{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreihatljzmcaqof2w36p5ierwbc66dzfr53rxuflxyxcyo4wqr4ua7a",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mmjf3geaojz2"
  },
  "path": "/t/why-qwen-is-dynamically-stable-an-empirical-phase-map-of-10-llms/176177#post_1",
  "publishedAt": "2026-05-23T11:09:22.000Z",
  "site": "https://discuss.huggingface.co",
  "textContent": "I’ve been analyzing the internal trajectory dynamics of several open-source models (Qwen, Llama, Gemma, etc.) using a new metric called ct_t (local trajectory instability based on hidden state curvature).\nThe results show a clear structural difference in how these models process information internally.\nKey Finding:\nWhile models like Gemma-2B often drift into a “Chaotic” regime (high instability spikes) and Llama-3.2 into an “Underactive” one (rigid, low variance), Qwen models consistently maintain an “Adaptive” regime.\nThis suggests that Qwen’s architecture achieves a unique balance between stability and flexibility, independent of model size. A 1.5B Qwen model was dynamically more stable than larger counterparts in our panel.\nThe 4 Regimes Identified:\nUnderactive: Rigid, low adaptability.\nAdaptive: Balanced flux/stability (Qwen’s zone).\nTransition: Boundary zone.\nChaotic: High instability, prone to divergence.\nFull Technical Report:\nI’ve published the full working paper with data from 158 runs on Zenodo:\nFour Dynamical Regimes in Large Language Models: An Empirical Phase Map\nI’d love to hear your thoughts: Does this “Adaptive” dynamic correlate with your experience of Qwen’s reasoning capabilities?\nBest,\nJean-Denis Bosange Batuli",
  "title": "Why Qwen is Dynamically Stable: An Empirical Phase Map of 10 LLMs"
}