jobLLMatch creates a fine-tuned BART model that provides feedback to candidates based on their resume and a job description
python -m venv venv
source venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txtTo generate a score and feedback report per (job description, resume) pair use the palm_generate_reports.ipynb notebook. It was run in Google Colab since in my experience the inference time was significantly shorter as compared to local or Google Cloud implementation.
Refer to finetune_BART.py
Now that we have a fine-tuned model we can go to inference_BART.py and generate scores and reports for the test set (15% of the total data).
You can run inference by:
python inference_BART.py --checkpoint <path to model checkpoint>there are also some optional arguments suchs as:
--test_mode: if present, then it uses--test_mode_size <int>to slice the test dataset so that it runs an experimental run on a smaller part of the dataset--csv: if present the it exports the test dataset together with the generated data as csv, default is.pkl--calculate_loss: if present, it calculates the model loss of the test dataset (not present during training)
Refer to metrics_notebook.ipynb
- Occasionally ChatGPT was used to troubleshoot code bugs.