${save_dir} --verbose 1

    # --start-index 0 --end-index 2500 \