ml-explore
Releases Roadmap?
Hello, is there a roadmap for future releases of MLX? I see there are a number of requests regarding features like missing operations (e.g. pooling/upsampling), quantized inference, profiling tools, a Swift API, and so on. Is there anywhere public these are prioritized?
I've been playing with MLX and have been enjoying using it for toy models and am attempting to implement StripedHyena currently. However, for anything more serious, I find it hard to justify using MLX over other options: PyTorch and JAX support MPS, are usable on other platforms, utilize Nvidia GPU's for larger jobs, have a substantially larger community. MLX also doesn't have any conversion support to ONNX (or ironically, CoreML).
I'm probably speaking for many people, but a roadmap would help immensely for determining how much time to invest in this framework, what kind of work would be best suited for it (e.g. I presume pre-training a foundation model is non-starter), and what the level of commitment there will be from Apple AI/ML.
