Using jundot/Qwen3.6-27B-oQ8-mtp on Huggingface and 0.3.9dev1, there is no vision capability in omlx chat or as an API, even when I choose auto or vlm.
This is not just the MTP model, when using an older LLM of the 27B without MTP, the vision ability has disappeared.
Using jundot/Qwen3.6-27B-oQ8-mtp on Huggingface and 0.3.9dev1, there is no vision capability in omlx chat or as an API, even when I choose auto or vlm.
This is not just the MTP model, when using an older LLM of the 27B without MTP, the vision ability has disappeared.