Ant Group Open-Sources Humming for Quantized Inference

1 reports archived0 independent sourcesupdated 10/21/2026, 08:00:00

Synthesis & Latest Updates

Ant Group and the vLLM community have open-sourced Humming, a high-performance kernel library for quantized inference. It supports NVIDIA and domestic chips, handles diverse weight/activation formats, and outperforms existing implementations like Marlin and FlashInfer on dense and MoE workloads.

LATEST/Ant Group and the vLLM community have open-sourced Humming, a high-performance kernel library for quantized inference. It supports NVIDIA and domestic chips, handles diverse weight/activation formats, and outperforms existing implementations like Marlin and FlashInfer on dense and MoE workloads.

Hourly heat curve

168 fully observed hours

Now

0.0

Peak

9.815:00

24h change

–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first

vLLM 官方博客T1Oct 21, 08:00·78 pts
Ant Group Open-Sources Humming for Quantized Inference
- Supports NVIDIA GPUs from Turing onwards and CUDA-compatible devices like T-Head Zhenwu 810E
- Unified architecture handles diverse integer/float quantization combinations for dense and MoE models
- Benchmarks show significantly better execution performance than vLLM Marlin, FlashInfer, and DeepGEMM
View Related Reports & Analysis Original Source