search

microsoft / CodeXGLUE

CodeXGLUE
MIT License
1.5k stars 363 forks source link

Can UniXcoder do Defect detection? #186

Closed LO-FANG closed 2 months ago

LO-FANG commented 2 months ago

Thanks for this repo, thanks for your work. I wonder if you can help me answer some questions.

  1. This is my run script ："python run.py --output_dir=./saved_models --model_type=roberta --tokenizer_name=microsoft/unixcoder-base --model_name_or_path=microsoft/unixcoder-base --do_train --train_data_file=.. /dataset/dataset/train.jsonl --eval_data_file=.. /dataset/dataset/valid.jsonl --test_data_file=.. /dataset/dataset/test.jsonl --epoch 5 --block_size 400 --train_batch_size 32 --eval_batch_size 64 --learning_rate 2e-5 --max_grad_norm 1.0 --evaluate_during_training " I left the code unchanged, does that mean I used unixcoder?`
    1. With the above script, I successfully ran the project，but why can I fine-tune unixcoder when there is no code comment in my dataset and no code converted to AST?