Knowledge-to-Verification: Unlocking Reinforcement Learning with Verifiable Rewards for LLMs in Knowledge-Intensive Domains

What is K2V

K2V (Knowledge-to-Verification) is a framework that extends RLVR (Reinforcement learning ith verifiable Rewards) to knowledge-intensive domains and enabling verification of the model's reasoning process, without any human supervision.

How to use

1. Clone the repository

   git clone --recurse-submodules https://github.com/superfarther/K2V.git
   cd K2V

2. Install the dependencies of graphgen-mask according to the README, and then synthesize the fill-blank style QA pairs.

   cd graphgen-mask
   vim README.md 

3. Synthsize question-specific checklist for each QA pair.

   cd utils
   vim README.md 

4. Install the dependencies of verl according to the README, and then start training.

   cd verl
   vim README.md