Fork CUDA and ROCm guides into separate pages. #15196
Conversation
ScottTodd
added
documentation
| Using MobileNet v2 as an example, you can download the SavedModel with trained | ||
| weights from | ||
| [TensorFlow Hub](https://tfhub.dev/google/tf2-preview/mobilenet_v2/classification) | ||
| and convert it using IREE's | ||
| [TensorFlow importer](../ml-frameworks/tensorflow.md). Then run one of the | ||
| following commands to compile: | ||
|
|
||
| ```shell hl_lines="2-3" | ||
| iree-compile \ | ||
| --iree-hal-target-backends=cuda \ | ||
| --iree-hal-cuda-llvm-target-arch=<...> \ | ||
| mobilenet_iree_input.mlir -o mobilenet_cuda.vmfb | ||
| ``` |
There was a problem hiding this comment.
These TF sections should be replaced with something more generic. Future work...
Tbh, both pages still looks quite similar to me for now haha, I do think it's good to start splitting them up for better legibility and UI/UX. :) Just curious, are there any other reasons you started splitting the page up? |
Yeah, I could go either way here. On one end, we could fold all of the GPU pages (Vulkan, CUDA, ROCm, Metal) into a single page, and on the other (this PR), we could split them into separate pages. We're currently somewhere in the middle, since CUDA and ROCm were both at very similar stages of (early) development. The main reason I want separate pages is for clear messaging to users that each backend can stand on its own and can provide differentiating features.
The implementations are (by design) quite similar in IREE, and some users may appreciate side by side comparisons, but I expect that many users will already know what hardware/APIs they are using and would rather see documentation specialized for that. |
|
SGTM! |
Fixes iree-org#15115 These pages are very similar, but both stacks have undergone enough development recently that they can now stand on their own. This PR primarily forks the pages, but it also does some minor cleanup (fixing dead links, adjusting formatting). | | | |--------|--------| | Current page | https://www.iree.dev/guides/deployment-configurations/gpu-cuda-rocm/ | | Preview of this PR - CUDA | https://scotttodd.github.io/iree/guides/deployment-configurations/gpu-cuda | | Preview of this PR - ROCm | https://scotttodd.github.io/iree/guides/deployment-configurations/gpu-rocm |
Fixes #15115
These pages are very similar, but both stacks have undergone enough development recently that they can now stand on their own. This PR primarily forks the pages, but it also does some minor cleanup (fixing dead links, adjusting formatting).