Skip to content

Conversation

@takanotaiga
Copy link

@takanotaiga takanotaiga commented Nov 20, 2025

Description

Previously, the project relied on a mamba-based environment setup, which lacked reproducibility and made dependency management harder to maintain.
Additionally, inference on NVIDIA Blackwell GPUs was not supported.

This PR introduces the following improvements:

  • Migration to a uv-based environment setup for improved reproducibility and faster, more reliable environment creation.
  • Upgrade to CUDA 12.9 and PyTorch 2.8, enabling inference support on the latest NVIDIA Blackwell architecture.
  • Documentation updates to reflect the new build requirements, including the note that local CUDA installation is required for building.

Testing Environment

The updated environment and Blackwell-compatible inference pipeline were validated on the following hardware:

  • NVIDIA RTX 6000 PRO Blackwell GPUs

@meta-cla meta-cla bot added the CLA Signed This label is managed by the Meta Open Source bot. label Nov 20, 2025
@takanotaiga
Copy link
Author

@gleize
Could you please review this?

@takanotaiga takanotaiga changed the title Add uv-based environment setup and Blackwell inference support Add uv-based environment setup and Blackwell GPU inference support Nov 20, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed This label is managed by the Meta Open Source bot.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant