{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:3RV3KM22XZRYNYMQWRFT752L3W","short_pith_number":"pith:3RV3KM22","schema_version":"1.0","canonical_sha256":"dc6bb5335abe6386e190b44b3ff74bdda73966805e13371e1dae4bf63462fab0","source":{"kind":"arxiv","id":"2305.07017","version":2},"attestation_state":"computed","paper":{"title":"An Inverse Scaling Law for CLIP Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Cihang Xie, Xianhang Li, Zeyu Wang","submitted_at":"2023-05-11T17:56:09Z","abstract_excerpt":"CLIP, one of the pioneering foundation models that connect images and text, has enabled many recent breakthroughs in computer vision. However, its associated training cost is prohibitively high, imposing a significant barrier to its widespread exploration. In this paper, we present a surprising finding that there exists an inverse scaling law for CLIP training, whereby the larger the image/text encoders used, the shorter the sequence length of image/text tokens that can be applied in training. Moreover, we showcase that the strategy for reducing image/text token length plays a crucial role in "},"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":"2305.07017","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2023-05-11T17:56:09Z","cross_cats_sorted":[],"title_canon_sha256":"8c6a42be81e7252d8529d4e78059d58be818f6c1200c534c3ec1e2d9c495cfc3","abstract_canon_sha256":"df52e1d21854576f0de79c6d67da51c5ce86e01dc94088db928c8194d68830fe"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:06:26.778117Z","signature_b64":"/CspnMYfpVpfiI0wZCLBbtSJChXBK0YUbC9Ue5Nn1ASzkCOWJbgsssUe4GIDuQ3PnL5g7i3vAwdsYcGjlLXqBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dc6bb5335abe6386e190b44b3ff74bdda73966805e13371e1dae4bf63462fab0","last_reissued_at":"2026-07-05T07:06:26.777646Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:06:26.777646Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"An Inverse Scaling Law for CLIP Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Cihang Xie, Xianhang Li, Zeyu Wang","submitted_at":"2023-05-11T17:56:09Z","abstract_excerpt":"CLIP, one of the pioneering foundation models that connect images and text, has enabled many recent breakthroughs in computer vision. However, its associated training cost is prohibitively high, imposing a significant barrier to its widespread exploration. In this paper, we present a surprising finding that there exists an inverse scaling law for CLIP training, whereby the larger the image/text encoders used, the shorter the sequence length of image/text tokens that can be applied in training. Moreover, we showcase that the strategy for reducing image/text token length plays a crucial role in "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.07017","kind":"arxiv","version":2},"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/2305.07017/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":"2305.07017","created_at":"2026-07-05T07:06:26.777706+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.07017v2","created_at":"2026-07-05T07:06:26.777706+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.07017","created_at":"2026-07-05T07:06:26.777706+00:00"},{"alias_kind":"pith_short_12","alias_value":"3RV3KM22XZRY","created_at":"2026-07-05T07:06:26.777706+00:00"},{"alias_kind":"pith_short_16","alias_value":"3RV3KM22XZRYNYMQ","created_at":"2026-07-05T07:06:26.777706+00:00"},{"alias_kind":"pith_short_8","alias_value":"3RV3KM22","created_at":"2026-07-05T07:06:26.777706+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2509.02175","citing_title":"Understanding Space Is Rocket Science -- Only Top Reasoning Models Can Solve Spatial Understanding Tasks","ref_index":33,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3RV3KM22XZRYNYMQWRFT752L3W","json":"https://pith.science/pith/3RV3KM22XZRYNYMQWRFT752L3W.json","graph_json":"https://pith.science/api/pith-number/3RV3KM22XZRYNYMQWRFT752L3W/graph.json","events_json":"https://pith.science/api/pith-number/3RV3KM22XZRYNYMQWRFT752L3W/events.json","paper":"https://pith.science/paper/3RV3KM22"},"agent_actions":{"view_html":"https://pith.science/pith/3RV3KM22XZRYNYMQWRFT752L3W","download_json":"https://pith.science/pith/3RV3KM22XZRYNYMQWRFT752L3W.json","view_paper":"https://pith.science/paper/3RV3KM22","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.07017&json=true","fetch_graph":"https://pith.science/api/pith-number/3RV3KM22XZRYNYMQWRFT752L3W/graph.json","fetch_events":"https://pith.science/api/pith-number/3RV3KM22XZRYNYMQWRFT752L3W/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3RV3KM22XZRYNYMQWRFT752L3W/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3RV3KM22XZRYNYMQWRFT752L3W/action/storage_attestation","attest_author":"https://pith.science/pith/3RV3KM22XZRYNYMQWRFT752L3W/action/author_attestation","sign_citation":"https://pith.science/pith/3RV3KM22XZRYNYMQWRFT752L3W/action/citation_signature","submit_replication":"https://pith.science/pith/3RV3KM22XZRYNYMQWRFT752L3W/action/replication_record"}},"created_at":"2026-07-05T07:06:26.777706+00:00","updated_at":"2026-07-05T07:06:26.777706+00:00"}