model for frigate, a380
Posted by mcgeezy-e@reddit | LocalLLaMA | View on Reddit | 4 comments
Hello,
I am looking for a small vision model that would work for the genai features on frigate. I may use it for a few home assistant things as well (I figured it would be simple stuff like "how many lights are on")
my video card is an a380
I have been able to get gemma4 e2b to run, though it feels quite a bit slow. I am open to other models to test.
Thank you
onyxlabyrinth1979@reddit
a380 is doable but you’re gonna feel the limits fast, especially with vision. I’d stick to smaller multimodal models and keep expectations tight on latency. Also watch how the model handles simple counting tasks, they can be weirdly inconsistent. Are you running everything local or mixing in any offload?
mcgeezy-e@reddit (OP)
I ended up with Qwen3-VL 4b and its running pretty well. It's compiling solid descriptions for the review videos. Its surprisingly quick. ran some chat against it and it handles basic task quite well, so I think it should be fine for any small task that I may want to use for HA.
I have another server that is loaded with a few video cards that i run other models on. However, i wanted something on the same system, that was small and workable and dedicated for this task.
Acrobatic_Stress1388@reddit
Try Qwen. Either the new ones or the qwen3 vl ones. I would try the smallest one you can find and move up from there. The ol' speed vs quality trade off.
mcgeezy-e@reddit (OP)
ya i was thinking of giving it a go.