Skip to content

Migrate to building using ROCm multiarch wheels#21

Draft
superm1 wants to merge 1 commit into
lemonadefrom
superm1/venv-wheels
Draft

Migrate to building using ROCm multiarch wheels#21
superm1 wants to merge 1 commit into
lemonadefrom
superm1/venv-wheels

Conversation

@superm1

@superm1 superm1 commented Jun 22, 2026

Copy link
Copy Markdown
Member

Tarballs will be going away.

@superm1 superm1 force-pushed the superm1/venv-wheels branch 2 times, most recently from 0e9a77c to 67e9d6b Compare June 22, 2026 16:53
@superm1 superm1 force-pushed the superm1/venv-wheels branch from 67e9d6b to 9c3e4be Compare June 22, 2026 17:09

@jeremyfowers jeremyfowers left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We cannot ship python inside of llamacpp.

@superm1

superm1 commented Jun 22, 2026

Copy link
Copy Markdown
Member Author

We cannot ship python inside of llamacpp.

There is no python shipped. It's a vehicle for delivering the SDK at build time. ROCm still needs to be installed separately.

@superm1 superm1 requested a review from jeremyfowers June 22, 2026 21:09
@superm1

superm1 commented Jun 22, 2026

Copy link
Copy Markdown
Member Author

Something I would like to understand - can we always install latest and have an ABI guarantee? If so - we should be able to drop the "7.13" designation as well.

@superm1 superm1 marked this pull request as draft June 22, 2026 21:10
@superm1

superm1 commented Jun 22, 2026

Copy link
Copy Markdown
Member Author

Some stats:

  • Using multi-arch wheels that just get libraries, headers and compiler we download 886.4 MB.
  • Using tarball to download the GFX1151 snapshot we download 3256.55 MB.

The actual artifact sizes do not change in any tangible way (they are remaining 212MB).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants