Architecture deep-dive: decoder periodically reads final encoder states instead of cross-attention
stochasticchasm · x · 2026-09-11
- A close reading of a model architecture blog corrects an initial assumption: the decoder cannot access encoder KVs at multiple layers, instead it periodically reads the final encoder hidden states
- The author argues this makes sense: final states are the richest and the design echoes the original Transformer
- Unlike plain cross-attention, the decoder attends to encoder states in the context of its own sequence so far — the author deems it a good choice, likening it to MoDA-style approaches
Related event: Leaked DeepSeek V4.1 Flash Tech Report Reveals KV Cache-Obsessed Flagship(44 posts)→
More from Research
- Researchers identify "Forking": train loss falls while validation loss turns up — ZimingLiu11 · 2026-10-26
- Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
- FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
- LEMON: zero-shot LLM misbehavior monitor flags hallucinations from attention maps — yoavartzi · 2026-10-26
- Behind the EMNLP keynote: LMs as simulated learners to generate child language acquisition hypotheses — najoungkim · 2026-10-26