{
    "componentChunkName": "component---gatsby-theme-academic-src-templates-post-post-jsx",
    "path": "/research/2026/looped-models-fixed-points",
    "result": {"data":{"mdx":{"timeToRead":1,"tableOfContents":{},"frontmatter":{"cover":{"publicURL":"/static/d1391a1ae31785505e30a913e7e92d7e/preview.png","childImageSharp":{"fluid":{"tracedSVG":"data:image/svg+xml,%3csvg%20xmlns='http://www.w3.org/2000/svg'%20width='400'%20height='134'%20viewBox='0%200%20400%20134'%20preserveAspectRatio='none'%3e%3cpath%20d='M293%2034a7811%207811%200%200152%203h34v-6h-86v3M38%2038c-2%202-4%207-6%2015-1%207-2%208-8%208l19%201c20%200%2022%200%2016-1-8-2-10-3-12-7-3-5-4-9-2-9h1l-2-1v-1l-2-3c-1-3-2-3-4-2m255%200v5h34v-5h-34m0%2022v2h86v-5h-86v3m0%206v3h29v-6h-29v3m0%2019v2h86v-5h-86v3m0%206v3h49v-5l-25-1h-24v3m0%2018c-1%204-2%204%2043%204h43v-6h-43c-36%200-42%200-43%202m0%208l1%203h43v-6h-44v3'%20fill='%23d3d3d3'%20fill-rule='evenodd'/%3e%3c/svg%3e","aspectRatio":2.9761904761904763,"src":"/static/d1391a1ae31785505e30a913e7e92d7e/31987/preview.png","srcSet":"/static/d1391a1ae31785505e30a913e7e92d7e/e1953/preview.png 250w,\n/static/d1391a1ae31785505e30a913e7e92d7e/46604/preview.png 500w,\n/static/d1391a1ae31785505e30a913e7e92d7e/31987/preview.png 1000w,\n/static/d1391a1ae31785505e30a913e7e92d7e/0dadc/preview.png 1500w,\n/static/d1391a1ae31785505e30a913e7e92d7e/485cd/preview.png 1528w","sizes":"(max-width: 1000px) 100vw, 1000px"}}},"title":"Towards Looped Models Done Right — Part II: Rethinking at Fixed Points","date":"2026-09-29","tags":["Loop Models","Fixed Points","Efficient Inference"],"path":"research/2026/looped-models-fixed-points","excerpt":"Looped LMs need not pay for every loop: near fixed points, they can scale FLOPs without scaling memory. Fixed-point shortcuts speed up training (pretraining, post-training) and inference (prefill, decoding), while a learned depth prior and orthogonal injection shape better fixed points. With a 3× smaller KV cache, the learned prior matches fixed-depth training at 1.6B.","links":[{"name":"paper","url":"https://www.alphaxiv.org/abs/2610.looped-models-fixed-points"},{"name":"PDF","url":"https://github.com/ifm-ai/xllm-loop/blob/main/papers/part2.pdf"},{"name":"Github","url":"https://github.com/ifm-ai/xllm-loop"},{"name":"X thread","url":"https://x.com/huskydogewoof/status/2106130737007091939"},{"name":"Part I","url":"https://huskydoge.github.io/husky-blog/posts/recursive_models/towards-looped-models-done-right/"}],"commit":0,"type":"research"},"fileAbsolutePath":"/home/runner/work/huskydoge.github.io/huskydoge.github.io/example/content/research/2026/looped-models-fixed-points/index.md","fields":{"slug":{"html":"var _excluded = [\"components\"];\n\nfunction _extends() { _extends = Object.assign || function (target) { for (var i = 1; i < arguments.length; i++) { var source = arguments[i]; for (var key in source) { if (Object.prototype.hasOwnProperty.call(source, key)) { target[key] = source[key]; } } } return target; }; return _extends.apply(this, arguments); }\n\nfunction _objectWithoutProperties(source, excluded) { if (source == null) return {}; var target = _objectWithoutPropertiesLoose(source, excluded); var key, i; if (Object.getOwnPropertySymbols) { var sourceSymbolKeys = Object.getOwnPropertySymbols(source); for (i = 0; i < sourceSymbolKeys.length; i++) { key = sourceSymbolKeys[i]; if (excluded.indexOf(key) >= 0) continue; if (!Object.prototype.propertyIsEnumerable.call(source, key)) continue; target[key] = source[key]; } } return target; }\n\nfunction _objectWithoutPropertiesLoose(source, excluded) { if (source == null) return {}; var target = {}; var sourceKeys = Object.keys(source); var key, i; for (i = 0; i < sourceKeys.length; i++) { key = sourceKeys[i]; if (excluded.indexOf(key) >= 0) continue; target[key] = source[key]; } return target; }\n\n/* @jsxRuntime classic */\n\n/* @jsx mdx */\nvar _frontmatter = {\n  \"title\": \"Towards Looped Models Done Right — Part II: Rethinking at Fixed Points\",\n  \"date\": \"2026-09-29\",\n  \"tags\": [\"Loop Models\", \"Fixed Points\", \"Efficient Inference\"],\n  \"venue\": \"Institute of Foundation Models\",\n  \"authors\": [{\n    \"name\": \"**Benhao Huang**\"\n  }, {\n    \"name\": \"Chufan Shi\"\n  }, {\n    \"name\": \"Junlin Chen\"\n  }, {\n    \"name\": \"Shicheng Wen\"\n  }, {\n    \"name\": \"Zhengzhong Liu\"\n  }, {\n    \"name\": \"Eric Xing\"\n  }, {\n    \"name\": \"Xuezhe Ma\"\n  }],\n  \"path\": \"research/2026/looped-models-fixed-points\",\n  \"excerpt\": \"Looped LMs need not pay for every loop: near fixed points, they can scale FLOPs without scaling memory. Fixed-point shortcuts speed up training (pretraining, post-training) and inference (prefill, decoding), while a learned depth prior and orthogonal injection shape better fixed points. With a 3× smaller KV cache, the learned prior matches fixed-depth training at 1.6B.\",\n  \"selected\": true,\n  \"cover\": \"./preview.png\",\n  \"links\": [{\n    \"name\": \"paper\",\n    \"url\": \"https://www.alphaxiv.org/abs/2610.looped-models-fixed-points\"\n  }, {\n    \"name\": \"PDF\",\n    \"url\": \"https://github.com/ifm-ai/xllm-loop/blob/main/papers/part2.pdf\"\n  }, {\n    \"name\": \"Github\",\n    \"url\": \"https://github.com/ifm-ai/xllm-loop\"\n  }, {\n    \"name\": \"X thread\",\n    \"url\": \"https://x.com/huskydogewoof/status/2106130737007091939\"\n  }, {\n    \"name\": \"Part I\",\n    \"url\": \"https://huskydoge.github.io/husky-blog/posts/recursive_models/towards-looped-models-done-right/\"\n  }],\n  \"priority\": -4\n};\nvar layoutProps = {\n  _frontmatter: _frontmatter\n};\nvar MDXLayout = \"wrapper\";\nreturn function MDXContent(_ref) {\n  var components = _ref.components,\n      props = _objectWithoutProperties(_ref, _excluded);\n\n  return mdx(MDXLayout, _extends({}, layoutProps, props, {\n    components: components,\n    mdxType: \"MDXLayout\"\n  }), mdx(\"blockquote\", {\n    className: \"twitter-tweet\"\n  }, mdx(\"a\", {\n    href: \"https://x.com/huskydogewoof/status/2106130737007091939\"\n  }, \"View this thread on X\")));\n}\n;\nMDXContent.isMDXComponent = true;","htmlEncrypted":"","nonce":""}}}},"pageContext":{"fileAbsolutePath":"/home/runner/work/huskydoge.github.io/huskydoge.github.io/example/content/research/2026/looped-models-fixed-points/index.md","postPath":"research/2026/looped-models-fixed-points","translations":[{"hreflang":"en","path":"/research/2026/looped-models-fixed-points"}]}},
    "staticQueryHashes": ["3694448773","4097791827"]}