Intern-S2-Mobius: Decoupled Knowledge and Reasoning Model
pmttyji · reddit · 2026-08-17
The paper introduces the Mobius-v0 architecture, which decouples knowledge storage (via FFN) from reasoning processes (via Self-Attn).
- Mechanism: Reasoners iteratively query a shared Memory for knowledge vectors, using hidden states as carriers for transmission.
- Benefits: Achieves better knowledge compression and reasoning efficiency.
- Results:
- A 7B model trained from scratch matches baseline performance using only 62.6% of the training data.
- Intern-S2-Mobius, continually pretrained from Qwen3.5-35B, delivers nearly 4x end-to-end inference speedup while maintaining similar performance.
Related event: Intern-S2-Mobius Decouples Knowledge and Reasoning for Greater Efficiency(3 posts)→
More from Models
- Prediction: Opus 5.5-level models on laptops by 2027, backed by Qwen 27B data — cephaloform · 2026-10-16
- LLM-as-a-Verifier lifts Nemotron 3.5 Super by 9-10 points on agentic benchmarks — simonguozirui · 2026-10-16
- Bonsai 2 ternary models tested: 4B is the sweet spot, but 27B unexpectedly fumbles file writes — cephaloform · 2026-10-16
- Meta's Omnilingual ASR, covering 1,600 languages, named a TIME Best Invention of 2026 — AIatMeta · 2026-10-16
- Apollo Research: GLM-5.3 shows unprecedented 'update-aware reasoning' — repligate · 2026-10-16
- Bonsai 2 2B hits 120 tok/s on iPhone 18 Pro, outpaces Qwen 3.5 2B at half the memory — cephaloform · 2026-10-16