paper-with-me

Papers

Enhancing Cross-Tokenizer Knowledge Distillation with Contextual Dynamical Mapping

2025-02-16 · Yijie Chen, Yijin Liu, Fandong Meng, Yufeng Chen, Jinan Xu, Jie zhou

Knowledge Distillation (KD) has emerged as a prominent technique for model compression. However, conventional KD approaches primarily focus on homogeneous architectures with identical tokenizers, constraining their applicability in cross-architecture scenarios. As for the cross-tokenizer KD, the differences in the tokenizers give rise to two fundamental challenges: (1) sequence misalignment caused by divergent tokenization strategies, and (2) mismatched vocabulary size and composition. While existing probability-matching methods attempt to address these issues, their efficacy remains limited due to suboptimal alignment in both the sequence and vocabulary aspects. To overcome these limitations, we propose Contextual Dynamic Mapping (CDM), a novel cross-tokenizer distillation framework that employs contextual information to enhance sequence alignment precision and dynamically improves vocabulary mapping. We evaluated the effectiveness of our approach across five advanced and widely-used model families (i.e, LLama3, Phi3, Gemma2, OPT and Qwen2), which were configured into three distinct teacher-student pairs. Our method shows significant advantages over existing cross-tokenizer distillation baselines across diverse benchmarks, including instruction-following, code generation and math. Notably, our analysis reveals that combining conventional same-tokenizer distillation and cross-tokenizer distillation through CDM yields further performance improvements. The code is available at https://github.com/pppa2019/ContexualDynamicMapping

📄 PDF Abstract BibTeX arXiv:2502.11104

Code (1)

pppa2019/contexualdynamicmapping 공식 구현 pytorch

Tasks

Code GenerationInstruction FollowingKnowledge DistillationMathModel Compression

Methods 이 논문이 사용한 방법론

Focus 설명 없음
OPT OPT is a suite of decoder-only pre-trained transformers ranging from 125M to 175B parameters. The model uses an AdamW optimizer and weight decay of 0.1. It follows a linear…

Similar Papers 제목 키워드 기반

Breaking the Tokenizer Barrier: On-Policy Distillation across Model Families

2026-06-08 · Yifan Niu, Han Xiao, Dongyi Liu, Zelong Wang 외 arxiv

On-Policy Distillation (OPD) has become a core technique in the post-training of Large Language Models (LLMs) for transferring knowledge from domain experts to student models. However, existing OPD distillation methods r…

CoT2Align: Cross-Chain of Thought Distillation via Optimal Transport Alignment for Language Models with Different Tokenizers

2025-02-24 · Anh Duc Le, Tu Vu, Nam Le Hai, Nguyen Thi Ngoc Diep 외

Large Language Models (LLMs) achieve state-of-the-art performance across various NLP tasks but face deployment challenges due to high computational costs and memory constraints. Knowledge distillation (KD) is a promising…

Knowledge Distillation

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

2026-02-25 · Duc Trung Vu, Pham Khanh Chi, Dat Phi Van, Linh Ngo Van 외 arxiv

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 constraine…

Knowledge Distillation

Cross-Tokenizer Distillation via Approximate Likelihood Matching

2025-03-25 · Benjamin Minixhofer, Ivan Vulić, Edoardo Maria Ponti

Distillation has shown remarkable success in transferring knowledge from a Large Language Model (LLM) teacher to a student LLM. However, current distillation methods predominantly require the same tokenizer between the t…

Large Language Model

Cross-Tokenizer LLM Distillation through a Byte-Level Interface

2026-04-08 · Avyav Kumar Singh, Yen-Chen Wu, Alexandru Cioba, Alberto Bernacchia 외 arxiv

Cross-tokenizer distillation (CTD), the transfer of knowledge from a teacher to a student language model when the two use different tokenizers, remains a largely unsolved problem. Existing approaches rely on heuristic st…