AI & RoboticsMore in AI & Robotics→
Google's DiffusionGemma Generates Text from Noise, Up to 4x Faster on Single GPU
Key Takeaways
- Google releases DiffusionGemma, a 26B-parameter model generating text via diffusion.
- Produces 256 tokens in parallel, up to 4x faster on single GPU than autoregressive models.
- Nvidia optimized the model; achieves ~1000 tokens/sec on H100.
- Lower quality but suited for non-linear tasks like code gap-filling.
DE
DT Editorial Team··via the-decoder.com
