microsoft / onnxruntime

ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator
https://onnxruntime.ai
MIT License
13.4k stars 2.74k forks source link

[Documentation] Install page missing instructions for onnxruntime-rocm #21036

Open shaunclimbon opened 2 weeks ago

shaunclimbon commented 2 weeks ago

Describe the documentation issue

The installation matrix provides correct guidance for ORT with ROCm (i.e. build from source) but the actual install documentation is missing it. In particular, the Python Installs section covers ORT>CPU and ORT>GPU>CUDA but not ORT>GPU>ROCm. Albeit Linux specific, it still seems like this info should be present.

As of reporting this, building the onnxruntime-rocm wheel was the recommended route (as documented here) but it's worth noting that I found pre-built wheels in a couple locations:

Page / URL

https://onnxruntime.ai/docs/install/

MaanavD commented 2 weeks ago

Hey @shaunclimbon , do you have an email I could follow up with you on? I'm happy to add these missing instructions, but I want to make sure we accurately represent them in a way you're happy with :)

Alternatively,

I'm assuming you want a section in the first link for Install ONNX Runtime GPU (ROCm)

with instructions along the lines of

Prerequisites Install ROCm The ROCm execution provider for ONNX Runtime is built and tested with ROCm6.0.0

Build Instructions LINUX ./build.sh --config <Release|Debug|RelWithDebInfo> --use_rocm --rocm_home

See more information on the ROCm Execution Provider here.

shaunpamd commented 2 weeks ago

Thanks for the prompt response MaanavD. Yes, above suggestions looks good to me. Mostly just trying to help users like me that miss the installation matrix. And I figure the links to pre-built wheels could be useful to others that come across this.