Add docker config for ROCm #26
Open
Conversation
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
Owner
|
Nice work! I don't actually have an AMD card myself and cant test it so I think we will likely need a third person to test this then Ill happily merge (If no one tests it for awhile it lgtm so I reckon it can just be merged but id prefer a test if possible) |
Author
|
Noticed that dependencies have been updated so at the moment this won't run, I'll try bringing PR up-to-date. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
My nvidia cards were lacking in VRAM so got this up and running on 7900XTX, works ok-ish, could be faster (didn't do proper testing yet but logs had these sort of numbers: "Processed prompts: 5.96s, Total time: 14.05s, Generated 262 frames (20.96s of audio), Real-time factor: 0.670x" how does that compare to RTX cards? not sure where to look for bottlenecks)
Building vllm and finding compatible bitsandbytes and triton versions were the main challenges, hopefully there's no need to update them in a while. vllm docs said to disable FA so did, not sure what it's status is, could be nice to get working if possible.
For some reason building vllm from their most recent master (v0.8.5+) resulted in "only" 30.8GB image, didn't check why v0.6.6 would be 100GB but I guess +-70GB is no biggie these days.