New model ships Day-0 in Diffusers with novel KV-caching and disentangled prefill+decode
RisingSayak · x · 2026-09-20
RisingSayak announced that a new model is available Day-0 in Hugging Face Diffusers, thanking a community collaborator for the PR. He highlights the model's interesting architecture: a novel approach to KV-caching plus a disentangled prefill + decode design, and says he'll do a separate thread explaining the details.
More from Models
- Raschka: with $100M for a top LLM, spend it all on post-training, not pretraining — MaziyarPanahi · 2026-09-20
- Raschka: with $100M to build an LLM, I'd skip pretraining and invest in post-training — rasbt · 2026-09-20
- Leaked Claude Opus 5.5 Pricing Shows $0.20/Mtok Cache Read, Sparking Hype — xeophon · 2026-09-20
- Rumors: Jev model tackled bounded prime gaps, FLT formalization and Navier-Stokes in a week — f_charton · 2026-09-20
- Jev rejects its 'System One model' label, prefers to be called a Decision model — IanArawjo · 2026-09-20
- Agent swarms could be why DeepSeek cuts prices again: attention was the usage bottleneck — teortaxesTex · 2026-09-20