{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreieebljpphzi7vnpar6xs5azdjv36zz4sk5lckoamfehi6bjsx7jwy",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mo2afx6pwu62"
  },
  "path": "/t/can-an-llm-lose-conceptual-continuity-while-remaining-coherent/176469#post_11",
  "publishedAt": "2026-06-11T21:14:08.000Z",
  "site": "https://discuss.huggingface.co",
  "textContent": "If I manage to solve the LITM problem, or to be better than SnapKV at least, I will share everything with you, if you are actually working on that epistemic layer.\n\nIf I fail, though, I will share the NIAH solution (the basic TIS) so that we can try to integrate it with your system but at a different level.\n\nThese are my baselines so far:\n\n### **6.2 LITM Results**\n\n**Cache Budget** | **Vanilla** | **StreamingLLM** | **H2O** | **SnapKV** | **Infini-Attn** | **TIS (oracle)**\n---|---|---|---|---|---|---\n100% | 100.0% | 100.0% | 100.0% | 100.0% | 100.0% | **100.0%**\n75% | 66.1% | 66.7% | 66.7% | **79.4%** | 47.8% | 66.1%\n50% | 43.9% | 33.3% | 44.4% | **55.6%** | 32.8% | 46.1%\n25% | 33.3% | 33.3% | 33.3% | 33.3% | 33.3% | 33.3%",
  "title": "Can an LLM lose conceptual continuity while remaining coherent?"
}