{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:LGCTBA34QVZS5EIU6J4PNOXTOG","short_pith_number":"pith:LGCTBA34","schema_version":"1.0","canonical_sha256":"598530837c85732e9114f278f6baf3719a7d5d58a30bac8f053c8061978b08d4","source":{"kind":"arxiv","id":"2502.06786","version":3},"attestation_state":"computed","paper":{"title":"Matryoshka Quantization","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aditya Kusupati, Jeff Dean, Pranav Nair, Prateek Jain, Puranjay Datta","submitted_at":"2025-02-10T18:59:10Z","abstract_excerpt":"Quantizing model weights is critical for reducing the communication and inference costs of large models. However, quantizing models -- especially to low precisions like int4 or int2 -- requires a trade-off in model quality; int2, in particular, is known to severely degrade model quality. Consequently, practitioners are often forced to maintain multiple models with different quantization levels or serve a single model that best satisfies the quality-latency trade-off. On the other hand, integer data types, such as int8, inherently possess a nested (Matryoshka) structure where smaller bit-width "},"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":"2502.06786","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-10T18:59:10Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"4bbd5ac8e3d5cb94a422cf8ad8eede4e33aee3a3e5d34471969664360ea9e0bf","abstract_canon_sha256":"ce8e36c5c184b346839ff0d364456e8a6b4cdda35f356af10f7ba06200e55e0d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:22:56.758643Z","signature_b64":"YU+Zklr6ehP55anzLCyTATN+wVzipohjxd6d6MsibxltVET2w3T5puNavViI3YySRPWjNptsJJJiTLtCWHu3Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"598530837c85732e9114f278f6baf3719a7d5d58a30bac8f053c8061978b08d4","last_reissued_at":"2026-07-05T10:22:56.757950Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:22:56.757950Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Matryoshka Quantization","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aditya Kusupati, Jeff Dean, Pranav Nair, Prateek Jain, Puranjay Datta","submitted_at":"2025-02-10T18:59:10Z","abstract_excerpt":"Quantizing model weights is critical for reducing the communication and inference costs of large models. However, quantizing models -- especially to low precisions like int4 or int2 -- requires a trade-off in model quality; int2, in particular, is known to severely degrade model quality. Consequently, practitioners are often forced to maintain multiple models with different quantization levels or serve a single model that best satisfies the quality-latency trade-off. On the other hand, integer data types, such as int8, inherently possess a nested (Matryoshka) structure where smaller bit-width "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.06786","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/2502.06786/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":"2502.06786","created_at":"2026-07-05T10:22:56.758029+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.06786v3","created_at":"2026-07-05T10:22:56.758029+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.06786","created_at":"2026-07-05T10:22:56.758029+00:00"},{"alias_kind":"pith_short_12","alias_value":"LGCTBA34QVZS","created_at":"2026-07-05T10:22:56.758029+00:00"},{"alias_kind":"pith_short_16","alias_value":"LGCTBA34QVZS5EIU","created_at":"2026-07-05T10:22:56.758029+00:00"},{"alias_kind":"pith_short_8","alias_value":"LGCTBA34","created_at":"2026-07-05T10:22:56.758029+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25674","citing_title":"BitNet Text Embeddings","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2606.23419","citing_title":"GRINQH: Graded Input-based Quantization Hierarchy for Efficient LLM Generation","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12876","citing_title":"Multi-Bitwidth Quantization for LLMs Using Additive Codebooks","ref_index":84,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LGCTBA34QVZS5EIU6J4PNOXTOG","json":"https://pith.science/pith/LGCTBA34QVZS5EIU6J4PNOXTOG.json","graph_json":"https://pith.science/api/pith-number/LGCTBA34QVZS5EIU6J4PNOXTOG/graph.json","events_json":"https://pith.science/api/pith-number/LGCTBA34QVZS5EIU6J4PNOXTOG/events.json","paper":"https://pith.science/paper/LGCTBA34"},"agent_actions":{"view_html":"https://pith.science/pith/LGCTBA34QVZS5EIU6J4PNOXTOG","download_json":"https://pith.science/pith/LGCTBA34QVZS5EIU6J4PNOXTOG.json","view_paper":"https://pith.science/paper/LGCTBA34","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.06786&json=true","fetch_graph":"https://pith.science/api/pith-number/LGCTBA34QVZS5EIU6J4PNOXTOG/graph.json","fetch_events":"https://pith.science/api/pith-number/LGCTBA34QVZS5EIU6J4PNOXTOG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LGCTBA34QVZS5EIU6J4PNOXTOG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LGCTBA34QVZS5EIU6J4PNOXTOG/action/storage_attestation","attest_author":"https://pith.science/pith/LGCTBA34QVZS5EIU6J4PNOXTOG/action/author_attestation","sign_citation":"https://pith.science/pith/LGCTBA34QVZS5EIU6J4PNOXTOG/action/citation_signature","submit_replication":"https://pith.science/pith/LGCTBA34QVZS5EIU6J4PNOXTOG/action/replication_record"}},"created_at":"2026-07-05T10:22:56.758029+00:00","updated_at":"2026-07-05T10:22:56.758029+00:00"}