Go to ICLR 2026 Conference homepageBridging Draft Policy Misalignment: Group Tree Optimization for Speculative Decoding
Keywords: Speculative Decoding; Draft Tree Reward; Tree Optimization
TL;DR: We introduce GTO, which optimizes draft-tree reward to provably increase acceptance length and achieve >7% faster speculative decoding than EAGLE-3, while fine-tuning existing draft models.
Abstract: Speculative decoding accelerates large language model (LLM) inference by letting a lightweight draft model propose multiple tokens that the target model verifies in parallel. Yet existing training objectives optimize only a single greedy draft path, while decoding follows a tree policy that re-ranks and verifies multiple branches. This draft policy misalignment limits achievable speedups.
We introduce **Group Tree Optimization** (GTO), which aligns training with the decoding-time tree policy through two components: (i) Draft Tree Reward, a sampling-free objective equal to the expected acceptance length of the draft tree under the target model, directly measuring decoding performance; (ii) Group-based Draft Policy Training, a stable optimization scheme that contrasts trees from the current and a frozen reference draft model, forming debiased group-standardized advantages and applying a PPO-style surrogate along the longest accepted sequence for robust updates. We further prove that increasing our Draft Tree Reward provably improves acceptance length and speedup.
Across dialogue (MT-Bench), code (HumanEval), and math (GSM8K), and multiple LLMs (e.g., LLaMA-3.1-8B, LLaMA-3.3-70B, Vicuna-1.3-13B, DeepSeek-R1-Distill-LLaMA-8B, Qwen3-8B), GTO increases acceptance length by \(7.4\%\) and yields an additional \(7.7\%\) speedup over prior state-of-the-art EAGLE-3. By bridging draft policy misalignment, GTO offers a practical, general solution for efficient LLM inference. Code and draft models are available at https://github.com/hsj576/GTO.
Primary Area: foundation or frontier models, including LLMs
Submission Number: 3942
Loading