{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreieypvtnqirrbwv6f5er7l6ohjoz224vwq7tpbdk7zfjrxceakyr7a",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mhm6rjozqdm2"
  },
  "path": "/t/need-arxiv-cs-lg-endorsement-autocompress-autonomous-llm-compression-independent-researcher/174476#post_1",
  "publishedAt": "2026-03-21T22:58:14.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "Log in to arXiv | arXiv e-print repository"
  ],
  "textContent": "Hi HuggingFace community!\n\nI’m Archit Thorat, an independent researcher from India.\nI need an arXiv cs.LG endorsement for my first paper.\n\nPaper: “AutoCompress: Autonomous Compression of Small\nLanguage Models via Critical Layer Isolation”\n\nKey finding: Layer 0 in small transformers carries ~98%\nof task-critical information — consistent across 6-layer\nand 16-layer models.\n\nThis motivated Critical Layer Isolation (CLI):\n\n  * Layer 0 runs at full capacity\n  * All other layers compressed to shared bottleneck dim\n  * Result: 34.8% compression matching baseline quality\n\n\n\nAll experiments done on free Colab T4 GPU overnight.\nZero paid compute.\n\nWould anyone be willing to endorse my submission?\n\nEndorsement link: Log in to arXiv | arXiv e-print repository\n\nThank you so much!\nArchit",
  "title": "Need arXiv cs.LG endorsement — AutoCompress: Autonomous LLM Compression (Independent Researcher)"
}