Metadata-Version: 2.4
Name: raxis-cuda12
Version: 0.4.0
Summary: CUDA 12 backend of raxis (NVIDIA GPUs)
Home-page: https://github.com/pgmesa/raxis
Author: Pablo García Mesa
Author-email: pgmesa.sm@gmail.com
License: MIT
Classifier: Development Status :: 3 - Alpha
Classifier: Environment :: GPU :: NVIDIA CUDA
Classifier: Environment :: GPU :: NVIDIA CUDA :: 12
Classifier: Operating System :: Microsoft :: Windows
Classifier: Operating System :: POSIX :: Linux
Classifier: Programming Language :: C++
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering
Requires-Python: >=3.10,<3.15
Description-Content-Type: text/markdown
License-File: LICENSE
License-File: THIRD_PARTY_NOTICES.txt
Requires-Dist: raxis==0.4.0
Requires-Dist: nvidia-cublas-cu12>=12.1
Dynamic: author
Dynamic: author-email
Dynamic: classifier
Dynamic: description
Dynamic: description-content-type
Dynamic: home-page
Dynamic: license
Dynamic: license-file
Dynamic: requires-dist
Dynamic: requires-python
Dynamic: summary

# raxis-cuda12

The CUDA 12 backend of [raxis](https://github.com/pgmesa/raxis), a
readable re-implementation of the core of PyTorch (ATen + autograd) in C++20.

```bash
pip install "raxis[cuda12]"   # or pip install raxis-cuda12
```

installs raxis together with its GPU kernels, caching allocator and cuBLAS
bindings; `import raxis` loads them, and tensors move to the GPU with
`.cuda()` / `.to("cuda")` as in PyTorch. It needs an NVIDIA driver (the CUDA
runtime is linked statically, cuBLAS comes from NVIDIA's wheel).

raxis-cuda12 runs on drivers r525 and newer and on GPUs from Volta (sm_70)
on; raxis-cuda13 needs a driver r580 or newer and a GPU from Turing (sm_75)
on. Install only one of them.
