Qwen layer analysis shows next-token agreement rising sharply after layer 20
Sauers_ · x · 2026-07-21
A small research note on Qwen’s internal representations, illustrated with a chart of next-token agreement across layers.
- The image highlights an “analysis band” around layers 12–29.
- Agreement stays near zero in earlier layers, then rises sharply in later layers and crosses a noted motor-onset threshold.
- The post is pointing to a possible relationship between a model’s internal dynamics and motor-like behavior in the J-space representation.
More from Models
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11
- giffmana: the env being used in training is part of the point — giffmana · 2026-09-11
- awesome-llm-leaderboards: an open-source directory of LLM leaderboards, pricing tables, comparison tools — Last_Establishment_1 · 2026-09-11
- Anthropic claims it works to keep eval environments unidentifiable to models — MaxKannen · 2026-09-11
- Nex N2.5 Pro released on Hugging Face with 407GB of weights — jinnyjuice · 2026-09-11
- RoMa v2 image matching model unveiled in the usual black poster — ducha_aiki · 2026-09-11