HUAWEI Computing Systems Lab

Pinned Loading

1. KVarN KVarN Public

   KVarN is a native vLLM KV-cache quantization backend for your agents: 3-5x more context, throughput above FP16, and FP16-level accuracy. Calibration-free, one flag.

   Python 513 37

2. SINQ SINQ Public

   Welcome to the official repository of SINQ! A novel, fast and high-quality quantization method designed to make any Large Language Model smaller while preserving accuracy [ICML 2026]

   Python 636 51

3. AC-LoRA AC-LoRA Public

   Welcome to the official repository of AC-LORA: (Almost) Training-Free Access Control-Aware Multi-Modal LLMs, a mechanism that provides training-free secure access control for LLMs using separate Lo…

   Jupyter Notebook 21 1

4. spire-hdl spire-hdl Public

   Spire is a Python embedded domain-specific language (DSL) for RTL generation. Its built-in optimizations reduce area and delay of circuits and often beat the traditional Verilog/Yosys flow.

   Python 51 6