issue-auto-labeler

Issue Auto-Labeler

Multi-label text classification with a fine-tuned RoBERTa-base.
Predicts labels like bug, enhancement, question, etc. from issue titles + descriptions.

Demo

Try the live demo here: Issue Auto-Labeler Demo
Enter an issue title/body and get predicted labels with confidence scores.

Model

• Base: RoBERTa-base
• Fine-tuned on an internal dataset of issue/task texts
• Multi-label (independent probabilities per label, not softmax)
• Uses tuned thresholds from thresholds.json for improved precision/recall

Files

├─ app.py                 # Gradio UI
├─ requirements.txt       # Inference dependencies
└─ models/
   └─ NEWBEST/            # Fine-tuned model
      ├─ config.json
      ├─ special_tokens_map.json
      ├─ tokenizer.json / vocab.json + merges.txt
      ├─ thresholds.json
      └─ pytorch_model.bin

License

Code: MIT License (open for reuse).