New report details per-dimension forgetting, selective depth retrieval, and a 16-of-896 MoE
Ahmad_Al_Dahle · x · 2026-07-21
A reply highlights three architectural ideas in a new technical report:
- KDA replaces Gated DeltaNet’s single scalar decay with learned per-dimension forgetting.
- AttnRes selectively retrieves across depth instead of weighting every residual layer equally.
- LatentMoE activates only 16 of 896 experts, with routing balanced by quantiles of router scores.
The post frames these as fixes for common one-size-fits-all design choices in linear attention, residual weighting, and expert balancing.
Related event: Moonshot Releases 2.8T Open-Weights Model Kimi K3(14 posts)→
More from Models
- NVIDIA says Nemotron 3 Ultra scored 30/42 on the 2026 IMO problems — NVIDIAAI · 2026-07-22
- OpenAI is reportedly briefing U.S. lawmakers on its next model family — kimmonismus · 2026-07-22
- Muse Spark 1.1 lands at 1495 on Text Arena with standout agentic-coding price performance — ycombinator · 2026-07-22
- Advanced AI Models Are Becoming Impossible to Plug and Play — emollick · 2026-07-22
- Google Gemini's AI Problem: No Leading Model for Core Workloads — bindureddy · 2026-07-22
- Model Offers 1M Token Context Window at Just $0.33/1M Tokens — MickeySteamboat · 2026-07-22