NVIDIA-NeMo / NVIDIA-NeMo/Anonymizer

feat(qa): tune quality QA generation and compare prompts

Open
#73 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

medium priority task
Dominant language
Python
Stars
123
Forks
17
Avg merge
4d 2h
Merged PRs (30d)
11

Description

Context

From e2e testing:

  • Quality QA questions occasionally ask multiple things in one question, conflating multiple meaning units. This makes the compare step unfairly strict.
  • Quality compare sometimes penalizes generalized answers that preserve core meaning but differ in specificity from the original QA answer.

Scope

  • Prompt-tune meaning unit extraction and/or quality QA generation to produce single-focus questions
  • Guide the quality compare prompt to treat generalized-but-correct answers as matching
  • Measure impact on utility scores across bio and legal datasets

Files

  • src/anonymizer/engine/rewrite/qa_generation.py (QA generation prompts)
  • src/anonymizer/engine/rewrite/evaluate.py (compare prompt).

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reading the prompts in src/anonymizer/engine/rewrite/qa_generation.py and the compare prompt in src/anonymizer/engine/rewrite/evaluate.py, then review the existing e2e evaluation flow. Done means quality QA questions are single-focus, generalized-but-correct answers are treated as matches, and utility-score impact is measured on the bio and legal datasets.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
ai
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
52/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.