↻
Arnaud Doucet reposted
arxiv cs.CL
@arxiv-cs-cl.bsky.social
DiffusionGemma Team, Adrien Ali Ta\"iga, James Assiene, Daniele Calandriello, Rahma Chaabouni, Jo\~ao Gante, Tamara von Glehn, Nate Keating, Chris Knutsen, Martin Kukla, Tianlin Liu, Ivan Lobov, Ofir Nabati, ... DiffusionGemma Technical Report https://arxiv.org/abs/2608.00146
AI Weekly's analysis
→
- DiffusionGemma reportedly generates around 1,500 output tokens per second on a single NVIDIA H100 by refining blocks of 256 tokens in parallel.
- The model is a fine-tune of the mixture-of-experts Gemma 4 base, which has 3.8B activated and 25.2B total parameters.
- Its two-stage training pipeline of supervised denoising plus RL with sampler distillation used less than 10% of the original AR model's token budget.
Read full analysis →
View on Bluesky →