Port ML models into mflux/MLX with correctness-first validation, then refactor toward mflux style.
Provide a repeatable, MLX-focused workflow for porting ML models (typically from diffusers repo located near mflux repo in the system) into mflux with correctness first, then refactor to mflux style.
mx.compile, kernel fusion tweaks, scheduler micro-optimizations); add optimizations only after correctness is locked.z_image_turbo and flux2_klein for structure and naming.mx.compile and other speed-focused changes until deterministic parity is passing.mx.save/mx.load at critical points; it is OK to add these to the reference (without changing logic) to export latents.prompt → text_encoder → transformer_loop → VAE → image.MFLUX_PRESERVE_TEST_OUTPUT=1 uv run <test command>.CallbackManager.register_callbacks(...) so shared features like --stepwise-image-output-dir work; pass a latent_creator that supports unpack_latents(...).ModelConfig entry + aliasesuv for running scripts and tests: uv run <command>.uv run python -m <module> for local modules.