653e673ee3
- Add MPS device detection alongside CUDA - Add torch.mps.empty_cache() for memory cleanup - Add torch.mps.manual_seed() for reproducibility - Auto-disable 4-bit quantization on non-CUDA with warning - Fix HeartCodec hardcoded device="cuda" default - Fix .type() -> .to(dtype=) for MPS compatibility - Fix setup_caches context manager for MPS - Detect actual model device after loading Co-Authored-By: Kevin Gabeci <kevin@kevingabeci.com> Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>