Improve setup for multi-GPU support and fix inference docs
Finnish-NLP
d15775e
- install_dependencies.sh: auto-detect GPU compute capability and
install PyTorch 2.10.0+cu128 for Blackwell (sm_120+) or
2.5.1+cu124 for older GPUs; add cuDNN library path fix
- requirements.txt: remove torch version pins (managed by install
script), add missing huggingface_hub/requests deps, clarify that
chatterbox TTS is vendored in src/
- devcontainer.json: fix name, fix wrong repo path, add git-lfs pull,
auto-clone from HF if starting from blank container, run full setup
in postCreateCommand
- README.md: add Quick Start section with GPU compat notes; fix
inference example (was using nonexistent ChatterboxMultilingualTTS,
now uses correct ChatterboxTTS from src.chatterbox_.tts)
