InternLM / InternLM/InternLM-Math

critic model on multiple goals

未关闭
#46 2 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看
主要语言
Python
星标
550
派生
39
PR 合并指标
30 天内没有已合并 PR

描述

Hi,

I am trying to reproduce the proof search using the critic model. May I check If the current state has multiple goals, then how do we use the critic model evaluate it?
1. just join all the goals together as a single state (prompt) and send it into the critic model to get a single score.
2. for each goal get a score from the critic model individually and add them together.
3. just evaluate the current goal we are trying to solve.

Many thanks for your help in advance!

efessas

贡献指南

这个仓库没有索引到贡献指南

调研方向

No files, tests, or entry points are named. First confirm how the proof-search implementation represents multiple goals and whether the critic model behavior is documented elsewhere. Done means recording a maintainer-confirmed rule for evaluating multi-goal states, rather than choosing among the three options without project context.

由索引模型根据 Issue 内容生成。

评估

技术栈
python
领域
machine-learning
Issue 类型
文档
难度
5/5
预计耗时
一周以上
活跃度
停滞
描述清晰度
需要澄清
新手友好度
20/100

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。