Saved in:
Bibliographic Details
Main Authors: Sukhbaatar, Sainbayar, Golovneva, Olga, Sharma, Vasu, Xu, Hu, Lin, Xi Victoria, Rozière, Baptiste, Kahn, Jacob, Li, Daniel, Yih, Wen-tau, Weston, Jason, Li, Xian
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2403.07816
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909134826242048
author Sukhbaatar, Sainbayar
Golovneva, Olga
Sharma, Vasu
Xu, Hu
Lin, Xi Victoria
Rozière, Baptiste
Kahn, Jacob
Li, Daniel
Yih, Wen-tau
Weston, Jason
Li, Xian
author_facet Sukhbaatar, Sainbayar
Golovneva, Olga
Sharma, Vasu
Xu, Hu
Lin, Xi Victoria
Rozière, Baptiste
Kahn, Jacob
Li, Daniel
Yih, Wen-tau
Weston, Jason
Li, Xian
contents We investigate efficient methods for training Large Language Models (LLMs) to possess capabilities in multiple specialized domains, such as coding, math reasoning and world knowledge. Our method, named Branch-Train-MiX (BTX), starts from a seed model, which is branched to train experts in embarrassingly parallel fashion with high throughput and reduced communication cost. After individual experts are asynchronously trained, BTX brings together their feedforward parameters as experts in Mixture-of-Expert (MoE) layers and averages the remaining parameters, followed by an MoE-finetuning stage to learn token-level routing. BTX generalizes two special cases, the Branch-Train-Merge method, which does not have the MoE finetuning stage to learn routing, and sparse upcycling, which omits the stage of training experts asynchronously. Compared to alternative approaches, BTX achieves the best accuracy-efficiency tradeoff.
format Preprint
id arxiv_https___arxiv_org_abs_2403_07816
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Branch-Train-MiX: Mixing Expert LLMs into a Mixture-of-Experts LLM
Sukhbaatar, Sainbayar
Golovneva, Olga
Sharma, Vasu
Xu, Hu
Lin, Xi Victoria
Rozière, Baptiste
Kahn, Jacob
Li, Daniel
Yih, Wen-tau
Weston, Jason
Li, Xian
Computation and Language
Artificial Intelligence
We investigate efficient methods for training Large Language Models (LLMs) to possess capabilities in multiple specialized domains, such as coding, math reasoning and world knowledge. Our method, named Branch-Train-MiX (BTX), starts from a seed model, which is branched to train experts in embarrassingly parallel fashion with high throughput and reduced communication cost. After individual experts are asynchronously trained, BTX brings together their feedforward parameters as experts in Mixture-of-Expert (MoE) layers and averages the remaining parameters, followed by an MoE-finetuning stage to learn token-level routing. BTX generalizes two special cases, the Branch-Train-Merge method, which does not have the MoE finetuning stage to learn routing, and sparse upcycling, which omits the stage of training experts asynchronously. Compared to alternative approaches, BTX achieves the best accuracy-efficiency tradeoff.
title Branch-Train-MiX: Mixing Expert LLMs into a Mixture-of-Experts LLM
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2403.07816