Repository navigation
Pre-built CPU-only wheel for Windows (cp313) for version 0.3.16+ (Gemma 3 support) #2130
Description
Activity
Btw, do you know which workflow is responsible for building the CPU-only wheel?
Currently i'm trying to build wheels for CUDA 12.5 ~13.x on Windows (x64) and Linux (x64 & aarch64), but for some reason the CPU wheel didn't shows up at my release page https://github.com/anr2me/llama-cpp-python/releases
Was it
build-and-release.yaml? 🤔 because that is the only workflow that keeps failing during my tests (seems to be related to CPU features optimization)Edit: The CPU-only for v0.3.16 have shown now at https://anr2me.github.io/llama-cpp-python/whl/cpu
But i need to disable the native CPU features auto-detection to fix the workflow, so it doesn't have CPU feature optimizations (ie. AVX2, etc).
And i decided to enable BLAS acceleration (OpenBLAS on Windows & Linux, Accelerate on MacOS).
Hi @abetlen and the contributors,
I am currently working in a strictly restricted corporate environment where I do not have access to C++ build tools (MSVC) or direct internet access to compile packages from source. Because of this, I heavily rely on the pre-built CPU
.whlreleases provided here.I am trying to run a Gemma 3 based model (Trendyol Asure 12B), but I'm encountering the
unknown model architecture: 'gemma3'error. I noticed that the latest available pre-built Windows CPU wheel for Python 3.13 (cp313) in the releases is version0.3.2, which does not yet support this architecture.Would it be possible to trigger the build pipeline and upload a pre-built CPU-only wheel for Windows (
win_amd64) for Python 3.13 (cp313) for version0.3.16?This would be a lifesaver for users wanting to utilize the new Gemma 3 models and can't build.
Thank you for your time and hard work!