mirror of
https://github.com/hanzoai/live.git
synced 2026-10-30 18:29:41 +00:00
This commit adds full support for running Hanzo Live on macOS with Apple Silicon (M1/M2/M3/M4) using MLX and Metal backend. Key changes: - Add --mlx flag to enable Apple Silicon acceleration - Implement device detection for MLX/Metal/MPS backends - Fix CUDA hardcoding issues in T5 encoder models - Update device handling to work with CUDA, MPS, and CPU - Add MLX dependency for macOS in pyproject.toml - Update README with macOS installation and usage instructions - Fix flash_attn import errors on non-CUDA systems Tested pipelines: - ✅ Passthrough pipeline - ✅ LongLive pipeline - ✅ StreamDiffusionV2 pipeline All pipelines now work correctly on macOS with MLX mode enabled via: uv run daydream-scope --mlx
Code in this folder is modified from https://github.com/NVlabs/LongLive which has a CC-BY-NC-SA 4.0 license.