Skip to content

INSTALL.md

Latest commit

 

History

History
67 lines (52 loc) · 2.2 KB

INSTALL.md

File metadata and controls

67 lines (52 loc) · 2.2 KB

Mirage Installation

The quickest way to try Mirage is through our prebuilt docker images with all dependencies preinstalled. Mirage can also be built from source code using the following instructions.

Docker images

We require docker and nvidia-docker to run the Mirage docker images.

  • First, clone the Mirage gitpub repository to obtain necessary scripts.
git clone --recursive https://www.github.com/mirage-project/mirage
  • Second, use the following command to run a Mirage docker image. The default CUDA version is 12.4.
/path-to-mirage/docker/run_docker.sh mlso/mirage
  • You are ready to use Mirage now. Try some of our demos to superoptimize DNNs.
python demo/demo_group_query_attention_spec_decode.py --checkpoint demo/checkpoint_group_query_attn_spec_decode.json

Install from source code

Prerequisties

  • CMAKE 3.24 or higher
  • Cython 0.28 or higher
  • CUDA 11.0 or higher and CUDNN 8.0 or higher

Clone github repo

To get started, you can clone the Mirage source code from github.

git clone --recursive https://www.github.com/mirage-project/mirage
cd mirage

Install Z3 or build Z3 from source code

If the environment does not have a pre-installed Z3, you can build Z3 from source code using the following command lines

cd /path-to-mirage/deps/z3
mkdir build
cd build
cmake ..
make -j
export LD_LIBRARY_PATH=/path-to-mirage/deps/z3/build:LD_LIBRARY_PATH

This will install Z3 in the /path-to-mirage/deps/z3/build folder and add the fold into LD_LIBRARY_PATH.

Build the Mirage runtime library

Second, you will need to build the Mirage runtime library. You will need to set CUDACXX and Z3_DIR to let cmake find the paths to CUDA and Z3 librarires.

export CUDACXX=/usr/local/cuda/bin/nvcc
export Z3_DIR=/path-to-mirage/deps/z3/build
mkdir build; cd build; cmake ..
make -j

Install the Mirage python package

Finally, you will install the Mirage python package, which allows you to use Mirage's python package to superoptimize DNNs.

cd /path-to-mirage/python
python setup.py install