Please revise japanese clip's ReadMe and the position of model.eval in cli.py

[jackee777]

I wonder if one intorduction and one code may be something wrong.

At first, japanese clip's ReadMe is not correct.
The exiting example is following.

python3 clip_benchmark/cli.py \
  --model_type "ja_clip" \ # flag to use japanese-clip
  --pretrained "rinna/japanese-cloob-vit-b-16" \ # now, we have `rinna/japanese-cloob-vit-b-16` or `rinna/japanese-clip-vit-b-16`. 
  --language "jp" \
  --task "zeroshot_classification"  \
  --dataset "imagenet1k" 
  --dataset_root {ROOT_PATH} 

However, it lacks eval/build and "\".
Maybe, the following is correct. In addition to this, "#" or unnecessary spaces prevent executing this shell script.
Please check it.

python3 clip_benchmark/cli.py eval \
  --model_type "ja_clip" \
  --pretrained "rinna/japanese-cloob-vit-b-16" \
  --language "jp" \
  --task "zeroshot_classification"  \
  --dataset "imagenet1k" \
  --dataset_root {ROOT_PATH} 

Secondly, model.eval() in line. 181 of cli.py seems like unnecesary.
model.eval() has alerady been executed in load_clip() function.

I hope you to review them.

[mehdidc]

Thanks @jackee777! you are right, will update the example. On the other hand not sure to see where model.eval() is called in load_clip. I checked model.training after load_clip, and it is still True, so mode.eval() would still be necessary (it is needed for models using dropout and/or batchnorm).

[jackee777]

@mehdidc
Thank you for your response. You are correct, and I'm sorry for misunderstanding the 2nd question.
I may have seen load_clip in another version of this repository. The current version works successfully.

I close this comment.