[Feature Request]: Re-introduce Self-Refine
Opened this issue · 0 comments
alckasoc commented
Feature Description
There's an argument to be made that the mechanism of generating critique is somewhat agentic. Let's keep Self-Refine, but we will re-introduce it at least after https://github.com/agential-ai/agential/milestone/4
Implement:
Run:
- HotpotQA
- TriviaQA
- AmbigNQ
- GSM8k
- SVAMP
- TabMWP
- MBPP
- HumanEval
- ALFWorld
- WebShop
- AgentBench (includes ALFWorld & WebShop)