DistiLLM-2: A Contrastive Approach Boosts the Distillation of LLMs
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913866411147264 |
|---|---|
| author | Ko, Jongwoo Chen, Tianyi Kim, Sungnyun Ding, Tianyu Liang, Luming Zharkov, Ilya Yun, Se-Young |
| author_facet | Ko, Jongwoo Chen, Tianyi Kim, Sungnyun Ding, Tianyu Liang, Luming Zharkov, Ilya Yun, Se-Young |
| contents | Despite the success of distillation in large language models (LLMs), most prior work applies identical loss functions to both teacher- and student-generated data. These strategies overlook the synergy between loss formulations and data types, leading to a suboptimal performance boost in student models. To address this, we propose DistiLLM-2, a contrastive approach that simultaneously increases the likelihood of teacher responses and decreases that of student responses by harnessing this synergy. Our extensive experiments show that DistiLLM-2 not only builds high-performing student models across a wide range of tasks, including instruction-following and code generation, but also supports diverse applications, such as preference alignment and vision-language extensions. These findings highlight the potential of a contrastive approach to enhance the efficacy of LLM distillation by effectively aligning teacher and student models across varied data types. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_07067 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | DistiLLM-2: A Contrastive Approach Boosts the Distillation of LLMs Ko, Jongwoo Chen, Tianyi Kim, Sungnyun Ding, Tianyu Liang, Luming Zharkov, Ilya Yun, Se-Young Computation and Language Artificial Intelligence Machine Learning Despite the success of distillation in large language models (LLMs), most prior work applies identical loss functions to both teacher- and student-generated data. These strategies overlook the synergy between loss formulations and data types, leading to a suboptimal performance boost in student models. To address this, we propose DistiLLM-2, a contrastive approach that simultaneously increases the likelihood of teacher responses and decreases that of student responses by harnessing this synergy. Our extensive experiments show that DistiLLM-2 not only builds high-performing student models across a wide range of tasks, including instruction-following and code generation, but also supports diverse applications, such as preference alignment and vision-language extensions. These findings highlight the potential of a contrastive approach to enhance the efficacy of LLM distillation by effectively aligning teacher and student models across varied data types. |
| title | DistiLLM-2: A Contrastive Approach Boosts the Distillation of LLMs |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2503.07067 |