self-hosted/ai
§01·model · /models

Voxtral Mini 3B

multimodalactiveApache-2.0

3B audio-and-text multimodal by Mistral AI — speech-to-text, translation and audio Q&A. Audio goes in and text comes out, so it is not a speech synthesiser despite the name-adjacent company it keeps. About 9.5 GB in bf16/fp16, and fifteen cards carry a recipe from a 12 GB floor, both Radeon boards included under ROCm. Apache-2.0.

§02·GPUs that run this model
15 total

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