Ant Group and the vLLM community release Humming as open source
What happened
On October 21, AI HOT reported that Ant Group and the vLLM community released Humming, an open-source library of quantized inference kernels. It uses one unified architecture that covers both dense and MoE models across several quantization formats, so different model types and formats fall under the same design. On hardware, Humming supports NVIDIA Turing and later GPUs, and also the T-Head Zhenwu 810E. The report discloses only the open-sourcing, the model and quantization formats covered, and the hardware support range.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- AI HOT · Products蚂蚁集团与 vLLM 社区开源量化推理内核库 Humming
蚂蚁集团与 vLLM 社区开源 Humming，以统一架构支持稠密模型和 MoE 的多种量化格式，覆盖 Turing 及之后的 NVIDIA GPU 和平头哥真武 810E。
Heat over time
Not enough continuous observations to draw a trend yet.