斯坦福团队发了SPIRAL，让LLM训练时就学会并行采样和聚合答案，不是只会单链思考，更符合实际推理场景。
斯坦福团队提出SPIRAL框架，通过强化学习让LLM在测试时自动协调顺序推理、并行采样和结果聚合。与传统只优化单链推理的训练方法不同，SPIRAL使用set RL训练模型生成对聚合器集体有用的多个候选答案，并用标准RL优化聚合器从这些候选合成改进答案。该方法使所有测试时计算维度（长链、并行样本、聚合）端到端可学习，缩小训练与部署的差距。
At test time, we wrap LLMs in scaffolds that scale compute every which way -- longer chains, paralle...
At test time, we wrap LLMs in scaffolds that scale compute every which way -- longer chains, parallel samples, and aggregation across them. So why do we still train them to use only one of these? Introduce Spiral: it uses set RL to teach a model to generate responses that are collectively useful for an aggregator, and standard RL to teach it to aggregate those responses into an improved answer! Jubayer Ibn Hamid @jubayer_hamid The most capable reasoning systems in AI scale inference compute along several axes: sequential compute to think longer, parallel compute to sample many independent attempts, and aggregative compute to synthesize prior traces into a new improved one. But during training, we only optimize how models use sequential compute. This creates a fundamental mismatch between how we ultimately deploy these systems and how we train them, leaving much of search and synthesis unoptimized. We introduce SPIRAL, an RL framework for making all inference-compute primitives end-to-end learnable: models learn to coordinate sequential, parallel, and aggregative reasoning using only the reward of the final output. Work with @ifdita_hasan (co-lead), @michaelyli_ , @oshaikh13 , @yoonholeee , @DorsaSadigh , @chelseabfinn , @noahdgoodman 🧵 🔗 View Quoted Tweet 💬 4 🔄 7 ❤️ 59 👀 8462 📊 14 ⚡