Skip to content

Experimental build for AMD GPUs/ROCm #290

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 58 commits into
base: release
Choose a base branch
from
Open

Experimental build for AMD GPUs/ROCm #290

wants to merge 58 commits into from

Conversation

@kprinssu
Copy link

@kprinssu kprinssu commented Apr 14, 2025 •
edited
Loading

I am hoping to get this PR merged in as in it contains a few changes.

This PR contains the following major changes:

  • split up GitHub builds into a matrix
    • docker bake will now use specific groups (cpu, gpu (cuda), rocm (AMD))
    • use a build matrix strategy to parallelize and split up the builds
  • add experimental support for ROCm support (tested against 6.3.x and 6.4.x, PR will support 6.4.1)
  • upstream changes from master into release

bgs4free and others added 14 commits January 19, 2025 16:49
Add experimental rocm support
b064066 
Tested with rocm 6.3.1.

Known issues with this commit:
 * The generated docker image becomes quite big
 * Suppressed MIOPEN warnings, which might not be a good idea
 * No windows support, at least not tested
 * Incomplete documentation
 * Not thoroughly tested
@kprinssu
Fix merge conflicts
1d01755
@kprinssu
Update Dockerfile to build against newer code
aba2628
@kprinssu
Merge branch 'remsky:master' into master
a665cfc
@kprinssu
Use my fork of Kokoro
4f149ab
@kprinssu
Merge branch 'master' of github.com:kprinssu/Kokoro-FastAPI
3db522b
@kprinssu
Merge branch 'remsky:master' into master
2185b1b
@kprinssu
Merge branch 'remsky:master' into master
e9d9921
@kprinssu
Add notes about performance inside docker compose
19ac239
@kprinssu
Revert to upstream Kokoro
9e595d9
@kprinssu
Remove casual_conv1d
94e5305
@kprinssu
Fix merge conflicts
b19fd1d
@kprinssu
Update to ROCm 6.4 and use uv to install dependencies
105b96e
@kprinssu
Merge branch 'remsky:release' into release
e464b3a
kprinssu
kprinssu commented Apr 14, 2025
View reviewed changes
kprinssu
kprinssu commented Apr 14, 2025
View reviewed changes
kprinssu
kprinssu commented Apr 14, 2025
View reviewed changes
@kprinssu
Copy link
Author

kprinssu commented May 6, 2025 •
edited
Loading

@remsky or @fireblade2534 Can I get your eyes on this PR please?

@kprinssu kprinssu changed the base branch from master to release June 29, 2025 18:23
@kprinssu
Copy link
Author

Hi @fireblade2534

It looks like you are quite active here on this repo, would you be able to do a quick PR review?

kprinssu and others added 21 commits June 29, 2025 11:24
@kprinssu
Make myself the repo owner
ad65800
@kprinssu
Enable builds for master
fae42d3
@kprinssu
Revert "Make myself the repo owner"
2ad4467 
This reverts commit ad65800.
@kprinssu
Minor revert to clean up disk space step
1d6d455
@kprinssu
Increase build step timeout
130e6d9
@kprinssu
Split arm64 and x86_64 builds
726803a
@kprinssu
Update pipelines to also build arm64
1aaf004
@kprinssu
Fix arm64 cpu build target
92505e3
@kprinssu
Revert changes to pyproject toml to use pytorch nightly
70b9521
@kprinssu
Revert changes to pyproject toml to use pytorch nightly
7e9108e
@kprinssu
Split up arm64 builds into their own targets
67ac35f
@kprinssu
Bump version to 0.2.4-1
812a72d
@kprinssu
Reduce bloat in ROCm Dockerfile
84f5cc4
@fireblade2534
Merge pull request remsky#354 from MiggiV2/master
29066f7 
Update PyTorch to 2.7.1+cu128 for RTX 50 Series GPU Support
@faltiska
Smaller docker image:
acc671f 
CPU image 5.6Gb -> 4.9Gb
GPU image 14.8Gb -> 9.9Gb

Saved 4GBby using the cuda base image instead of cudnn-runtime. Saved 700MB by deleting the UV cache.

Improved the speed of rebuilding the image by merging the RUN commands into one.
@fireblade2534
Merge pull request remsky#361 from faltiska/smaller-docker-image
bbb6ab6 
Smaller docker image
@kprinssu
Merge branch 'remsky:master' into master
ef20f78
@kprinssu
Update Dockerfile to ROCm 6.4.3 and use newly released Pytorch 2.8
5528a5a
@kprinssu
Install missing rocrand package
dc2601b
@kprinssu
Fix merge conflicts
7e6d339
@kprinssu
Clean up pyproject with duplicate indices
b3a3af6
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@kprinssu @fireblade2534 @MiggiV2 @faltiska