{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:XK5AKJGWLD6QO5HVXXNAYLNNJK","short_pith_number":"pith:XK5AKJGW","canonical_record":{"source":{"id":"2605.11416","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-12T02:11:31Z","cross_cats_sorted":[],"title_canon_sha256":"36485d8ccb4a57da68441f81584d1fb2c8ec3248707af62e02576da1f165e343","abstract_canon_sha256":"4164c9cf64e3e4bbb74baa4b1cb91e78590c2c280db997638621b160f17a6ae6"},"schema_version":"1.0"},"canonical_sha256":"baba0524d658fd0774f5bdda0c2dad4a83620b8fd85811863feea82ed7b9d809","source":{"kind":"arxiv","id":"2605.11416","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.11416","created_at":"2026-05-25T02:01:23Z"},{"alias_kind":"arxiv_version","alias_value":"2605.11416v2","created_at":"2026-05-25T02:01:23Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.11416","created_at":"2026-05-25T02:01:23Z"},{"alias_kind":"pith_short_12","alias_value":"XK5AKJGWLD6Q","created_at":"2026-05-25T02:01:23Z"},{"alias_kind":"pith_short_16","alias_value":"XK5AKJGWLD6QO5HV","created_at":"2026-05-25T02:01:23Z"},{"alias_kind":"pith_short_8","alias_value":"XK5AKJGW","created_at":"2026-05-25T02:01:23Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:XK5AKJGWLD6QO5HVXXNAYLNNJK","target":"record","payload":{"canonical_record":{"source":{"id":"2605.11416","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-12T02:11:31Z","cross_cats_sorted":[],"title_canon_sha256":"36485d8ccb4a57da68441f81584d1fb2c8ec3248707af62e02576da1f165e343","abstract_canon_sha256":"4164c9cf64e3e4bbb74baa4b1cb91e78590c2c280db997638621b160f17a6ae6"},"schema_version":"1.0"},"canonical_sha256":"baba0524d658fd0774f5bdda0c2dad4a83620b8fd85811863feea82ed7b9d809","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:01:23.127976Z","signature_b64":"Muu4Jrx/1w/0UC5zxYlkSScTmq43rqS+VdOyuYEaiUuwFY/zCiBB6FPWQb2Lxu0G0V1NJz8BPAtSNz5/9SbuAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"baba0524d658fd0774f5bdda0c2dad4a83620b8fd85811863feea82ed7b9d809","last_reissued_at":"2026-05-25T02:01:23.127245Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:01:23.127245Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.11416","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":"2026-05-25T02:01:23Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"VU9XldIpzhTYJaXuxlmtV/rrNg+0jQDm5iKGm/lbM62dl1DdBiEeg8y/pIc9bh2fIbYW1paKJDUEoaxE40/YCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-28T02:34:04.972747Z"},"content_sha256":"d3e9979061fe5cb4c524e996942c29bbb38307798db650bf237c004032476a4c","schema_version":"1.0","event_id":"sha256:d3e9979061fe5cb4c524e996942c29bbb38307798db650bf237c004032476a4c"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:XK5AKJGWLD6QO5HVXXNAYLNNJK","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Freeze Deep, Train Shallow: Interpretable Layer Allocation for Continued Pre-Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Training shallow layers while freezing deep layers outperforms full-parameter updates in continued pre-training of LLMs.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Bo Jiang, Jiang-Feng Yang, Qing-Wei Cong, Qin-Yuan Liu, Qiu-Yang Zhao, Yu-Hang Wu","submitted_at":"2026-05-12T02:11:31Z","abstract_excerpt":"Selective layer-wise updates are essential for low-cost continued pre-training of Large Language Models (LLMs), yet determining which layers to freeze or train remains an empirical black-box problem due to the lack of interpretable guidance. To address this issue, we propose LayerTracer, an architecture-agnostic diagnostic framework that reveals the evolution patterns of layer-wise representations and stability by locating task execution positions and quantifying layer sensitivity. Analysis results reveal that deep layers act as critical regions for task execution and maintain high stability a"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"training shallow layers while freezing deep layers consistently outperforms full-parameter fine-tuning and the opposite allocation on both C-Eval and CMMLU benchmarks","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The diagnostic patterns identified by LayerTracer (deep layers as critical and stable) are general enough to guide allocation decisions across models, tasks, and continued-pretraining regimes, and the three controlled trials isolate layer allocation as the causal factor.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Freezing deep layers and training shallow layers during continued pre-training of LLMs outperforms full fine-tuning and the opposite allocation on C-Eval and CMMLU, guided by a new layer-sensitivity diagnostic.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Training shallow layers while freezing deep layers outperforms full-parameter updates in continued pre-training of LLMs.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"178f1f51d7c92aadf39cf9b0ad4de7ab488e01ab473a7012c6a2cd7af4e43872"},"source":{"id":"2605.11416","kind":"arxiv","version":2},"verdict":{"id":"a32f5629-f3fa-45d5-9a26-1f9dfbf8e4ec","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T02:35:03.549753Z","strongest_claim":"training shallow layers while freezing deep layers consistently outperforms full-parameter fine-tuning and the opposite allocation on both C-Eval and CMMLU benchmarks","one_line_summary":"Freezing deep layers and training shallow layers during continued pre-training of LLMs outperforms full fine-tuning and the opposite allocation on C-Eval and CMMLU, guided by a new layer-sensitivity diagnostic.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The diagnostic patterns identified by LayerTracer (deep layers as critical and stable) are general enough to guide allocation decisions across models, tasks, and continued-pretraining regimes, and the three controlled trials isolate layer allocation as the causal factor.","pith_extraction_headline":"Training shallow layers while freezing deep layers outperforms full-parameter updates in continued pre-training of LLMs."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.11416/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T04:22:00.467828Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T12:36:28.217876Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T10:01:16.593174Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:28:45.877855Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"2110b427a7760f7151e21455c6bfdfdd531afa3a2f2dc79b896fd4bec1ef02f8"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"a32f5629-f3fa-45d5-9a26-1f9dfbf8e4ec"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-25T02:01:23Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"erpJfEi9t8FfUKgYr1Drqsp8Lxueuo8lmIQQQEK9yxZU0b9HiF1RA99FyZhl7WcT+z6KrwmAzfBT4GKa64HHAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-28T02:34:04.973260Z"},"content_sha256":"d874586b0618154f78ca712d3ca9d0c31085ccd823c6d98f01e0b3b8fd7f13f3","schema_version":"1.0","event_id":"sha256:d874586b0618154f78ca712d3ca9d0c31085ccd823c6d98f01e0b3b8fd7f13f3"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/XK5AKJGWLD6QO5HVXXNAYLNNJK/bundle.json","state_url":"https://pith.science/pith/XK5AKJGWLD6QO5HVXXNAYLNNJK/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/XK5AKJGWLD6QO5HVXXNAYLNNJK/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":"2026-06-28T02:34:04Z","links":{"resolver":"https://pith.science/pith/XK5AKJGWLD6QO5HVXXNAYLNNJK","bundle":"https://pith.science/pith/XK5AKJGWLD6QO5HVXXNAYLNNJK/bundle.json","state":"https://pith.science/pith/XK5AKJGWLD6QO5HVXXNAYLNNJK/state.json","well_known_bundle":"https://pith.science/.well-known/pith/XK5AKJGWLD6QO5HVXXNAYLNNJK/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:XK5AKJGWLD6QO5HVXXNAYLNNJK","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":"4164c9cf64e3e4bbb74baa4b1cb91e78590c2c280db997638621b160f17a6ae6","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-12T02:11:31Z","title_canon_sha256":"36485d8ccb4a57da68441f81584d1fb2c8ec3248707af62e02576da1f165e343"},"schema_version":"1.0","source":{"id":"2605.11416","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.11416","created_at":"2026-05-25T02:01:23Z"},{"alias_kind":"arxiv_version","alias_value":"2605.11416v2","created_at":"2026-05-25T02:01:23Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.11416","created_at":"2026-05-25T02:01:23Z"},{"alias_kind":"pith_short_12","alias_value":"XK5AKJGWLD6Q","created_at":"2026-05-25T02:01:23Z"},{"alias_kind":"pith_short_16","alias_value":"XK5AKJGWLD6QO5HV","created_at":"2026-05-25T02:01:23Z"},{"alias_kind":"pith_short_8","alias_value":"XK5AKJGW","created_at":"2026-05-25T02:01:23Z"}],"graph_snapshots":[{"event_id":"sha256:d874586b0618154f78ca712d3ca9d0c31085ccd823c6d98f01e0b3b8fd7f13f3","target":"graph","created_at":"2026-05-25T02:01:23Z","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":"training shallow layers while freezing deep layers consistently outperforms full-parameter fine-tuning and the opposite allocation on both C-Eval and CMMLU benchmarks"},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The diagnostic patterns identified by LayerTracer (deep layers as critical and stable) are general enough to guide allocation decisions across models, tasks, and continued-pretraining regimes, and the three controlled trials isolate layer allocation as the causal factor."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Freezing deep layers and training shallow layers during continued pre-training of LLMs outperforms full fine-tuning and the opposite allocation on C-Eval and CMMLU, guided by a new layer-sensitivity diagnostic."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Training shallow layers while freezing deep layers outperforms full-parameter updates in continued pre-training of LLMs."}],"snapshot_sha256":"178f1f51d7c92aadf39cf9b0ad4de7ab488e01ab473a7012c6a2cd7af4e43872"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T04:22:00.467828Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T12:36:28.217876Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T10:01:16.593174Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T08:28:45.877855Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.11416/integrity.json","findings":[],"snapshot_sha256":"2110b427a7760f7151e21455c6bfdfdd531afa3a2f2dc79b896fd4bec1ef02f8","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Selective layer-wise updates are essential for low-cost continued pre-training of Large Language Models (LLMs), yet determining which layers to freeze or train remains an empirical black-box problem due to the lack of interpretable guidance. To address this issue, we propose LayerTracer, an architecture-agnostic diagnostic framework that reveals the evolution patterns of layer-wise representations and stability by locating task execution positions and quantifying layer sensitivity. Analysis results reveal that deep layers act as critical regions for task execution and maintain high stability a","authors_text":"Bo Jiang, Jiang-Feng Yang, Qing-Wei Cong, Qin-Yuan Liu, Qiu-Yang Zhao, Yu-Hang Wu","cross_cats":[],"headline":"Training shallow layers while freezing deep layers outperforms full-parameter updates in continued pre-training of LLMs.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-12T02:11:31Z","title":"Freeze Deep, Train Shallow: Interpretable Layer Allocation for Continued Pre-Training"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.11416","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-13T02:35:03.549753Z","id":"a32f5629-f3fa-45d5-9a26-1f9dfbf8e4ec","model_set":{"reader":"grok-4.3"},"one_line_summary":"Freezing deep layers and training shallow layers during continued pre-training of LLMs outperforms full fine-tuning and the opposite allocation on C-Eval and CMMLU, guided by a new layer-sensitivity diagnostic.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Training shallow layers while freezing deep layers outperforms full-parameter updates in continued pre-training of LLMs.","strongest_claim":"training shallow layers while freezing deep layers consistently outperforms full-parameter fine-tuning and the opposite allocation on both C-Eval and CMMLU benchmarks","weakest_assumption":"The diagnostic patterns identified by LayerTracer (deep layers as critical and stable) are general enough to guide allocation decisions across models, tasks, and continued-pretraining regimes, and the three controlled trials isolate layer allocation as the causal factor."}},"verdict_id":"a32f5629-f3fa-45d5-9a26-1f9dfbf8e4ec"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d3e9979061fe5cb4c524e996942c29bbb38307798db650bf237c004032476a4c","target":"record","created_at":"2026-05-25T02:01:23Z","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":"4164c9cf64e3e4bbb74baa4b1cb91e78590c2c280db997638621b160f17a6ae6","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-12T02:11:31Z","title_canon_sha256":"36485d8ccb4a57da68441f81584d1fb2c8ec3248707af62e02576da1f165e343"},"schema_version":"1.0","source":{"id":"2605.11416","kind":"arxiv","version":2}},"canonical_sha256":"baba0524d658fd0774f5bdda0c2dad4a83620b8fd85811863feea82ed7b9d809","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"baba0524d658fd0774f5bdda0c2dad4a83620b8fd85811863feea82ed7b9d809","first_computed_at":"2026-05-25T02:01:23.127245Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-25T02:01:23.127245Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Muu4Jrx/1w/0UC5zxYlkSScTmq43rqS+VdOyuYEaiUuwFY/zCiBB6FPWQb2Lxu0G0V1NJz8BPAtSNz5/9SbuAw==","signature_status":"signed_v1","signed_at":"2026-05-25T02:01:23.127976Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.11416","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d3e9979061fe5cb4c524e996942c29bbb38307798db650bf237c004032476a4c","sha256:d874586b0618154f78ca712d3ca9d0c31085ccd823c6d98f01e0b3b8fd7f13f3"],"state_sha256":"e952dd1a26ad582ca63e417d9a913c1720ff7bd4f3b6053d4afba252bf81e70d"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"KuvtHU+MVQe+tBZEeYfKEWE6bKxy6wwhjcKPo5URuCrGP98RdsrpvozkxvNnoHA0rN5RNskMoNz9rsYHtEFKCQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-28T02:34:04.975798Z","bundle_sha256":"9658aeead1d5fa59bc3ca3236576caf282b2bf150ce9fd6ee407587182052340"}}