search

luohongyin / EntST

Entailment self-training
MIT License
26 stars 4 forks source link

bash prompt_script.sh #2

Open masterj7874 opened 11 months ago

masterj7874 commented 11 months ago

No such file or directory: 'log/confidence/rte_ent_deberta_base.json'

luohongyin commented 11 months ago

This could be solved by mkdir confidence in the EntST/ folder.

masterj7874 commented 11 months ago

It worked when I added "/log/confidence". However, I ran into another issue: usage: python prompt_cst.py [-h] [--domain DOMAIN] [--train-model TRAIN_MODEL] [--train-size TRAIN_SIZE] [--ft-mode FT_MODE] [--exp-id EXP_ID] [--eval-mode EVAL_MODE] [--model-type-str MODEL_TYPE_STR] python prompt_cst.py: error: argument --train-model: expected one argument

luohongyin commented 11 months ago

Could you try deleting line 49 --train-mode $train_mode \ from the prompt_script.sh file?

masterj7874 commented 11 months ago

I deleted both --train-mode $train_mode \ and --train-model $train_model, then it worked. However, I ran into what it seems to be CUDA-related error as the following: RuntimeError: CUDA error: device-side assert triggered Compile with TORCH_USE_CUDA_DSA to enable device-side assertions. Thank you for your support in advance.

luohongyin commented 11 months ago

Hi, I think this might because of your torch installation. Please check with torch.cuda.is_available()

masterj7874 commented 11 months ago

torch.cuda.is_available() returns True. I believe torch installation is properly done on my end. Could it be something else?

luohongyin commented 11 months ago

Sorry for the delayed response. What is the version of your pytorch?

masterj7874 commented 11 months ago

It's 2.0.1.

luohongyin commented 11 months ago

Thanks, let me test pytorch 2 and get back to you!