Qwen-2.5-math-7B

The 'Qwen2.5-Math-7B' dataset/benchmark contains prompts designed for evaluating reinforcement learning with verifiable rewards (RLVR) in deterministic outcome reasoning tasks.

Papers using Qwen-2.5-math-7B (4)

Beyond Variance: Prompt-Efficient RLVR via Rare-Event Amplification and Bidirectional Pairing2026DCPO: Dynamic Clipping Policy Optimization2025Incentivizing LLMs to Self-Verify Their Answers2025On the optimization dynamics of RLVR: Gradient gap and step size thresholds2025