{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:CJ6GZJMJ43IFVGCNJTVJQAMLZ3","short_pith_number":"pith:CJ6GZJMJ","schema_version":"1.0","canonical_sha256":"127c6ca589e6d05a984d4cea98018bcec766905a7b89c62bc55e2d155c71608c","source":{"kind":"arxiv","id":"2410.13230","version":3},"attestation_state":"computed","paper":{"title":"Starbucks-v2: Improved Training for 2D Matryoshka Embeddings","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.IR","authors_text":"Bevan Koopman, Fabio Zheng, Guido Zuccon, Shengyao Zhuang, Shuai Wang","submitted_at":"2024-10-17T05:33:50Z","abstract_excerpt":"2D Matryoshka training enables a single embedding model to generate sub-network representations across different layers and embedding dimensions, offering adaptability to diverse computational and task constraints. However, its effectiveness remains well below that of individually trained models of equivalent sizes. To address this, we propose Starbucks, a new training strategy for Matryoshka-style embedding models that combines structured fine-tuning with masked autoencoder (MAE) pre-training. During fine-tuning, we compute the loss over a fixed set of layer-dimension pairs, from small to lar"},"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":"2410.13230","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.IR","submitted_at":"2024-10-17T05:33:50Z","cross_cats_sorted":[],"title_canon_sha256":"87b71284f4d254c42cd3646b9614262c2e18d58bd31febde8b5a6ac1b04f78b4","abstract_canon_sha256":"96853a340f259805a28e294bf599ad5f66d045347b6145868b5e0251bced5657"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:12:21.436727Z","signature_b64":"ufl99p538+s9kVjnruC/jK0QXIe7mD27VGnmnrXhGJPHhTu6gzCFR8EcUKTRYn3b3tQulT2+wEChXJqc5B7cBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"127c6ca589e6d05a984d4cea98018bcec766905a7b89c62bc55e2d155c71608c","last_reissued_at":"2026-07-05T11:12:21.436157Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:12:21.436157Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Starbucks-v2: Improved Training for 2D Matryoshka Embeddings","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.IR","authors_text":"Bevan Koopman, Fabio Zheng, Guido Zuccon, Shengyao Zhuang, Shuai Wang","submitted_at":"2024-10-17T05:33:50Z","abstract_excerpt":"2D Matryoshka training enables a single embedding model to generate sub-network representations across different layers and embedding dimensions, offering adaptability to diverse computational and task constraints. However, its effectiveness remains well below that of individually trained models of equivalent sizes. To address this, we propose Starbucks, a new training strategy for Matryoshka-style embedding models that combines structured fine-tuning with masked autoencoder (MAE) pre-training. During fine-tuning, we compute the loss over a fixed set of layer-dimension pairs, from small to lar"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.13230","kind":"arxiv","version":3},"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/2410.13230/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":"2410.13230","created_at":"2026-07-05T11:12:21.436227+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.13230v3","created_at":"2026-07-05T11:12:21.436227+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.13230","created_at":"2026-07-05T11:12:21.436227+00:00"},{"alias_kind":"pith_short_12","alias_value":"CJ6GZJMJ43IF","created_at":"2026-07-05T11:12:21.436227+00:00"},{"alias_kind":"pith_short_16","alias_value":"CJ6GZJMJ43IFVGCN","created_at":"2026-07-05T11:12:21.436227+00:00"},{"alias_kind":"pith_short_8","alias_value":"CJ6GZJMJ","created_at":"2026-07-05T11:12:21.436227+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07654","citing_title":"MM-Matryoshka: Towards Budget-Elastic Visual Document Retrieval via a 2D Multimodal Matryoshka Training Framework","ref_index":100,"is_internal_anchor":false},{"citing_arxiv_id":"2601.21262","citing_title":"CausalEmbed: Auto-Regressive Multi-Vector Generation in Latent Space for Visual Document Embedding","ref_index":38,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CJ6GZJMJ43IFVGCNJTVJQAMLZ3","json":"https://pith.science/pith/CJ6GZJMJ43IFVGCNJTVJQAMLZ3.json","graph_json":"https://pith.science/api/pith-number/CJ6GZJMJ43IFVGCNJTVJQAMLZ3/graph.json","events_json":"https://pith.science/api/pith-number/CJ6GZJMJ43IFVGCNJTVJQAMLZ3/events.json","paper":"https://pith.science/paper/CJ6GZJMJ"},"agent_actions":{"view_html":"https://pith.science/pith/CJ6GZJMJ43IFVGCNJTVJQAMLZ3","download_json":"https://pith.science/pith/CJ6GZJMJ43IFVGCNJTVJQAMLZ3.json","view_paper":"https://pith.science/paper/CJ6GZJMJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.13230&json=true","fetch_graph":"https://pith.science/api/pith-number/CJ6GZJMJ43IFVGCNJTVJQAMLZ3/graph.json","fetch_events":"https://pith.science/api/pith-number/CJ6GZJMJ43IFVGCNJTVJQAMLZ3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CJ6GZJMJ43IFVGCNJTVJQAMLZ3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CJ6GZJMJ43IFVGCNJTVJQAMLZ3/action/storage_attestation","attest_author":"https://pith.science/pith/CJ6GZJMJ43IFVGCNJTVJQAMLZ3/action/author_attestation","sign_citation":"https://pith.science/pith/CJ6GZJMJ43IFVGCNJTVJQAMLZ3/action/citation_signature","submit_replication":"https://pith.science/pith/CJ6GZJMJ43IFVGCNJTVJQAMLZ3/action/replication_record"}},"created_at":"2026-07-05T11:12:21.436227+00:00","updated_at":"2026-07-05T11:12:21.436227+00:00"}