ThreadWeaver: Adaptive Threading for Efficient Parallel Reasoning in Language Models
Abstract
ThreadWeaver, a framework for adaptive parallel reasoning, achieves accuracy comparable to sequential models while reducing inference latency through parallel trajectory generation, trie-based training-inference co-design, and parallelization-aware reinforcement learning.
Scaling inference-time computation has enabled Large Language Models (LLMs) to achieve strong reasoning performance, but inherently sequential decoding leads to substantial latency, especially on complex tasks. Recent work on adaptive parallel reasoning aims to improve inference efficiency by decomposing the problem-solving process into concurrent reasoning threads when beneficial. However, existing methods on realistic tasks are either limited to supervised behavior cloning or exhibit significant accuracy drops compared to widely-used sequential long chain-of-thought (CoT) baselines. Moreover, many require customized inference engines, complicating deployment. We introduce ThreadWeaver, a framework for adaptive parallel reasoning that achieves accuracy on par with popular sequential reasoning models of comparable size while significantly reducing inference latency. ThreadWeaver's performance stems from three key innovations: 1) a two-stage parallel trajectory generator that produces large-scale, high-quality CoT data with parallel annotations for supervised fine-tuning; 2) a trie-based training-inference co-design that enables parallel reasoning on any off-the-shelf autoregressive inference engine without modifying position embeddings or KV caches; and 3) a parallelization-aware reinforcement learning framework that teaches the model to balance accuracy with effective parallelization. Across six challenging mathematical reasoning benchmarks, ThreadWeaver trained atop Qwen3-8B achieves accuracy comparable to cutting-edge sequential reasoning models (71.9% on average and 79.9% on AIME24) while delivering up to 1.53x average speedup in token latency, establishing a new Pareto frontier between accuracy and efficiency.
Community
Thank you AK for sharing our new research from Meta, UC Berkeley, and UCSF!
Within the realm of parallel reasoning that is getting more and more popular, ThreadWeaver 🧵⚡️ is the first adaptive reasoning approach to:
🧠 Match strong sequential reasoning LLM accuracy on real competition math problems,
⚡ Run in parallel with no engine modifications, and
📈 Perform parallel GRPO on an off-the-shelf long CoT reasoning model to further improve performance and efficiency
Project Page: https://threadweaver-parallel.github.io/
Post on X for a one-minute preview: https://x.com/LongTonyLian/status/1995561005557186963?s=20
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper



