Skip to content

Update Docker Script#1817

Open
2vibeornot2vibe wants to merge 1 commit into
deepbeepmeep:mainfrom
2vibeornot2vibe:patch-1
Open

Update Docker Script#1817
2vibeornot2vibe wants to merge 1 commit into
deepbeepmeep:mainfrom
2vibeornot2vibe:patch-1

Conversation

@2vibeornot2vibe

@2vibeornot2vibe 2vibeornot2vibe commented May 22, 2026

Copy link
Copy Markdown

What changed:

  1. Auto detects package manager to accommodate for more distros (assuming docker is already installed and configured)
  2. Skips docker build if the image already exists, allowing offline operation
  3. Changed NVIDIA runtime test to runtime version 12.8.1, since 12.4 is not on docker.io anymore
  4. Mount cache dir to /tmp (tmpfs-backed on newer distros) instead of ~/.cache for speedups
  5. HuggingFace connectivity check, auto-switch to HF-Mirror if failed
  6. Uses sdpa attention if CUDA Arch<=7.0 since sage's main branch will not work

(PS: it's just some quick and dirty fixes with the help of gemma-4 to make it work on my Arch setup, more improvements are definitely needed for this script)

What changed:
1. Auto detects package manager to accommodate for more distros (assumimg docker is already installed and configured)
2. Skips docker build if the image already exists, allows for offline operation
3. Changed NVIDIA runtime test to runtime version 12.8.1, since 12.4 is not on docker.io anymore
4. Mount cache dir to /tmp (tmpfs-backed on newer distros) instead of ~/.cache for speedups
5. HuggingFace connectivity check, auto-switch to HF-Mirror if failed
6. Uses sdpa attention if CUDA Arch<=7.0 since sage's main branch will not work

(PS: it's just some quick and dirty fixes with the help of gemma-4 to make it work on my Arch setup, more improvements are definitely needed for this script)
@deepbeepmeep

Copy link
Copy Markdown
Owner

thx

  1. you removed the former docker script, does the new one cover at least what the other one did ?
  2. why is the support for cuda 12.8 and not for cuda 13 / pytorch 2.10 (official current wangp config)

@2vibeornot2vibe

2vibeornot2vibe commented May 30, 2026

Copy link
Copy Markdown
Author

thx

1. you removed the former docker script, does the new one cover at least what the other one did ?

2. why is the support for cuda 12.8 and not for cuda 13 / pytorch 2.10 (official current wangp config)
  1. Yes it does. I tried to change the name of the script (because it now covers more than just debian-based distros), which resulted in the original script being removed.
  2. The original script uses an even older version (12.4) to ensure compatibility with pre-Turing cards, as CUDA 13 dropped support for those. Also the original script checks for CUDA 12.4 runtime every time on startup, but a later merged PR for the Dockerfile updated that to 12.8 with pytorch 2.10, causing a version mismatch. This is just a quick fix to make that check pass again :)

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