Skip to content

K-QA #45

@warner-benjamin

Description

@warner-benjamin

Implement verifiers env for K-QA

Implement a verifiers environment for the K-QA datasets. Use the authors’ question, system, and LLM-as-judge prompts if provided. Only create a training dataset if a training split exists.

Paper: https://arxiv.org/abs/2401.14493
Dataset / Project: https://github.com/Itaymanes/K-QA

Format & Env

  • Single-turn open-ended clinical Q&A:
  • Use vf.SingleTurnEnv + FActScore +vf.JudgeRubric()

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions