Missing support for Orion model architecture

[aspwow]

Platform: macOS 26.6.2, Apple Silicon (M5)
Python: 3.11

Hi,
Orion-14B-LongChat was officially released by OrionStarAI in 2023, making it one of the relatively early long-context multilingual open-source large language models in the industry.

As a pioneering 14B-parameter model that supports a context window of up to 320,000 tokens, it has been publicly available on Hugging Face for a long time, and its multilingual capabilities covering Chinese, English, Japanese and Korean have been fully verified by the open-source community.

I believe it would be great to support this model in MLX so that it can be used with this framework as well.

Please support for Orion models,Thanks a lot!
https://huggingface.co/OrionStarAI/Orion-14B-LongChat
https://huggingface.co/OrionStarAI/Orion-14B-Chat

More details:
https://huggingface.co/OrionStarAI

[manuhalapeth]

I'd like to work on this: I'll add support for the Orion architecture (following the porting steps in CONTRIBUTING.md: checking the safetensors weights, adding a new orion.py model file under mlx_lm/models/, and matching the model_type in config.json). Will open a PR once I have something working.