- New --flash flag enables Flash Attention 2 installation
- Works with nvidia and all backends (when CUDA available)
- Installs with --no-build-isolation flag
- Graceful error handling if installation fails
- Updated usage instructions to show --flash-attn flag
- Requirements: CUDA 11.6+, Linux, Ampere/Ada/Hopper GPU

- Complete overview of refactoring work
- Architecture before/after comparison
- List of all commits and changes
- Installation and usage instructions
- Configuration examples
- Testing checklist and known limitations
- Next steps for future work

- Added PIP_NO_INPUT and PIP_REQUIRE_VIRTUALENV environment variables
- Better error handling for problematic packages (procname, stable-diffusion-cpp-python)
- Install core dependencies first, then optional packages individually
- Graceful fallback when optional packages fail to build
- Updated PyTorch installation to handle version compatibility
- Clear warnings for optional packages that fail (not critical errors)

- Removed version pins from stable-diffusion-cpp-python and procname
- Both packages have build issues with strict version requirements on Python 3.13
- Allowing pip to find compatible versions automatically