AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

AdaSpec: Adaptive Multilingual Speculative Decoding with Self-Synthesized Language-Aware Training and Vocabulary Simplification

Dinh-Truong Do, Nguyen-Khang Le, Le-Minh Nguyen

AAAI 2026 · Large language models › Reasoning and chain of thought · approach: Speculative decoding

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes AdaSpec, a multilingual speculative decoding method that pairs a lightweight drafter with self-synthesized language-aware training and vocabulary simplification to address static-vocabulary and language-data gaps.

Problem

Speculative decoding for LLMs works well in English but struggles in multilingual settings due to static vocabularies and a lack of language-specific instruction data.

Approach

Adapts speculative decoding to multiple languages by self-synthesizing language-aware training data and simplifying the drafter's vocabulary, rather than relying on static vocabularies and external language-specific instruction data.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train the drafter on target-language synthesized instruction data instead of English data (ablation)AdaSpec (Llama3-8B) · Multi-SpecBench Germanspeedup ratio: 1.30 -> 1.83Improved
train the drafter on target-language data (ablation)AdaSpec (Llama3-8B) · Multi-SpecBench Japanesespeedup ratio: 1.25 -> 1.91Improved
use language-specific simplified vocabulary instead of the full vocabulary (ablation)AdaSpec (Llama3-8B) · Multi-SpecBench Germanspeedup ratio: 1.58 -> 1.83Improved
use adaptive vocabulary size instead of a fixed size (ablation)AdaSpec (Llama3-8B) · Multi-SpecBench Germanspeedup ratio: 1.71 -> 1.83Improved
use language-specific vocabulary instead of full vocabulary (ablation)AdaSpec (Llama3-8B) · Multi-SpecBench Germancompression ratio tau: 2.79 -> 2.71No clear effect
replace EAGLE-2 with AdaSpec for multilingual decodingLlama3-8B-Instruct · Multi-SpecBench Germanspeedup ratio: 1.15 -> 1.83Improved
increase adaptive window size from 5 to 15 (u=1, z=0) (ablation)AdaSpec · Multi-SpecBench multilingual mixthroughput (tokens/s): 135.22 -> 136.70Improved
add larger step size (u=3) and offset (z=1) (ablation)AdaSpec (w=10) · Multi-SpecBench multilingual mixthroughput (tokens/s): 136.30 -> 138.30Improved
increase window to 15 at u=5, z=2 (ablation)AdaSpec · Multi-SpecBench multilingual mixthroughput (tokens/s): 139.93 -> 138.91Hurt
use EAGLE-2 and FR-Spec without language adaptation on small modelLlama3.2-1B-Instruct · Multi-SpecBench multilingualspeedup vs autoregressiveHurt
implement in C/CUDA instead of Pythonspeculative decoding methods · Multi-SpecBench monolingualspeedupImproved

Experimental setup

Builds on: speculative decoding

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy