Skip to content

Instantly share code, notes, and snippets.

@zhaoyanpeng
Created August 13, 2022 19:02
Show Gist options
  • Select an option

  • Save zhaoyanpeng/3cfaddbc5c7c8c7aeb20e5410bb75930 to your computer and use it in GitHub Desktop.

Select an option

Save zhaoyanpeng/3cfaddbc5c7c8c7aeb20e5410bb75930 to your computer and use it in GitHub Desktop.
Evaluate an AT model on AT retrieval tasks
#!/usr/bin/sh
root=./
export OMP_NUM_THREADS=32
export CUDA_VISIBLE_DEVICES=$2
run_type=$1
gpu_list=(${CUDA_VISIBLE_DEVICES//,/ })
port=$(expr $RANDOM + 1000)
ngpu=${#gpu_list[@]}
mode="dp"
num_proc=8
echo "GPUs: "$CUDA_VISIBLE_DEVICES "#"$ngpu "PORT: "$port
# eval retrieval (AT) model
# bash bash/run_retrieval.sh clotho 3
eval="retrieval"
# the AT fine-tuning experiment name
model_name="test"
# the \d+.pth file has to be in $root/$model_name/
# https://storage.googleapis.com/ai2-mosaic-public/projects/vipant/model/01FM2Y9HJ896B2G6NKRXTEVXZ7/00006210.pth
model_file="00006210.pth"
echo "eval the model $model_file"
mtask="model_root=$root model_file=$model_file model_name=$model_name eval=True
verbose=False monitor=LAMonitor worker=CLAP port=$port num_gpus=$ngpu mode=$mode num_proc=$num_proc
+model/image=vit_val +model/audio=vit_val +model/text=transformer_val +model/loss=ce +optimizer=standard +running/audio=default
model.image.pre_encoder.patch_size=32 model.audio.pre_encoder.patch_size=32 model.audio.pre_encoder.stride=[16,24] running.clip_model_name=ViT-B32
running.retrieval=True
running.audio.norms=[-4.93839311,5.75751113]
running.eval_name=clotho_captions_evaluation running.eval_samples=1e9
running.prompt=\"\" running.batch_size=64 running.audio.max_len=1000
running.audio.audio_transforms=[] running.audio.fbank_transforms=[]"
# config
extra="$mtask"
python train.py +running=$run_type $extra
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment