📚 Building Portable Large Models on Heterogeneous AI Accelerators Wit...
Talkat KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026 · 2026-09-07..09-09·
summary
Shows how the FlagOS open-source stack is used to port large LLM and multimodal models (examples: Qwen, MiniMax, DeepSeek) onto diverse AI accelerators by covering the full runtime path beneath PyTorch, vLLM and SGLang. Explains concrete components and tools: Triton for portable kernel programming, FlagGems for PyTorch operator coverage, unified FlagOS plugins for runtimes like vLLM, Megatron and VeRL to support training/inference on different accelerators, and KernelGen for auto-generating accelerator-specific kernels.
AI-written summary · about
related talks
25mKeynote: Running Large Models on Diverse AI Chips: PyTorch + Open-Source Stack (FlagO...
37mDeep Learning Neural Network Acceleration at the Edge
11mKeynote: Building a Large Model Inference Platform for Heterogeneous Chinese Chips Based on VLLM
20mArchitecting AI: Efficient Models, Everywhere
25mRace to Superintelligence: Building Low-Friction AI Infrastructure for a Heterogen...
32mBuilding Once, Running Everywhere: The Future of Portable AI Applications
23mHugging Face Accelerate: Making Device-Agnostic ML Training and Inference Easy...
06mThe Building Blocks of Agentic Al