What's Changed
- Extend Float8_e4m3fn copy_() to handle non-FP8 → FP8 conversions on MPS by @Copilot in #3
- Fix FP8 e4m3fn encoding rounding to match PyTorch CPU by @Copilot in #9
- Implement spatial tiling for VAE decode to handle MPS INT_MAX tensor limits by @Copilot in #11
- Fix VAE decode patch kwargs compatibility with ComfyUI by @Copilot in #13
Full Changelog: https://github.com/audiohacking/fp8-mps-metal/commits/0.0.2