Nghi Dao

Graduate student at Columbia University.

[Nghi Dao @NghiMDao]

Check out our new paper. We let language models retry failed attempts during RLVR, dynamically expanding the group size. The result: better first-attempt accuracy than GRPO across Qwen3 0.6B–4B on 8 math & science benchmarks. https://t.co/F9TsZcL9OC

[Adam Block @adam_block66985]

🚨New paper led by my amazing master's student @NghiMDao! 🚨
Introducing Retry-Aware Policy Optimization (ReAPO). Conditioning retries on failures improves RLVR exploration: ReAPO improves accuracy and training efficiency relative to GRPO.

1/12