ASD: training-free approximate speculative decoding boosts LLM throughput up to 15.26%
新智元 · wechat · 2026-09-05
A joint team from Beihang, Tsinghua, HKU and Peking University proposes Approximate Speculative Decoding (ASD), which relaxes the 'stop at first token mismatch' rule: it selectively accepts low-regret mismatches within a strict request-level budget and reuses the already-computed greedy suffix behind them.
Results:
- +7.78% average throughput on Qwen3-14B + DSpark-14B across 7 tasks (up to +11.73%);
- Positive gains in all 10 setups across DSpark/EAGLE3/Medusa drafters (3.05%–15.26%, avg 7.52%);
- +10–16% verifier acceptance rate on the 284B DeepSeek-V4-Flash (8x H20), with accuracy drift under 0.30pt;
- Plugin verifier costs only 0.045–0.083ms per token, no training/fine-tuning/extra forward passes; budget=0 falls back to strict decoding.
The paper is explicit that accepting non-greedy tokens changes the trajectory — not lossless — but audits show accuracy holds on most tasks, arguing speed/behavior tradeoffs should be explicitly modeled and auditable.
More from Infra
- llama.cpp team launches Llama: one-click local AI on your Mac menu bar — Hugging Face · 2026-10-23
- China's daily token use hits 500T, up 4x as average price doubles to ¥13.46 — teortaxesTex · 2026-10-23
- New all-reduce kernels deliver 12% decode speedup for DeepSeek-V4.1 on PCIe-connected GPUs — AlpinDale · 2026-10-23
- Offshore oil rigs could be converted into AI data centers powered by stranded gas — clemnt · 2026-10-23
- Same native UI, two stacks: Swift build is ~80MB Mac-only, Zig build is ~10MB cross-platform — jedisct1 · 2026-10-23
- SurfSense: an open-source local NotebookLM with MCP agents and a prompt-caching lesson — FurtiveMirth · 2026-10-23