Search papers, labs, and topics across Lattice.
This paper introduces FlashBoot, a novel weight-loading subsystem designed to significantly reduce latency in large Mixture-of-Experts (MoE) models by addressing structural inefficiencies in weight memory and transfer processes. By implementing a contiguous tensor memory layout and a zero-copy transfer mechanism, FlashBoot achieves remarkable speed improvements, reducing single-node weight loading times by up to 50x and concurrent rack-level loading by over 270x. The results demonstrate that FlashBoot can effectively utilize GPU bandwidth and streamline the deployment of large models across multiple nodes, which is critical for elastic deployment scenarios.
FlashBoot slashes weight loading times from 20.1 seconds to just 0.4 seconds, revolutionizing how large models are deployed at scale.
Flagship Mixture-of-Experts (MoE) models are growing fast along two axes at once: total parameter count and the number of experts. In elastic deployment scenarios, many GPUs across many nodes must become serving-ready quickly, and this growth makes weight loading a noticeable part of the latency budget. Even on NVIDIA's GB300 NVL72, today's state-of-the-art loaders leave most of that bandwidth unused. The losses are structural: (C1) weight memory is fragmented into tens of thousands of per-tensor objects, so transfers run far below link bandwidth; (C2) cross-node replication is gated by NCCL communicator setup, which costs 10-110 s before a single weight byte moves; and (C3) the existing cross-node GPU->GPU clone path is serial and scales poorly to concurrent multi-node bring-up. We present FlashBoot, a hardware-friendly, framework-workflow co-designed weight-loading subsystem built on SGLang. At its core is FabricArena, a contiguous, exportable and inter-node addressable tensor memory layout. On top of it, FlashLoad loads from CPU as a single bulk, zero-copy transfer, and FlashClone replicates a resident model from a remote GPU via a remote-mapping mechanism that removes NCCL setup. In experiments on NVL72 with DeepSeek-V4-Pro and DeepSeek-V4-Flash, FlashClone maps remote weight memory in ~10 ms (versus 10-110 s for NCCL) and sustains >=700 GB/s per clone. Against the state of the art, FlashBoot accelerates single-node weight loading by up to 50x (from 20.1 s to 0.4 s) and concurrent rack-level weight loading by >270x (from 87 s to 0.32 s). Our code will be made publicly available.
