facebookresearch / facebookresearch/sam3

Is CUDA 12.8 a hard requirement, or just the dev environment default?

Open
#596 0 comments 4 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
11.7k
Forks
1.8k
PR merge metrics
No merged PRs in 30d

Description

Hi, thanks for releasing SAM 3!

I noticed that the installation instructions pull `flash-attn-3` from the cu128 wheel index, which effectively pins the setup to CUDA 12.8.

Due to infrastructure constraints on my server, I need to reproduce the results with CUDA 12.1 instead.
Before I start debugging dependency issues, I wanted to ask:

- Is CUDA 12.8 required because of a measured performance difference (e.g., a flash-attn-3 feature that isn't available/stable on 12.1),
or is it simply the environment the model was developed and tested on?

Any guidance would be appreciated. Thanks!

Contributor guide

Open the contributing guide

Research direction

Start with the installation instructions and the flash-attn-3 cu128 wheel reference mentioned in the issue. Check the project's stated CUDA support and whether the dependency is tested with CUDA 12.1. Done means the documentation clearly says whether CUDA 12.8 is required and what users on CUDA 12.1 should expect.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
documentation, machine-learning
Issue type
Documentation
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.