Moondream, Qwen and Gemma ported to B200, developer says
The developer says the ports use megakernel infrastructure to run fast on B200 and asks which cards and models users want supported next.
TLDR
A developer says they've ported Moondream, Qwen and Gemma—all previously supported models—to run fast on B200 using megakernel infrastructure. They're inviting requests for additional cards and models to support.
The developer also describes support for both streaming and offline transcription, linking to Moondream's transcription documentation.
Combined views
469
1 Source, first seen 28d ago