{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:R3NPBOX7VTQUXOBJN3EPEDCZBB","short_pith_number":"pith:R3NPBOX7","canonical_record":{"source":{"id":"2605.09165","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-09T20:58:18Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"e7dd6a11e540667d6896d88a6bd3e9a7319c664e9e99ccc8f30c8d8686b891f7","abstract_canon_sha256":"3838b12c3466f7c6829356da5920d41aa3d9b147bfdcd764a625d78678d4438b"},"schema_version":"1.0"},"canonical_sha256":"8edaf0bafface14bb8296ec8f20c59084c6cc156bbdac87d857d10a567e35e96","source":{"kind":"arxiv","id":"2605.09165","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.09165","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"arxiv_version","alias_value":"2605.09165v2","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.09165","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_12","alias_value":"R3NPBOX7VTQU","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_16","alias_value":"R3NPBOX7VTQUXOBJ","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_8","alias_value":"R3NPBOX7","created_at":"2026-07-01T01:17:15Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:R3NPBOX7VTQUXOBJN3EPEDCZBB","target":"record","payload":{"canonical_record":{"source":{"id":"2605.09165","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-09T20:58:18Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"e7dd6a11e540667d6896d88a6bd3e9a7319c664e9e99ccc8f30c8d8686b891f7","abstract_canon_sha256":"3838b12c3466f7c6829356da5920d41aa3d9b147bfdcd764a625d78678d4438b"},"schema_version":"1.0"},"canonical_sha256":"8edaf0bafface14bb8296ec8f20c59084c6cc156bbdac87d857d10a567e35e96","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-01T01:17:15.664087Z","signature_b64":"/FV9Nzwz+X7VGpwOSbAwrY1O+AKd05p71QI4qzTDVYK61fk1Wyb+FBdm90GiKQ4/dLCL7RONeZ8ymiPnIYs/Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8edaf0bafface14bb8296ec8f20c59084c6cc156bbdac87d857d10a567e35e96","last_reissued_at":"2026-07-01T01:17:15.663657Z","signature_status":"signed_v1","first_computed_at":"2026-07-01T01:17:15.663657Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.09165","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-01T01:17:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"9zOtS85UYiIuSilqQfL1z23DnrKz8/+HG/3KSBJpF2N5+gwiLVQswiBo2YX383/Ehl0bJ+iGD+WA8zbnENU/Bg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-12T22:02:37.144598Z"},"content_sha256":"d98a51bdeaa3197f3b7f78153b0be73f3cb108d66fc29b21ff17ab938455bcb5","schema_version":"1.0","event_id":"sha256:d98a51bdeaa3197f3b7f78153b0be73f3cb108d66fc29b21ff17ab938455bcb5"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:R3NPBOX7VTQUXOBJN3EPEDCZBB","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Sparse Layers are Critical to Scaling Looped Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Looped models using sparse MoE layers scale better than dense looped models or standard transformers by activating different experts on each loop iteration.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Edward J. Hu, Jacob Biloki, Jonathan May, Ryan Lee","submitted_at":"2026-05-09T20:58:18Z","abstract_excerpt":"Looped language models repeat a set of transformer layers through depth, reducing memory costs and providing natural early-exit points at loop boundaries. However, looped models do not scale as favorably as standard transformers with unique layers. We compare standard and Mixture-of-Experts (MoE) transformers, with and without looping, and find two main results. First, we find Looped-MoE models scale better than the standard baseline while dense looped models do not. We trace this to routing divergence between loops: in Looped-MoE models, different experts are activated on each pass through th"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Looped-MoE models scale better than the standard baseline while dense looped models do not. We trace this to routing divergence between loops: in Looped-MoE models, different experts are activated on each pass through the same shared layers, recovering expressivity without additional parameters.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the performance gap is caused by routing divergence rather than differences in training procedure, optimizer state, or other unmentioned experimental details, and that the observed trends continue at larger scales.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Looped MoE models scale better than standard transformers because different experts activate on each loop pass, recovering expressivity without extra parameters, and support superior early exits.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Looped models using sparse MoE layers scale better than dense looped models or standard transformers by activating different experts on each loop iteration.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"761a153d4c6965cf87784c3f2694d40057c363aece4b3ad93a8e39473c6adbc9"},"source":{"id":"2605.09165","kind":"arxiv","version":2},"verdict":{"id":"f6cfbedf-7b64-49bf-8124-4a094a253dbc","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T04:15:08.541694Z","strongest_claim":"Looped-MoE models scale better than the standard baseline while dense looped models do not. We trace this to routing divergence between loops: in Looped-MoE models, different experts are activated on each pass through the same shared layers, recovering expressivity without additional parameters.","one_line_summary":"Looped MoE models scale better than standard transformers because different experts activate on each loop pass, recovering expressivity without extra parameters, and support superior early exits.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the performance gap is caused by routing divergence rather than differences in training procedure, optimizer state, or other unmentioned experimental details, and that the observed trends continue at larger scales.","pith_extraction_headline":"Looped models using sparse MoE layers scale better than dense looped models or standard transformers by activating different experts on each loop iteration."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.09165/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T08:22:01.610282Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T20:36:25.121358Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T13:31:18.550230Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T10:29:23.697666Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"c555607e955949c36f38459a817a257a273bbb6a49060ad9fec1c8e60df61ec7"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"4753145a5cb688652400a2532e11d21e41165b18562ca8cf0fb8937ce0d6caed"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"f6cfbedf-7b64-49bf-8124-4a094a253dbc"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-01T01:17:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"7JE3J5O38m2++kYu8aMa8xpoB69bw7oJYLbnV622D3TjY2G4trpkgmsUL11stw5dVICdx5N0b+nv5zeb4pU0DA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-12T22:02:37.145100Z"},"content_sha256":"6cf4c59d79d57898081ccaf9d595451136bd3a7e6b7800b4c05d65c9c7dc29e6","schema_version":"1.0","event_id":"sha256:6cf4c59d79d57898081ccaf9d595451136bd3a7e6b7800b4c05d65c9c7dc29e6"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/R3NPBOX7VTQUXOBJN3EPEDCZBB/bundle.json","state_url":"https://pith.science/pith/R3NPBOX7VTQUXOBJN3EPEDCZBB/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/R3NPBOX7VTQUXOBJN3EPEDCZBB/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-12T22:02:37Z","links":{"resolver":"https://pith.science/pith/R3NPBOX7VTQUXOBJN3EPEDCZBB","bundle":"https://pith.science/pith/R3NPBOX7VTQUXOBJN3EPEDCZBB/bundle.json","state":"https://pith.science/pith/R3NPBOX7VTQUXOBJN3EPEDCZBB/state.json","well_known_bundle":"https://pith.science/.well-known/pith/R3NPBOX7VTQUXOBJN3EPEDCZBB/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:R3NPBOX7VTQUXOBJN3EPEDCZBB","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"3838b12c3466f7c6829356da5920d41aa3d9b147bfdcd764a625d78678d4438b","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-09T20:58:18Z","title_canon_sha256":"e7dd6a11e540667d6896d88a6bd3e9a7319c664e9e99ccc8f30c8d8686b891f7"},"schema_version":"1.0","source":{"id":"2605.09165","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.09165","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"arxiv_version","alias_value":"2605.09165v2","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.09165","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_12","alias_value":"R3NPBOX7VTQU","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_16","alias_value":"R3NPBOX7VTQUXOBJ","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_8","alias_value":"R3NPBOX7","created_at":"2026-07-01T01:17:15Z"}],"graph_snapshots":[{"event_id":"sha256:6cf4c59d79d57898081ccaf9d595451136bd3a7e6b7800b4c05d65c9c7dc29e6","target":"graph","created_at":"2026-07-01T01:17:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Looped-MoE models scale better than the standard baseline while dense looped models do not. We trace this to routing divergence between loops: in Looped-MoE models, different experts are activated on each pass through the same shared layers, recovering expressivity without additional parameters."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the performance gap is caused by routing divergence rather than differences in training procedure, optimizer state, or other unmentioned experimental details, and that the observed trends continue at larger scales."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Looped MoE models scale better than standard transformers because different experts activate on each loop pass, recovering expressivity without extra parameters, and support superior early exits."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Looped models using sparse MoE layers scale better than dense looped models or standard transformers by activating different experts on each loop iteration."}],"snapshot_sha256":"761a153d4c6965cf87784c3f2694d40057c363aece4b3ad93a8e39473c6adbc9"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"4753145a5cb688652400a2532e11d21e41165b18562ca8cf0fb8937ce0d6caed"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T08:22:01.610282Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T20:36:25.121358Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T13:31:18.550230Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T10:29:23.697666Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.09165/integrity.json","findings":[],"snapshot_sha256":"c555607e955949c36f38459a817a257a273bbb6a49060ad9fec1c8e60df61ec7","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Looped language models repeat a set of transformer layers through depth, reducing memory costs and providing natural early-exit points at loop boundaries. However, looped models do not scale as favorably as standard transformers with unique layers. We compare standard and Mixture-of-Experts (MoE) transformers, with and without looping, and find two main results. First, we find Looped-MoE models scale better than the standard baseline while dense looped models do not. We trace this to routing divergence between loops: in Looped-MoE models, different experts are activated on each pass through th","authors_text":"Edward J. Hu, Jacob Biloki, Jonathan May, Ryan Lee","cross_cats":["cs.CL"],"headline":"Looped models using sparse MoE layers scale better than dense looped models or standard transformers by activating different experts on each loop iteration.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-09T20:58:18Z","title":"Sparse Layers are Critical to Scaling Looped Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.09165","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-12T04:15:08.541694Z","id":"f6cfbedf-7b64-49bf-8124-4a094a253dbc","model_set":{"reader":"grok-4.3"},"one_line_summary":"Looped MoE models scale better than standard transformers because different experts activate on each loop pass, recovering expressivity without extra parameters, and support superior early exits.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Looped models using sparse MoE layers scale better than dense looped models or standard transformers by activating different experts on each loop iteration.","strongest_claim":"Looped-MoE models scale better than the standard baseline while dense looped models do not. We trace this to routing divergence between loops: in Looped-MoE models, different experts are activated on each pass through the same shared layers, recovering expressivity without additional parameters.","weakest_assumption":"That the performance gap is caused by routing divergence rather than differences in training procedure, optimizer state, or other unmentioned experimental details, and that the observed trends continue at larger scales."}},"verdict_id":"f6cfbedf-7b64-49bf-8124-4a094a253dbc"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d98a51bdeaa3197f3b7f78153b0be73f3cb108d66fc29b21ff17ab938455bcb5","target":"record","created_at":"2026-07-01T01:17:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"3838b12c3466f7c6829356da5920d41aa3d9b147bfdcd764a625d78678d4438b","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-09T20:58:18Z","title_canon_sha256":"e7dd6a11e540667d6896d88a6bd3e9a7319c664e9e99ccc8f30c8d8686b891f7"},"schema_version":"1.0","source":{"id":"2605.09165","kind":"arxiv","version":2}},"canonical_sha256":"8edaf0bafface14bb8296ec8f20c59084c6cc156bbdac87d857d10a567e35e96","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"8edaf0bafface14bb8296ec8f20c59084c6cc156bbdac87d857d10a567e35e96","first_computed_at":"2026-07-01T01:17:15.663657Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-01T01:17:15.663657Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"/FV9Nzwz+X7VGpwOSbAwrY1O+AKd05p71QI4qzTDVYK61fk1Wyb+FBdm90GiKQ4/dLCL7RONeZ8ymiPnIYs/Bg==","signature_status":"signed_v1","signed_at":"2026-07-01T01:17:15.664087Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.09165","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d98a51bdeaa3197f3b7f78153b0be73f3cb108d66fc29b21ff17ab938455bcb5","sha256:6cf4c59d79d57898081ccaf9d595451136bd3a7e6b7800b4c05d65c9c7dc29e6"],"state_sha256":"d67deb4202628ad06836aad077ca8bd3316050dd26c444ea9d5338d9ed09ba03"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"WViT5AOV1w55xFeF7/OnGULQesfBGSqbaCfsYG3whWOr6zTlveXLoYJiIpXc7jKn9ylxWA7pRyyftp2S2SMHCQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-12T22:02:37.148110Z","bundle_sha256":"7b64136f8631cf63140b4592a07ec734502f40a44010712a4b179f6ae13c2094"}}