mlx-community

/

Ornith-1.0-35B-bf16

like 6

Follow

MLX Community 14.6k

Issue Summary

When attempting to load Ornith-1.0-35B-bf16 in LM Studio on my M5 Max (128GB unified memory), I encounter a fatal error in the MLX backend.

System Details

• Hardware: MacBook Pro M5 Max with 128GB unified memory
• Software: LM Studio with MLX backend
• Model: mlx-community/Ornith-1.0-35B-bf16

Error Message

ValueError: Unsupported model type: qwen3_5_moe_vision

The full stack trace indicates the error occurs in /mlx_vlm/models/qwen3_vl/vision.py at line 201 when trying to initialize the vision model.

Root Cause

The MLX VLM loader appears to not recognize the qwen3_5_moe_vision model type, even though the model is tagged with qwen3_5_moe and mlx-vlm tags.

Expected Behavior

The model should load successfully on hardware that meets its specifications (70GB storage, fits in 128GB unified memory).

Steps to Reproduce

1. Load the Ornith-1.0-35B-bf16 model in LM Studio on M5 Max
2. Attempt to generate a response

Please advise if:

• This is a known compatibility issue with the current MLX VLM version
• There's a workaround or configuration fix
• The model weights need to be reconverted

Thank you!