PaCoRe：透過平行協調推理學習擴展測試時計算規模

摘要

我們提出平行協同推理（PaCoRe），這是一個專為突破當代語言模型核心限制而設計的訓練與推論框架：現有模型無法在固定上下文窗口下，將測試時計算量（TTC）擴展至遠超序列推理的規模。PaCoRe 突破傳統序列範式，透過多輪訊息傳遞架構協調大規模平行探索來驅動 TTC。每輪啟動多條平行推理軌跡，將其發現壓縮為上下文受限的訊息，並整合這些訊息以指導下一輪推理，最終生成答案。透過大規模基於結果的強化學習進行端到端訓練，模型掌握了 PaCoRe 所需的整合能力，能將有效 TTC 擴展至數百萬詞元規模而不突破上下文限制。該方法在多領域實現顯著提升，尤其在數學推理上超越前沿系統：一個 80 億參數模型在 HMMT 2025 競賽中達到 94.5% 準確率，透過將有效 TTC 擴展至約兩百萬詞元，超越了 GPT-5 的 93.2%。我們開源了模型檢查點、訓練資料與完整推論流程，以加速後續研究。

English

We introduce Parallel Coordinated Reasoning (PaCoRe), a training-and-inference framework designed to overcome a central limitation of contemporary language models: their inability to scale test-time compute (TTC) far beyond sequential reasoning under a fixed context window. PaCoRe departs from the traditional sequential paradigm by driving TTC through massive parallel exploration coordinated via a message-passing architecture in multiple rounds. Each round launches many parallel reasoning trajectories, compacts their findings into context-bounded messages, and synthesizes these messages to guide the next round and ultimately produce the final answer. Trained end-to-end with large-scale, outcome-based reinforcement learning, the model masters the synthesis abilities required by PaCoRe and scales to multi-million-token effective TTC without exceeding context limits. The approach yields strong improvements across diverse domains, and notably pushes reasoning beyond frontier systems in mathematics: an 8B model reaches 94.5% on HMMT 2025, surpassing GPT-5's 93.2% by scaling effective TTC to roughly two million tokens. We open-source model checkpoints, training data, and the full inference pipeline to accelerate follow-up work.

PaCoRe：透過平行協調推理學習擴展測試時計算規模

PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning

摘要

Support