nerdola-de-cartola opened issue #10153:
Feature
Add GPU support for wasi_nn with PyTorch backend
Benefit
Improve performance over nn operations
Implementation
The PyTorch backend uses tch, which is a PyTorch wrapper for rust, so the basic implementation would be to add cada as a compilation target when compiling the tch crate.
nerdola-de-cartola commented on issue #10153:
If you are a contributor to wasmtime/wasi_nn please leave your comments and sugestions. This is my first contribution in this project, so and help would be nice.
nerdola-de-cartola commented on issue #10153:
I created a fork that solves the issue, at least for me. Maybe is interested for some one else. My fork
abrown commented on issue #10153:
@nerdola-de-cartola, thanks for looking into this. Looking at the top commit from your fork, I think the change makes sense to me (at a glance!). Are you willing to create a PR here that I can review?
nerdola-de-cartola commented on issue #10153:
Yes, for sure, just tell me what I need to do
abrown commented on issue #10153:
Can you just create a PR from that fork you reference above? (If you're coming from a version control environment other than GitHub, these instructions may help?).
nerdola-de-cartola commented on issue #10153:
PR 10204 crated! Thanks for the guidance.
nerdola-de-cartola closed issue #10153:
Feature
Add GPU support for wasi_nn with PyTorch backend
Benefit
Improve performance over nn operations
Implementation
The PyTorch backend uses tch, which is a PyTorch wrapper for rust, so the basic implementation would be to add cada as a compilation target when compiling the tch crate.
nerdola-de-cartola commented on issue #10153:
PR 10204 Merged into main, closing issue
Last updated: Apr 17 2025 at 00:13 UTC