openai-gemm.pytorch

PyTorch bindings for openai-gemm.

Installation

Clone original openai-gemm and add it to PYTHONPATH, install pycuda:

pip install pycuda

and follow instructions to install PyTorch on http://pytorch.org

No neon installation needed.

The library defines matmul function similar to the one that works with neon: https://github.com/openai/openai-gemm/blob/master/openai_gemm.py#L14, which instead of neon matrices takes torch.cuda.FloatTensor or torch.cuda.HalfTensor as A, B and C.

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
README.md		README.md
benchmark.py		benchmark.py
openai_gemm_pytorch.py		openai_gemm_pytorch.py
test.py		test.py