Semantic Code Search using GPT2 small model as Code and Query Encoder

Based on CodeSearchNet Challenge code

Train a Model

Steps to run with Singularity on TACC

1. If running on TACC clusters, load the Singularity module

   module load tacc-singularity

2. Clone dir in the work directory of TACC

    cd $WORK
    git clone https://github.com/ligerfotis/semantic-code-search_gpt2_tf.git
    cd semantic-code-search_gpt2_tf
   

3. To download dataset (only the first time), run from root directory

    script/setup
   

4. Start the container by running

    script/console
   

5. Train on part of the Python Dataset

    cd src/
    
    train.py --model gpt2 --testrun ../resources/saved_models/ ../resources/data/python/final/jsonl/train/ ../resources/data/python/final/jsonl/valid/ ../resources/data/python/final/jsonl/test
   

6. Alternatively one can sent the training as a job on TACC after step 3.

    sbatch  run_experiments_gtp2_p100.sh 
   

Docker images:

• preprocessing: ligerfotis/preprocessing:latest
• main image with gpu support: ligerfotis/csnet:gpu