11 posts
Introducing Humming, a quantized inference kernel library from Ant Group and the vLLM community, with broad GPU and quantization support, dense and MoE performance results, and vLLM integration.
How asynchronous preparation and NIXL EP enabled near-zero-downtime Elastic EP scaling in vLLM.
vLLM-Omni runs Sand AI's MAGI-2 Preview audio-video generation model on eight Moore Threads MTT S5000 GPUs through torchada's CUDA compatibility layer, then optimizes MoE, mHC, attention, parallelism, and compilation.
Novita AI has open-sourced Chord, a high-performance W4A16 MoE CUDA kernel for Kimi K2.x serving shapes, with a Humming-compatible indexed path and grouped SM90 operators.
A performance model for LLM serving: inspect local shapes, remove repeated work, verify data movement and dispatch, then follow the queue.
Day-0 vLLM support for Qwen3.8-2.4T-A95B: a 2.4-trillion-parameter hybrid MoE model served out of the box, with FP8/BF16 checkpoints plus NVFP4 and MXFP4 quantized weights, and co-developed kernels on NVIDIA and AMD hardware.
How we took GLM-5.2-NVFP4 from 40 ms to 17 ms mean TPOT on 24 B300 GPUs with vLLM: P/D disaggregation, MTP speculative decoding, Model Runner V2, and the SLA-first trade-offs behind the final configuration.
What vLLM AFD Plugin adds to the vLLM ecosystem: Attention–FFN disaggregation for MoE serving, GPU and Ascend NPU backends, connector-based execution, and graph and ubatching support.
How HPC-Ops integrates Hopper-optimized attention and FP8 MoE backends into vLLM for Tencent Hunyuan Hy3, improving mixed-length decode, MoE latency, TTFT, and TPOT on NVIDIA H20.
How vLLM serves MiniMax M3 with MiniMax Sparse Attention, multimodal and reasoning parsers, MXFP8 weights, and long-context deployment recipes.
How Elastic Expert Parallelism lets vLLM scale Mixture-of-Experts serving up or down at runtime by changing data-parallel workers, redistributing experts, and coordinating live topology changes without server restarts.
