Back to AI News
arxiv

BubbleTea Reuses MoE Expert-Parallel Idle Time for Fine-Tuning

The paper introduces BubbleTea, a colocation system that fills idle GPU bubbles during Mixture-of-Experts LLM inference with short chunks of fine-tuning work. It predicts each GPU's per-layer bubble from the router's token assignment and uses compiler passes to derive fine-tuning chunks from the served model without porting architectures to a separate training engine. Evaluated on three MoE models across four A100 GPUs, BubbleTea reports 1.9–6.2× faster fine-tuning throughput than LLMStation while keeping p90 inference latency within 10% of serving alone on 50 of 63 workloads; on bursty traces it trains 3.3× faster than LLMStation and 14× faster than a stock trainer colocated under CUDA MPS.

AI Brief

It offers a concrete systems technique for ML infrastructure teams to reclaim idle MoE serving capacity for fine-tuning while controlling inference latency.

arXiv:2611.24496v1 Announce Type: cross Abstract: GPUs are expensive, yet latency-critical large language model (LLM) inference leaves part of them idle. Colocation reclaims this capacity by running best-effort work, such as fine-tuning, alongside serving. Many leading LLMs now use a Mixture-of-Experts (MoE) architecture, which adds a new source of idle time. Under expert parallelism, MoE experts are spread across GPUs, and because the router sends uneven numbers of tokens to each GPU, lightly loaded GPUs wait for the most loaded one. Even after applying load balancing solutions, such bubbles still take 8--13% of prefill time. We present BubbleTea, the first colocation system that leverages expert-parallel bubbles for fine-tuning. Unlike existing colocation systems, which make one scheduling decision for all GPUs per forward pass, BubbleTea predicts each GPU's bubble at every layer from the router's token assignment, and fills it with short chunks of fine-tuning work that reuse the weights already loaded for serving. We also introduce compiler passes that automatically derive these chunks from the served LLM as-is, without porting each model architecture to a separate training engine. Our evaluation on three MoE models, served on four A100 GPUs, shows that fine-tuning throughput with BubbleTea is 1.9--6.2$\times$ faster than with LLMStation, a state-of-the-art colocation system. BubbleTea keeps p90 inference latency within 10% of serving alone in 50 of 63 workloads, compared to 44 for LLMStation. On bursty production traces, BubbleTea trains 3.3$\times$ faster than LLMStation and 14$\times$ faster than a stock trainer colocated under CUDA MPS, with p90 inference latency staying within 7% of serving alone.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!