Zap11's picture
Update README.md
6bcbf1f verified
# llama-cpp-python — Free-Tier Friendly Wheel
This Space provides a **prebuilt `llama-cpp-python` wheel** designed to work
**reliably on Hugging Face Free tier Spaces**.
No compilation. No system packages. No build failures.
If your Space crashes during `pip install llama-cpp-python`, this wheel is the fix.
---
## Optimized for Hugging Face Free Tier
Hugging Face Free tier Spaces are:
- CPU-only
- Limited in memory
- Not suitable for native compilation
This wheel is built **ahead of time** so it can be installed instantly without
triggering CMake, compilers, or BLAS detection.
---
## What this wheel gives you
- ✅ Works on **HF Free tier CPU Spaces**
- ✅ Linux (ubuntu-22.04 compatible)
- ✅ Python 3.10
- ✅ OpenBLAS enabled (`GGML_BLAS=ON`)
- ✅ No system dependencies required
- ✅ No build step during Space startup
- ✅ Fast, reliable `pip install`
---
## How to use in a Space (Free tier)
1. Download the wheel from the GitHub repository
2. Upload it to your Space
3. Install it in your Space startup:
pip install llama_cpp_python-*.whl>
## That’s it — your Space will start without build errors.
## Build details
This wheel was built using:
abetlen/llama-cpp-python (recursive submodules)
OpenBLAS (GGML_VENDOR=OpenBLAS)
scikit-build-core
ninja
python -m build --wheel --no-isolation
## Build environment:
OS: Ubuntu 22.04
Python: 3.10
## Why not build from source on HF?
On Free tier Spaces, building from source often fails due to:
Missing compilers
Missing BLAS libraries
Memory limits
Build timeouts
This prebuilt wheel avoids all of those issues.
## Notes
CPU-only (no CUDA)
Intended for inference workloads
Not an official upstream release
## Credits
All credit goes to the maintainers of llama-cpp-python and llama.cpp.
This Space exists solely to make Free tier usage painless.