NousResearch/atropos
1
0
Fork 0
mirror of https://github.com/NousResearch/atropos.git synced 2026-04-29 17:35:07 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Update README.md

Browse source
This commit is contained in:
Merkel Tranjes 2025-06-23 16:23:02 +02:00 • committed by GitHub
parent 8bf0312b8a
commit af1c98d7a8
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
environments/community/README.md
View file

@ -293,7 +293,7 @@ A unique environment for training LLMs to express needs and desires through auth
**Author**: [JakeBoggs](https://github.com/JakeBoggs)
**Purpose**: Train LLMs to generate humorous punchlines using Verifiable Rewards via Completion Likelihood Improvement (VR-CLI)
A specialized environment for training LLMs to understand humor by generating joke punchlines through a novel RL technique from the paper "Learning to Reason for Long-Form Story Generation" (Gurning & Lapata, 2025). The environment teaches models to first generate reasoning that leads to good punchlines, with rewards based on how much the reasoning improves the likelihood of the actual punchline.
A specialized environment for training LLMs to understand humor by generating joke punchlines through a novel RL technique from the paper "Learning to Reason for Long-Form Story Generation" (Gurung & Lapata, 2025). The environment teaches models to first generate reasoning that leads to good punchlines, with rewards based on how much the reasoning improves the likelihood of the actual punchline.
**Features**:
- **VR-CLI Methodology**: Uses Verifiable Rewards via Completion Likelihood Improvement for reduced overfitting