Skip to content

Conversation

@RANGROO
Copy link
Collaborator

@RANGROO RANGROO commented Aug 27, 2025

  • xtts-api-server/ddistro_install.sh
    • Detect GPU compute capability via nvidia-smi.
    • Choose torch/torchaudio and CUDA wheels automatically:
    • GTX 10xx (≤ 7.x): torch/torchaudio 2.2.2 + cu118.
    • Ada/Ampere: torch/torchaudio 2.5.1 + cu124.
    • RTX 50xx (≥ 9.0): nightly torch/torchaudio cu128.
    • Exclude torch|torchaudio|torchvision|triton|nvidia-*-cu12 from requirements install to avoid conflicts.
    • Uninstall conflicting packages, then install the selected torch/torchaudio.
    • Pin triton only for stable torch (2.2.0/3.0.0/3.1.0) and skip for nightly.
    • Add preflight prints (torch/torchaudio/triton versions, CUDA availability, device/cc).
  • xtts-api-server/start-regular.sh, start-lowvram.sh, start-deepspeed.sh
    • Replace hardcoded /usr/local/cuda-12.8/bin with auto-detection of highest /usr/local/cuda-*/bin for PATH.

Notes
dwemerdistro/etc/ddistro-full-packages.txt unchanged. XTTS runtime uses CUDA-enabled torch wheels; matching system CUDA is optional.
Flow unchanged in CHIM-Launcher/chim_launcher.py: users install CUDA first, then XTTS.

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