Installation¶
This guide covers the installation of vllm-spyre-next using uv, a fast Python package installer and resolver.
Prerequisites¶
- Python >= 3.11
uvpackage manager installed (installation guide)- Access to systems where
sendnnis available (required for torch-spyre compilation)
Installation with uv sync¶
The vllm-spyre-next plugin uses uv for dependency management and installation. The project configuration in vllm_spyre_next/pyproject.toml includes several important settings that ensure proper installation:
Basic Installation¶
From the vllm_spyre_next directory, run:
This command will:
- Install all project dependencies
- Build vLLM from source with CPU backend support
- Build torch-spyre from source
- Install PyTorch 2.10.0 from the CPU-specific index
Configuration Highlights¶
The pyproject.toml file includes several key configurations:
1. Build Configuration¶
[tool.uv]
build-constraint-dependencies = ["torch==2.10.0"]
extra-build-variables = { vllm = { VLLM_TARGET_DEVICE = "cpu" } }
These settings ensure:
- All packages are built with the same PyTorch version (2.10.0)
- vLLM is built specifically for the CPU backend
2. Source Repositories¶
The plugin pulls dependencies from specific Git repositories:
[tool.uv.sources]
vllm = { git = "https://github.com/vllm-project/vllm", rev = "..." }
torch-spyre = { git = "https://github.com/torch-spyre/torch-spyre", rev = "..." }
This ensures that both torch-spyre and vllm are compiled from source, instead of pulling pre-compiled wheels from pypi.
3. PyTorch CPU Index¶
We ensure that the cpu flavor of pytorch is installed, as we're not building cuda support.
Verification¶
After installation, verify the plugin is correctly installed:
Development Installation¶
For development work, install with the dev dependency group:
This includes additional tools like pytest, pytest-asyncio, and other testing utilities.
Troubleshooting¶
Build Failures¶
If you encounter build failures:
- torch-spyre compilation: Ensure
sendnnis available on your system. See internal development documentation for how to set up a dev environment withsendnn. - vLLM build: Check that you have sufficient memory and CPU resources for compilation
- Dependency conflicts: Review the
override-dependenciessection inpyproject.toml
Next Steps¶
After installation, you can start using vllm-spyre-next with your vLLM applications. The plugin will automatically be loaded by vLLM when the appropriate platform is detected.