Enhance DeepSeek Harness with LLM-as-a-Verifier Plugin
The LLM verifier plugin for DeepSeek Harness grades candidate solutions and returns the best one.
A new plugin for DeepSeek Harness introduces an LLM verifier that grades candidate solutions with a model, returning scores between 0 and 1. The main feature, verify_rollout, allows users to request something once, prompting the plugin to run several independent agent attempts in parallel and provide the best result.
This synthesis was produced from its source by AI; there is no human editor or manual review step. How we work