BTS: Harmonizing Specialized Experts into a Generalist LLM
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_ | 1866913673744744448 |
|---|---|
| author | Zhang, Qizhen Bhargava, Prajjwal Bi, Chloe Cai, Chris X. Foerster, Jakob Fu, Jeremy Koura, Punit Singh Silva, Ruan Shen, Sheng Dinan, Emily Gururangan, Suchin Lewis, Mike |
| author_facet | Zhang, Qizhen Bhargava, Prajjwal Bi, Chloe Cai, Chris X. Foerster, Jakob Fu, Jeremy Koura, Punit Singh Silva, Ruan Shen, Sheng Dinan, Emily Gururangan, Suchin Lewis, Mike |
| contents | We present Branch-Train-Stitch (BTS), an efficient and flexible training algorithm for combining independently trained large language model (LLM) experts into a single, capable generalist model. Following Li et al., we start with a single seed language model which is branched into domain-specific (e.g., coding or math) experts with continual pretraining. BTS combines experts into a generalist model using lightweight stitch layers, which are inserted between frozen experts and the seed LLM, and trained on a small datamix of the expert domains. Stitch layers enable the seed LLM to integrate representations from any number of experts during the forward pass, allowing it to generalize to new domains, despite remaining frozen. Because BTS does not alter the constituent LLMs, BTS provides a modular and flexible approach: experts can be easily removed and new experts can be added with only a small amount of training. Compared to alternative model merging approaches, BTS yields the best generalist performance on a variety of downstream tasks, retaining the specialized capabilities of each of the experts. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_00075 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | BTS: Harmonizing Specialized Experts into a Generalist LLM Zhang, Qizhen Bhargava, Prajjwal Bi, Chloe Cai, Chris X. Foerster, Jakob Fu, Jeremy Koura, Punit Singh Silva, Ruan Shen, Sheng Dinan, Emily Gururangan, Suchin Lewis, Mike Computation and Language Machine Learning We present Branch-Train-Stitch (BTS), an efficient and flexible training algorithm for combining independently trained large language model (LLM) experts into a single, capable generalist model. Following Li et al., we start with a single seed language model which is branched into domain-specific (e.g., coding or math) experts with continual pretraining. BTS combines experts into a generalist model using lightweight stitch layers, which are inserted between frozen experts and the seed LLM, and trained on a small datamix of the expert domains. Stitch layers enable the seed LLM to integrate representations from any number of experts during the forward pass, allowing it to generalize to new domains, despite remaining frozen. Because BTS does not alter the constituent LLMs, BTS provides a modular and flexible approach: experts can be easily removed and new experts can be added with only a small amount of training. Compared to alternative model merging approaches, BTS yields the best generalist performance on a variety of downstream tasks, retaining the specialized capabilities of each of the experts. |
| title | BTS: Harmonizing Specialized Experts into a Generalist LLM |
| topic | Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2502.00075 |