awesome-RLVR-boundary: Resource Collection for Reinforcement Learning with Verifiable Rewards (RLVR) and LLM Reasoning Boundaries

This project compiles selected resources on Reinforcement Learning with Verifiable Rewards (RLVR) and the reasoning capability boundaries of large language models (LLMs), providing researchers with a systematic learning reference.

awesome-RLVR-boundary is a carefully curated resource collection focusing on two cutting-edge research directions:

1. Reinforcement Learning with Verifiable Rewards (RLVR)
2. LLM Reasoning Capability Boundaries

RLVR (Reinforcement Learning with Verifiable Rewards) is a reinforcement learning paradigm where reward signals are verifiable, rather than relying on human preferences or subjective judgments. This is particularly important in tasks such as mathematical reasoning and code generation.

With the emergence of reasoning models like DeepSeek-R1 and OpenAI o1, understanding the reasoning capability boundaries of LLMs has become crucial:

• Which tasks can be reliably solved?
• Where are the model's limitations?
• How to further improve reasoning capabilities?

This project provides researchers with:

• Systematic literature compilation
• Links to key papers and code
• An overview of the field's development path

• Reinforcement learning researchers
• LLM reasoning capability researchers
• AI alignment and safety researchers

• GitHub Repository: https://github.com/rorofaiz/awesome-RLVR-boundary