arXiv subjects

Zepeng Zhao

Publications and source records attributed to Zepeng Zhao.

3 recordsLinked to original sources

arXiv2025

MPK: A Compiler and Runtime for Mega-Kernelizing Tensor Programs

We introduce Mirage Persistent Kernel (MPK), the first compiler and runtime system that automatically transforms multi-GPU model inference into a single high-performance mega-kernel. MPK introduces an SM-level graph representation that captures data dependencies at the granularity of individual streaming multiprocessors (SMs), enabling cross-operator software pipelining, \rev{fine-grained overlap of computation and communication, and other optimizations that are infeasible under the conventional kernel-per-operator execution model}. The MPK compiler lowers tensor programs into optimized SM-level task graphs and generates fast CUDA implementations for each task, while the MPK in-kernel parallel runtime executes these tasks within a single persistent mega-kernel using decentralized scheduling across SMs. Together, these components provide end-to-end kernel fusion with minimal developer effort, while preserving the flexibility of existing programming models. Our evaluation shows that MPK significantly outperforms existing kernel-per-operator LLM serving systems, achieving up to 1.7$\times$ lower end-to-end inference latency and pushing LLM inference performance close to the limits of the underlying hardware. MPK is publicly available at https://github.com/mirage-project/mirage.

cs.DC↗

arXiv2025

SuperGen: An Efficient Ultra-high-resolution Video Generation System with Sketching and Tiling

Diffusion models have recently achieved remarkable success in generative tasks (e.g., image and video generation), and the demand for high-quality content (e.g., 2K/4K videos) is rapidly increasing across various domains. However, generating ultra-high-resolution videos on existing standard-resolution (e.g., 720p) platforms remains challenging due to the excessive re-training requirements and prohibitively high computational and memory costs. To this end, we introduce SUPERGEN, an efficient tile-based framework for ultra-high-resolution video generation. SUPERGEN features a novel training-free algorithmic innovation with tiling to successfully support a wide range of resolutions without additional training efforts while significantly reducing both memory footprint and computational complexity. Moreover, SUPERGEN incorporates a tile-tailored, adaptive, region-aware caching strategy that accelerates video generation by exploiting redundancy across denoising steps and spatial regions. SUPERGEN also integrates cache-guided, communication-minimized tile parallelism for enhanced throughput and minimized latency. Evaluations show that SUPERGEN maximizes performance gains while achieving high output quality across various benchmarks.

cs.LG↗

arXiv2025

PATCHEDSERVE: A Patch Management Framework for SLO-Optimized Hybrid Resolution Diffusion Serving

The Text-to-Image (T2I) diffusion model has emerged as one of the most widely adopted generative models. However, serving diffusion models at the granularity of entire images introduces significant challenges, particularly under multi-resolution workloads. First, image-level serving obstructs batching across requests. Second, heterogeneous resolutions exhibit distinct locality characteristics, making it difficult to apply a uniform cache policy effectively. To address these challenges, we present PatchedServe, a Patch Management Framework for SLO-Optimized Hybrid-Resolution Diffusion Serving. PatchedServe is the first SLO-optimized T2I diffusion serving framework designed to handle heterogeneous resolutions. Specifically, it incorporates a novel patch-based processing workflow that substantially improves throughput for hybrid-resolution inputs. Moreover, PatchedServe devises a patch-level cache reuse policy to fully exploit diffusion redundancies and integrates an SLO-aware scheduling algorithm with lightweight online latency prediction to improve responsiveness. Our evaluation demonstrates that PatchedServe achieves 30.1 % higher SLO satisfaction than the state-of-the-art diffusion serving system, while preserving image quality.

cs.DC↗