Sparse attention without dense warmup signals midtraining absorption, analyst argues
stochasticchasm · x · 2026-09-11
Blogger stochasticchasm reads sparse attention training progress: when a sparse variant no longer needs a dense-attention warmup, it's a good sign. Long-context pretraining is likely enabled by lower prefill FLOPs, and agentic traces can probably be introduced much earlier in training — suggesting midtraining is being absorbed. The thread also notes engram making it into the model, an mHC simplification (mega-mHC), and speculation about dspark use during pretraining pending the paper.
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