vLLM Day-0 Support for Kimi K3: Deploying a 2.8T Parameter MoE Model
vllm_project · x · 2026-07-30
The vLLM team announced efficient Day-0 support for Moonshot AI's open-weight Kimi K3 model. Kimi K3 is a 2.8-trillion-parameter Mixture-of-Experts (MoE) model with 16 experts active per token, featuring a 1M-token context window and native vision capabilities.
The post details the engineering challenges and solutions for adapting vLLM to this architecture:
- Architecture Adaptation: Specialized kernel and cache optimizations were made for new architectural features like Kimi Delta Attention (KDA), Attention Residuals (AttnRes), and MXFP4 MoE.
- Inference Acceleration: Inferact has open-sourced a DSpark speculator, which, combined with prefill/decode disaggregation and long-context deployment recipes, significantly boosts inference efficiency.
- Deployment Guide: Provides complete Docker startup commands and configurations for running the model on 8 NVIDIA B300 or AMD MI355X GPUs.
Related event: Kimi K3 Gets Day-0 vLLM and AMD Support Across Clouds(14 posts)→
More from Infra
- Qualcomm goes agent-centric: Snapdragon 8 Elite Gen 6 and agent-native devices — jiqizhixin · 2026-09-23
- Unsloth Desktop Hotfix Adds Qwen-Image-2.1 Image Editing and Fixes GGUF Loading — danielhanchen · 2026-09-23
- Qwen 3.6 35B-A3B Q6 hits ~50 tok/s on a 128GB Strix Halo — what's the best local model now? — jankeydankey · 2026-09-23
- Together AI adds canary rollouts for zero-downtime model upgrades on dedicated inference — togethercompute · 2026-09-23
- Dedicated Hardware for Running AI Agents at Scale Arrives — cyrilzakka · 2026-09-23
- Ternary Bonsai 2 27B: 5.9GB weights retain ~95% of full-precision reasoning — cephaloform · 2026-09-23