# NCCL setting export GLOO_SOCKET_IFNAME=bond0 export NCCL_SOCKET_IFNAME=bond0 export NCCL_IB_HCA=mlx5_10:1,mlx5_11:1,mlx5_12:1,mlx5_13:1 export NCCL_IB_GID_INDEX=3 export NCCL_IB_TC=162 export NCCL_IB_TIMEOUT=22 export NCCL_PXN_DISABLE=0 export NCCL_IB_QPS_PER_CONNECTION=4 # export NCCL_ALGO=Ring export OMP_NUM_THREADS=1 export MKL_NUM_THREADS=1 export NCCL_ALGO=Tree REAL_DATASET_DIR=/storage/dataset/vae_eval/panda70m EXP_NAME=test_train SAMPLE_RATE=1 NUM_FRAMES=33 RESOLUTION=256 SUBSET_SIZE=1000 CKPT=/storage/clh/models/488dim8 torchrun \ --nnodes=1 --nproc_per_node=8 \ --rdzv_endpoint=localhost:29504 \ --master_addr=localhost \ --master_port=29600 \ scripts/rec_causalvideo_vae.py \ --batch_size 1 \ --real_video_dir ${REAL_DATASET_DIR} \ --generated_video_dir /storage/clh/gen/488dim8 \ --device cuda \ --sample_fps 24 \ --sample_rate ${SAMPLE_RATE} \ --num_frames ${NUM_FRAMES} \ --resolution ${RESOLUTION} \ --crop_size ${RESOLUTION} \ --num_workers 8 \ --ckpt ${CKPT} \ --output_origin \ --subset_size 1000\ #--change_decoder \ #--decoder_dir /remote-home1/clh/Causal-Video-VAE/results/decoder_only-lr1.00e-05-bs1-rs248-sr2-fr25/checkpoint-5000.ckpt \