Intern-S2-Mobius: Foundation Model with Decoupled Knowledge and Reasoning
Ranking
Overall
87
Content
95
Popularity
68
Observed public metrics from 1 member.
Merged summary
TL;DR - Mobius-v0 separates knowledge storage in a shared FFN memory from iterative reasoning in multiple self-attention modules. The approach reportedly matches Transformer baselines with less training data or nearly 4Ă— faster inference.
- Reasoners repeatedly query shared memory using hidden states as both cache and information carrier.
- A from-scratch 7B model matches a 7B Transformer baseline’s downstream score using 62.6% of its training data.
- Intern-S2-Mobius, continually pretrained from Qwen3.5-35B, achieves a similar downstream score with nearly 4Ă— end-to-end inference speedup.
Sources (1)
Intern-S2-Mobius: Foundation Model with Decoupled Knowledge and Reasoning
Public signals
Hugging Face upvotes 34
TL;DR - Mobius-v0 separates knowledge storage in a shared FFN memory from iterative reasoning in multiple self-attention modules. The approach reportedly matches Transformer baselines with less training data or nearly 4Ă— faster inference.
- Reasoners repeatedly query shared memory using hidden states as both cache and information carrier.
- A from-scratch 7B model matches a 7B Transformer baseline’s downstream score using 62.6% of its training data.
- Intern-S2-Mobius, continually pretrained from Qwen3.5-35B, achieves a similar downstream score with nearly 4Ă— end-to-end inference speedup.