Metadata-Version: 2.2
Name: loom-py-rt-cuda
Version: 1.0.0rc3
Summary: CUDA backend for loom-py-rt
License: MIT License
         
         Copyright (c) 2026 loom-engine contributors
         
         Permission is hereby granted, free of charge, to any person obtaining a copy
         of this software and associated documentation files (the "Software"), to deal
         in the Software without restriction, including without limitation the rights
         to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
         copies of the Software, and to permit persons to whom the Software is
         furnished to do so, subject to the following conditions:
         
         The above copyright notice and this permission notice shall be included in all
         copies or substantial portions of the Software.
         
         THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
         IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
         FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
         AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
         LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
         OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
         SOFTWARE.
         
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.10
Requires-Dist: loom-py-rt==1.0.0rc3
Requires-Dist: nvidia-cuda-runtime-cu12>=12.9
Requires-Dist: nvidia-cublas-cu12>=12.9
Description-Content-Type: text/markdown

# loom-py-rt-cuda

The CUDA backend for [`loom-py-rt`](https://pypi.org/project/loom-py-rt/). One shared library in one
small wheel; there is nothing here to import.

```sh
pip install "loom-py-rt[cuda]"
```

```python
import loom

loom.devices()                                   # a CUDA device now appears
model = loom.Model.from_file(path, device="gpu")
```

The base package is unchanged by installing this — it discovers the library on `sys.path` at import
and `device="auto"` starts using it. That is what `GGML_BACKEND_DL` buys: no second copy of the
runtime per accelerator.

## If it appears not to have worked

Check `loom.devices()`. A backend whose driver is too old, or which finds no supported device, loads
**without error** and registers nothing — the only other symptom is a model running at CPU speed.

`device="gpu"` asks for an offload device with its own memory, preferring one the kernel confirms is a
GPU. It is not a promise that CUDA specifically was chosen. Pass `device="CUDA0"` to require this
backend.

## Versioning

This package pins the base with `==`, not `~=`. It carries a `libggml-cuda.so` that is dlopened beside
the base wheel's `libggml-base.so`, and ggml makes no ABI promise across revisions — so a base release
that moves its ggml pin invalidates every backend wheel published before it. The exact pin is what
stops pip from pairing two libraries that do not agree.
