train_tc_clip_zero_shot.sh

# zero-shot k400 training with 4 V100 gpus
export CUDA_VISIBLE_DEVICES=0,1,2,3
export GPUS_PER_NODE=4

protocol=zero_shot
dataset_name=k400
data=${protocol}_${dataset_name}

expr_name=tc_clip_reproduce
trainer=tc_clip
use_wandb=true

torchrun --nproc_per_node=${GPUS_PER_NODE} main.py \
data=${data} \
output=workspace/expr/${data}/${expr_name}/${data}_${expr_name}_${trainer} \
trainer=${trainer} \
use_wandb=${use_wandb}