ME-Switch: A Memory-Efficient Expert Switching Framework for Large Language Models
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866912087359356928 |
|---|---|
| author | Liu, Jing Gong, Ruihao Zhang, Mingyang He, Yefei Cai, Jianfei Zhuang, Bohan |
| author_facet | Liu, Jing Gong, Ruihao Zhang, Mingyang He, Yefei Cai, Jianfei Zhuang, Bohan |
| contents | LLM development involves pre-training a foundation model on massive data, followed by fine-tuning on task-specific data to create specialized experts. Serving these experts can pose significant memory challenges, as loading all experts onto devices is impractical, and frequent switching between experts in response to user requests can incur substantial I/O costs. Previous approaches decompose the expert weights as the pre-trained weights plus delta weights, followed by quantizing the delta weights using output channel-wise step sizes to reduce the model size. However, these methods overlook the fact that certain input channels of delta weights can cause significant quantization errors at extremely low bitwidths. Additionally, existing methods assume that the appropriate model for a user request is known in advance, which is not the case in practice. To this end, we introduce ME-Switch, a memory-efficient expert switching framework tailored for serving multiple LLMs. To condense the number of bits required for describing the delta weights, we propose a salient-aware delta compression method that identifies salient input channels based on reconstruction error and applies mixed-precision quantization, reducing non-salient channels to low bits while keeping salient ones intact, cutting storage demand without compromising performance. Moreover, we develop a model-level routing method that efficiently directs user queries to the most suitable expert by performing domain classification. Extensive experiments show the promising memory efficiency and routing performance of ME-Switch. For example, when serving three models from the Mistral-7B family, ME-Switch reduces the model size by $1.74\times$ and maintains nearly lossless performance on instruction, mathematical reasoning, and code generation tasks. Notably, our method can efficiently serve 16 Mistral-7B models on a single NVIDIA A100 GPU. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_09041 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | ME-Switch: A Memory-Efficient Expert Switching Framework for Large Language Models Liu, Jing Gong, Ruihao Zhang, Mingyang He, Yefei Cai, Jianfei Zhuang, Bohan Computation and Language Artificial Intelligence Machine Learning LLM development involves pre-training a foundation model on massive data, followed by fine-tuning on task-specific data to create specialized experts. Serving these experts can pose significant memory challenges, as loading all experts onto devices is impractical, and frequent switching between experts in response to user requests can incur substantial I/O costs. Previous approaches decompose the expert weights as the pre-trained weights plus delta weights, followed by quantizing the delta weights using output channel-wise step sizes to reduce the model size. However, these methods overlook the fact that certain input channels of delta weights can cause significant quantization errors at extremely low bitwidths. Additionally, existing methods assume that the appropriate model for a user request is known in advance, which is not the case in practice. To this end, we introduce ME-Switch, a memory-efficient expert switching framework tailored for serving multiple LLMs. To condense the number of bits required for describing the delta weights, we propose a salient-aware delta compression method that identifies salient input channels based on reconstruction error and applies mixed-precision quantization, reducing non-salient channels to low bits while keeping salient ones intact, cutting storage demand without compromising performance. Moreover, we develop a model-level routing method that efficiently directs user queries to the most suitable expert by performing domain classification. Extensive experiments show the promising memory efficiency and routing performance of ME-Switch. For example, when serving three models from the Mistral-7B family, ME-Switch reduces the model size by $1.74\times$ and maintains nearly lossless performance on instruction, mathematical reasoning, and code generation tasks. Notably, our method can efficiently serve 16 Mistral-7B models on a single NVIDIA A100 GPU. |
| title | ME-Switch: A Memory-Efficient Expert Switching Framework for Large Language Models |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2406.09041 |