Disaggregation of Diffusion Pipeline on Ascend NPU - Sglang

Quick Start

Please follow the NPU installation guide first.

Then install Mooncake from sources.

bash

git clone https://github.com/kvcache-ai/Mooncake.git
cd Mooncake
git checkout v0.3.10.post2
bash dependencies.sh
mkdir build
cd build
export GLOG_logtostderr=1
cmake -DUSE_ASCEND_DIRECT=ON ..
make -j
make install
cd ../mooncake-wheel/
pip install .

Before run servers ensure that *.so files of mooncake in LD_LIBRARY_PATH

bash

export LD_LIBRARY_PATH=/usr/local/python3.11.14/lib/python3.11/site-packages/mooncake:$LD_LIBRARY_PATH

Examples

For usage examples please follow the main disaggregation guide