DWA-KD: Dual-Space Weighting and Time-Warped Alignment for Cross-Tokenizer Knowledge Distillation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Vu, Duc Trung, Chi, Pham Khanh, Van, Dat Phi, Van, Linh Ngo, Dinh, Sang, Le, Trung
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917293998473216
author Vu, Duc Trung
Chi, Pham Khanh
Van, Dat Phi
Van, Linh Ngo
Dinh, Sang
Le, Trung
author_facet Vu, Duc Trung
Chi, Pham Khanh
Van, Dat Phi
Van, Linh Ngo
Dinh, Sang
Le, Trung
contents Knowledge Distillation (KD) has emerged as a crucial technique for compressing Large Language Models (LLMs). Although existing cross-tokenizer KD methods have made notable progress, their effectiveness remains constrained by suboptimal alignment across sequence and vocabulary levels. To address these limitations, we introduce Dual-Space Weighting and Time-Warped Alignment (DWA-KD), a novel cross-tokenizer distillation framework that enhances token-wise distillation through dual-space entropy-based weighting and achieves precise sequence-level alignment by leveraging both lexical and semantic information. At the token level, DWA-KD maps teacher representations into the student space and vice versa, performing dual-space KD via Kullback-Leibler divergence (KL). The process is modulated by dual-space weights that up-weight tokens where the student is uncertain and the teacher is confident, thereby focusing learning on informative tokens rather than treating all positions equally. At the sequence level, DWA-KD applies Soft Dynamic Time Warping (Soft-DTW) to both the embedding and final hidden-state layers, enabling robust alignment of lexical and contextual semantics between teacher and student sequences. Extensive experiments across diverse NLP benchmarks demonstrate that DWA-KD outperforms state-of-the-art KD baselines, while ablation studies confirm the complementary contributions of entropy-based token weighting and embedding and final hidden state layer Soft-DTW alignment.
format Preprint
id arxiv_https___arxiv_org_abs_2602_21669
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle DWA-KD: Dual-Space Weighting and Time-Warped Alignment for Cross-Tokenizer Knowledge Distillation
Vu, Duc Trung
Chi, Pham Khanh
Van, Dat Phi
Van, Linh Ngo
Dinh, Sang
Le, Trung
Computation and Language
Knowledge Distillation (KD) has emerged as a crucial technique for compressing Large Language Models (LLMs). Although existing cross-tokenizer KD methods have made notable progress, their effectiveness remains constrained by suboptimal alignment across sequence and vocabulary levels. To address these limitations, we introduce Dual-Space Weighting and Time-Warped Alignment (DWA-KD), a novel cross-tokenizer distillation framework that enhances token-wise distillation through dual-space entropy-based weighting and achieves precise sequence-level alignment by leveraging both lexical and semantic information. At the token level, DWA-KD maps teacher representations into the student space and vice versa, performing dual-space KD via Kullback-Leibler divergence (KL). The process is modulated by dual-space weights that up-weight tokens where the student is uncertain and the teacher is confident, thereby focusing learning on informative tokens rather than treating all positions equally. At the sequence level, DWA-KD applies Soft Dynamic Time Warping (Soft-DTW) to both the embedding and final hidden-state layers, enabling robust alignment of lexical and contextual semantics between teacher and student sequences. Extensive experiments across diverse NLP benchmarks demonstrate that DWA-KD outperforms state-of-the-art KD baselines, while ablation studies confirm the complementary contributions of entropy-based token weighting and embedding and final hidden state layer Soft-DTW alignment.
title DWA-KD: Dual-Space Weighting and Time-Warped Alignment for Cross-Tokenizer Knowledge Distillation
topic Computation and Language
url https://arxiv.org/abs/2602.21669