Conversation
dfgHiatus
changed the title
dfgHiatus
added
enhancement
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR has a bit of a story attached to it.
As of the latest release, inference on Windows is accelerated through the DirectML execution provider, and there is no acceleration on Linux. The AMD backends, ROCM and MiGraphX, do not have any NuGet packages available, and the former is deprecated. The effort to maintain our own execution providers is not worth it at this time.
On the flip side, CUDA has been a more interesting. For us to have proper CUDA support, we need to bundle a parts of the CUDNN runtime with the application, as the publicly available NuGet package does not do this by default (and hasn't for a while, by the looks of some things online). This process is literally dropping in a single missing file, as has been my experience on Windows. I can't speak for Linux, at least on WSL Ubuntu 24.04.
I learned about this after modifying how I was appending the execution provider to CUDA. Dropping
AppendExecutionProvider_CUDAfor the longer code block has made things work for me on a 5060ti and CUDA 12-13. This is the only card I have to reliably test against, I can't speak against other hardware.There are workarounds. This could be done in the actions build step, or we could copy the file over on the local user machine. These implementations appear fragile - This hasn't even factored in instructing the user to install compatible versions of CUDA and CUDNN on their system, build size, etc.
In the spirit of transparency, I'm going to push up what I have and see what others make and or think of it. Cheers, and good luck