Re:Google calls Gemma 3 the most powerful AI model you can r...
by thot_experiment on Thursday March 20, 2025 @09:29PM
(#43430985)
Are you running vision tasks? Works just fine in ollama as an LLM but the vision component is unimplemented.
Re:Google calls Gemma 3 the most powerful AI model you can r...
by pitzips on Thursday March 20, 2025 @10:28PM
(#43431258)
I'm using gemma3:27b with the latest ollama (0.6.2) and vision is working.
Re:Google calls Gemma 3 the most powerful AI model you can r...
by thot_experiment on Thursday March 20, 2025 @10:46PM
(#43431342)
Yes but we're talking about mistral 3.1 here, which doesn't have the vision component implemented, looks like vllm supports it though.
Re:Google calls Gemma 3 the most powerful AI model you can r...
by simonw on Friday March 21, 2025 @12:31PM
(#43437709)
Mistral 3.1 vision works with MLX and mlx-vlm - https://github.com/Blaizzy/mlx-vlm/pull/256
https://huggingface.co/mlx-community/Mistral-Small-3.1-24B-I...
I've managed to run Mistral 3.1 on my laptop using MLX, notes here https://simonwillison.net/2025/Mar/17/mistral-small-31/