BASIS: Batchwise Advantage Estimation from Single-Rollout Information Sharing for LLM Reasoning
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910261351284736 |
|---|---|
| author | Gong, Shijin Xu, Erhan Ye, Kai Quinzan, Francesco Livieri, Giulia Shi, Chengchun |
| author_facet | Gong, Shijin Xu, Erhan Ye, Kai Quinzan, Francesco Livieri, Giulia Shi, Chengchun |
| contents | Reinforcement learning with verifiable rewards has become a standard recipe for improving the reasoning abilities of large language models. Existing algorithms face a tradeoff between computational efficiency and sample efficiency in value estimation and policy learning. We introduce BASIS, a critic-free post-training algorithm designed to address this tradeoff. At each online training step, BASIS samples only one rollout per prompt, but leverages rich information across prompts in the entire batch to improve value function estimation. Our experiments demonstrate that BASIS reduces MSE in value function estimation by 69% compared to REINFORCE++, a representative single-rollout baseline, and achieves lower MSE with one rollout than group mean estimators with 8 rollouts. This improvement in value estimation translates to better policy optimization: using substantially less training time, BASIS achieves performance close to multi-rollout GRPO-type baselines and often outperforms single-rollout REINFORCE-type baselines. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_27293 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | BASIS: Batchwise Advantage Estimation from Single-Rollout Information Sharing for LLM Reasoning Gong, Shijin Xu, Erhan Ye, Kai Quinzan, Francesco Livieri, Giulia Shi, Chengchun Machine Learning Reinforcement learning with verifiable rewards has become a standard recipe for improving the reasoning abilities of large language models. Existing algorithms face a tradeoff between computational efficiency and sample efficiency in value estimation and policy learning. We introduce BASIS, a critic-free post-training algorithm designed to address this tradeoff. At each online training step, BASIS samples only one rollout per prompt, but leverages rich information across prompts in the entire batch to improve value function estimation. Our experiments demonstrate that BASIS reduces MSE in value function estimation by 69% compared to REINFORCE++, a representative single-rollout baseline, and achieves lower MSE with one rollout than group mean estimators with 8 rollouts. This improvement in value estimation translates to better policy optimization: using substantially less training time, BASIS achieves performance close to multi-rollout GRPO-type baselines and often outperforms single-rollout REINFORCE-type baselines. |
| title | BASIS: Batchwise Advantage Estimation from Single-Rollout Information Sharing for LLM Reasoning |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2605.27293 |