{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreifkt2uufursd3cmm5a6c6vely333sw7m3np46caejgkcezflsrrsu",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mgsqadwlch72"
  },
  "path": "/t/open-ai-box-universal-llm-introspection-injection-points-dimension-roles-in-any-model/174198#post_1",
  "publishedAt": "2026-03-11T19:10:26.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "Open AI Box - a Hugging Face Space by anzizdaouda0",
    "openaibox · PyPI",
    "GitHub - Tryboy869/openaibox: Universal LLM introspection. Open AI Box — understand any model. · GitHub"
  ],
  "textContent": "Hey everyone\n\nI’m releasing **Open AI Box** — a Python package that opens the black box of any LLM.\n\n## What it does\n\nBy tracing a live inference pass with PyTorch forward hooks, it automatically discovers:\n\n- **Injection points** — where data enters, where decisions are made, where memory lives\n\n- **Dimension roles** — which dimensions carry causality, emotion, certainty, temporal reasoning\n\n- Everything exported to a single `graph.json`\n\n## Live demo\n\nOpen AI Box - a Hugging Face Space by anzizdaouda0\n\nPaste any model ID → get a full architecture analysis with interactive Plotly graphs.\n\n## Install\n\npip install openaibox\n\n```python\n\nfrom openaibox import OpenAIBox\n\noaib = OpenAIBox(“HuggingFaceTB/SmolLM-360M”)\n\noaib.discover().map_dimensions().export(“graph.json”)\n\noaib.print_summary()\n\nKey finding: the model.norm → lm_head control point is universal across all tested architectures (LlamaForCausalLM, Qwen2ForCausalLM, GPT2…).\n\nPyPI: openaibox · PyPI\n\nGitHub: GitHub - Tryboy869/openaibox: Universal LLM introspection. Open AI Box — understand any model. · GitHub\n\nHappy to hear feedback!",
  "title": "Open AI Box – Universal LLM introspection: injection points & dimension roles in any model"
}