DiffusionGemma rewrites text generation as denoising
Google’s open 26B MoE model generates text in parallel, trading top quality for local speed above 1,000 tokens per second on H100.
AI technology News and Analysis
1 verified stories covering AI technology, product updates and industry developments.