{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:2U2O6KMDXX4HVBNCZKAU6ZP35X","short_pith_number":"pith:2U2O6KMD","schema_version":"1.0","canonical_sha256":"d534ef2983bdf87a85a2ca814f65fbedea4c9ef80cde65e28404866e212152c7","source":{"kind":"arxiv","id":"2506.13216","version":1},"attestation_state":"computed","paper":{"title":"Capability Salience Vector: Fine-grained Alignment of Loss and Capabilities for Downstream Task Scaling Law","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hang Yan, Kai Chen, Qiming Ge, Qipeng Guo, Qi Zhang, Shuhao Xing, Songyang Gao, Songyang Zhang, Yicheng Zou, Yunhua Zhou, Zhi Chen","submitted_at":"2025-06-16T08:16:03Z","abstract_excerpt":"Scaling law builds the relationship between training computation and validation loss, enabling researchers to effectively predict the loss trending of models across different levels of computation. However, a gap still remains between validation loss and the model's downstream capabilities, making it untrivial to apply scaling law to direct performance prediction for downstream tasks. The loss typically represents a cumulative penalty for predicted tokens, which are implicitly considered to have equal importance. Nevertheless, our studies have shown evidence that when considering different tra"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2506.13216","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-06-16T08:16:03Z","cross_cats_sorted":[],"title_canon_sha256":"fd09d91b75bcc94ec2f447055637311afaea1a80fc058fe4e9b9eac425565d94","abstract_canon_sha256":"781276a550908c4c7eb64da1354103afed7a54655dd86105fac0b6df7aa6d38c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:22:06.931579Z","signature_b64":"kXAd4qzsMlUUuDrEKCO9m8c4jpW/99JarGuQTOMYzpOtxL2wWGQJgwej50GHj/nctNj9D+ISFUJgGaNow8IUCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d534ef2983bdf87a85a2ca814f65fbedea4c9ef80cde65e28404866e212152c7","last_reissued_at":"2026-07-05T11:22:06.931109Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:22:06.931109Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Capability Salience Vector: Fine-grained Alignment of Loss and Capabilities for Downstream Task Scaling Law","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hang Yan, Kai Chen, Qiming Ge, Qipeng Guo, Qi Zhang, Shuhao Xing, Songyang Gao, Songyang Zhang, Yicheng Zou, Yunhua Zhou, Zhi Chen","submitted_at":"2025-06-16T08:16:03Z","abstract_excerpt":"Scaling law builds the relationship between training computation and validation loss, enabling researchers to effectively predict the loss trending of models across different levels of computation. However, a gap still remains between validation loss and the model's downstream capabilities, making it untrivial to apply scaling law to direct performance prediction for downstream tasks. The loss typically represents a cumulative penalty for predicted tokens, which are implicitly considered to have equal importance. Nevertheless, our studies have shown evidence that when considering different tra"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.13216","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2506.13216/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2506.13216","created_at":"2026-07-05T11:22:06.931170+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.13216v1","created_at":"2026-07-05T11:22:06.931170+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.13216","created_at":"2026-07-05T11:22:06.931170+00:00"},{"alias_kind":"pith_short_12","alias_value":"2U2O6KMDXX4H","created_at":"2026-07-05T11:22:06.931170+00:00"},{"alias_kind":"pith_short_16","alias_value":"2U2O6KMDXX4HVBNC","created_at":"2026-07-05T11:22:06.931170+00:00"},{"alias_kind":"pith_short_8","alias_value":"2U2O6KMD","created_at":"2026-07-05T11:22:06.931170+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.02364","citing_title":"InfoLaw: Information Scaling Laws for Large Language Models with Quality-Weighted Mixture Data and Repetition","ref_index":48,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2U2O6KMDXX4HVBNCZKAU6ZP35X","json":"https://pith.science/pith/2U2O6KMDXX4HVBNCZKAU6ZP35X.json","graph_json":"https://pith.science/api/pith-number/2U2O6KMDXX4HVBNCZKAU6ZP35X/graph.json","events_json":"https://pith.science/api/pith-number/2U2O6KMDXX4HVBNCZKAU6ZP35X/events.json","paper":"https://pith.science/paper/2U2O6KMD"},"agent_actions":{"view_html":"https://pith.science/pith/2U2O6KMDXX4HVBNCZKAU6ZP35X","download_json":"https://pith.science/pith/2U2O6KMDXX4HVBNCZKAU6ZP35X.json","view_paper":"https://pith.science/paper/2U2O6KMD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.13216&json=true","fetch_graph":"https://pith.science/api/pith-number/2U2O6KMDXX4HVBNCZKAU6ZP35X/graph.json","fetch_events":"https://pith.science/api/pith-number/2U2O6KMDXX4HVBNCZKAU6ZP35X/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2U2O6KMDXX4HVBNCZKAU6ZP35X/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2U2O6KMDXX4HVBNCZKAU6ZP35X/action/storage_attestation","attest_author":"https://pith.science/pith/2U2O6KMDXX4HVBNCZKAU6ZP35X/action/author_attestation","sign_citation":"https://pith.science/pith/2U2O6KMDXX4HVBNCZKAU6ZP35X/action/citation_signature","submit_replication":"https://pith.science/pith/2U2O6KMDXX4HVBNCZKAU6ZP35X/action/replication_record"}},"created_at":"2026-07-05T11:22:06.931170+00:00","updated_at":"2026-07-05T11:22:06.931170+00:00"}