DiffusionGemma does not load in Unsloth Studio on Mac yet
A Mac with an M5 Pro chip and 64 gigabytes of memory failed to load DiffusionGemma through Unsloth Studio. The update script ran and the model was downloaded, but loading stopped with a message saying the model is not supported yet. The error says llama.cpp does not support the GGUF file’s model architecture, diffusion-gemma, so it cannot run through llama-server.
The model file appears to be valid, but the current runtime cannot handle this model type. Unsloth documentation seems to suggest support, so it is unclear whether Mac support is missing, whether the bundled llama.cpp is behind, or whether a separate update is needed.
Key points
- A Mac with an M5 Pro chip and 64 gigabytes of memory could not load DiffusionGemma.
- The downloaded GGUF file appears valid, but its diffusion-gemma architecture is not supported by the runtime.
- llama.cpp currently cannot run this model type through llama-server in this setup.
- Unsloth documentation may look more optimistic than the actual Mac experience.
- Before adding a new model to a Mac mini server, confirm that the local runtime supports it.
Quick term guide
- DiffusionGemma
- An AI model or experiment name, but the item does not give enough detail to define it fully.
- Unsloth Studio
- A tool used to download and run AI models.
- model architecture
- The underlying design and structure of an AI system that determines how it learns and responds.
- Architecture
- The overall structure and organization of a software project.
- llama-server
- A tool that runs Llama-style AI models as a server other programs can call.
- documentation
- Written notes that explain how a task or process is done.
- local model
- An AI model you run directly on your own computer, with no internet connection or external service needed.
- Mac mini server
- A Mac mini used as an always-on computer for files, apps, backups, or automation.