Installation#

Prerequisites#

  • Python 3.10+

  • CUDA 12.4+

Setup using Python#

Install Stable LMCache from PyPI#

The simplest way to install the latest stable release of LMCache is through PyPI:

pip install lmcache

Install Latest LMCache from TestPyPI#

These wheels are continually built from the latest LMCache source code (not officially stable release).

pip install --index-url https://pypi.org/simple --extra-index-url https://test.pypi.org/simple lmcache==0.2.2.dev57

See the latest pre-release of LMCache: latest LMCache pre-releases and replace 0.2.2.dev57 with the latest pre-release version.

This will install all dependencies from the real PyPI and only LMCache itself from TestPyPI.

Confirm that you have the latest pre-release:

python
>>> import lmcache
>>> from importlib.metadata import version
>>> print(version("lmcache"))
0.2.2.dev57 # should be the latest pre-release version you installed

Install Latest LMCache from Source#

To install from source, clone the repository and install in editable mode:

git clone https://github.com/LMCache/LMCache.git
cd LMCache
pip install -e .

Install LMCache with uv#

We recommend developers to use uv for a better package management:

git clone https://github.com/LMCache/LMCache.git
cd LMCache

uv venv --python 3.12
source .venv/bin/activate
uv pip install -e .

LMCache with vLLM v1#

LMCache is integrated with the latest vLLM (vLLM v1). To use it, install the latest vLLM package:

pip install vllm

Test whether LMCache works with vLLM v1 by running:

python3 -c "import vllm.distributed.kv_transfer.kv_connector.v1.lmcache_connector"

LMCache with vLLM v0#

Note

LMCache is also integrated with vLLM v0. Refer to the example in vLLM. See the examples README to understand how to run the script for vLLM v0.

Setup using Docker#

Pre-built vLLM + LMCache Images#

We provide pre-built Docker images that include vLLM integration:

docker pull lmcache/vllm-openai:2025-04-18

Note

Currently, we build and release Docker images manually. An automated Docker build/release GitHub workflow will be set up soon. Contributions to this effort are welcomed!