Skip to content

Conversation

@kwonseokki
Copy link

@kwonseokki kwonseokki commented Apr 24, 2025

안녕하세요 1주차 과제 제출입니다.

벤치마크를 처음해봐서 개념적으로 이렇게 하는것이 맞나? 라는 생각이 많이 들었습니다. 기탄없는 피드백 부탁드립니다!

@kwonseokki kwonseokki requested a review from sigridjineth April 24, 2025 09:48
@kwonseokki kwonseokki self-assigned this Apr 24, 2025
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

세부 계획을 평가한다는 것이 좀 모호한 것 같아요.

  • 예를 들면 주어진 시간에 따라 수행할 수 있는 과업인지를 측정 가능한가? 와 같은 질문이 있으면 좋겠습니다.
  • 실제로 점수를 매기는 것이니 만큼 LLM은 호출 횟수에 따라 점수가 달라질 수도 있습니다. 구체적인 Few-Shot (채점예시 기준) 을 주는 것이 중요합니다.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

이 정도 점수차이면 거의 의미가 없는 차이 정도로 보이네요.
뭔가 평가기준이 아쉬운 것 같습니다

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

넵 맞습니다. 모델 명이 계속 바뀌고 또 이게 날짜별로 모델 업데이트가 되기 때문에 분명하게 찾아보는 것이 중요합니다.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

네 맞습니다! Structured Decoding이라고 하는데 이렇게 쓰면 보통은 답변 퀄리티 즉 모델의 지능이 좀 떨어지기는 하는데 안정적인 사용이 가능합니다.

Copy link
Collaborator

@sigridjineth sigridjineth left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

너무 훌륭하십니다!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants