Feature Request: mlx-swift-lm 3.x is out, more clean api(decoupled from huggingface hubapi), will Conduit consider a upgrade?

[z-also]

hi @christopherkarani，mlx-swift-lm just bumped a 3.x main version, decoupling some of its api from huggingface hubapi, making its api more elegant and flexible for local model loading, is there any plan to upgrade mlx-swift-lm to 3.x in Conduit?

• https://github.com/ml-explore/mlx-swift-lm#breaking-changes
• Decouple from tokenizer and downloader packages ml-explore/mlx-swift-lm#118

of course, not an urgent thing, but nice to have~

[z-also]

I am happy to contribute but still lack a comprehensive understanding of the codebase yet, seems claude code at your hand would certainly do the hard work

[christopherkarani]

Thanks for flagging this @z-also! Yes, upgrading to mlx-swift-lm 3.x is on the radar. The decoupled Hub API is a welcome change — it'll simplify our model loading paths.

Not urgent but definitely planned. If you'd like to take a crack at it, the main touchpoints are in Sources/Conduit/Providers/MLX/ — MLXProvider.swift, MLXModelLoader.swift, and MLXModelCache.swift. Happy to review a PR if you get to it before we do.