Learning from the Gap Between Pass@K and Pass@1

תקציר מקורי באנגליתarXiv:2609.35793v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly trained with reinforcement learning from verifiable rewards (RLVR). An exact verifier can also support test-time scaling by selecting a passing response from multiple samples, while other deployments use beam search, adaptive sampling, or tools. We study single-sample decoding, where each query receives one response without search, to ask whether search-exposed behavior can be absorbed into the model. Existing verified-response post-training recipes do not generally distinguish problems already solved on the first decode from failures recovered within K samples. Under a fixed budget, this can spend examples repeating behavior the deployed policy already has. We introduce GapFT, which selects train