self-hosted/ai
§01·model · /models

MiniMind-O

multimodalactiveApache-2.0

0.1B omni model trained from scratch by jingyaogong — text, speech and image, in and out, in something small enough to read end to end. It is educational and research-grade rather than production, and its roughly 4 GB envelope means memory is never the question: nineteen cards carry a recipe here, from the 8 GB tier upward, both Radeon boards included. Apache-2.0.

§02·GPUs that run this model
19 total

benchmarked·~ runs via recipe (not benchmarked)· untested·doesn't fit