{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:4MOA4IJZPKR4DQFC6M6Z2MTS6N","short_pith_number":"pith:4MOA4IJZ","schema_version":"1.0","canonical_sha256":"e31c0e21397aa3c1c0a2f33d9d3272f347dcb551d66ab13c921ee70904446747","source":{"kind":"arxiv","id":"2004.07320","version":3},"attestation_state":"computed","paper":{"title":"Training with Quantization Noise for Extreme Model Compression","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Angela Fan, Armand Joulin, Benjamin Graham, Edouard Grave, Herve Jegou, Pierre Stock, Remi Gribonval","submitted_at":"2020-04-15T20:10:53Z","abstract_excerpt":"We tackle the problem of producing compact models, maximizing their accuracy for a given model size. A standard solution is to train networks with Quantization Aware Training, where the weights are quantized during training and the gradients approximated with the Straight-Through Estimator. In this paper, we extend this approach to work beyond int8 fixed-point quantization with extreme compression methods where the approximations introduced by STE are severe, such as Product Quantization. Our proposal is to only quantize a different random subset of weights during each forward, allowing for un"},"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":"2004.07320","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2020-04-15T20:10:53Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"918b6f74e82b6649f5dabf10ccc43355cf3f7fd947515c88e9f3f8c458fce01f","abstract_canon_sha256":"db4bc9a966536d3e78d90033acc233a59ae86bc341bcc192eb6493188f07d5a2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:18:51.695929Z","signature_b64":"dw56Zp55LTzLDt7ULfyEtGwJlSN8weBHlBIzcOgCcg4dHTXlaa3rO1pejWo0RKF3t+gqAgFDaoagmfDPb/YRDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e31c0e21397aa3c1c0a2f33d9d3272f347dcb551d66ab13c921ee70904446747","last_reissued_at":"2026-07-05T02:18:51.695487Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:18:51.695487Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Training with Quantization Noise for Extreme Model Compression","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Angela Fan, Armand Joulin, Benjamin Graham, Edouard Grave, Herve Jegou, Pierre Stock, Remi Gribonval","submitted_at":"2020-04-15T20:10:53Z","abstract_excerpt":"We tackle the problem of producing compact models, maximizing their accuracy for a given model size. A standard solution is to train networks with Quantization Aware Training, where the weights are quantized during training and the gradients approximated with the Straight-Through Estimator. In this paper, we extend this approach to work beyond int8 fixed-point quantization with extreme compression methods where the approximations introduced by STE are severe, such as Product Quantization. Our proposal is to only quantize a different random subset of weights during each forward, allowing for un"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2004.07320","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/2004.07320/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":"2004.07320","created_at":"2026-07-05T02:18:51.695545+00:00"},{"alias_kind":"arxiv_version","alias_value":"2004.07320v3","created_at":"2026-07-05T02:18:51.695545+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2004.07320","created_at":"2026-07-05T02:18:51.695545+00:00"},{"alias_kind":"pith_short_12","alias_value":"4MOA4IJZPKR4","created_at":"2026-07-05T02:18:51.695545+00:00"},{"alias_kind":"pith_short_16","alias_value":"4MOA4IJZPKR4DQFC","created_at":"2026-07-05T02:18:51.695545+00:00"},{"alias_kind":"pith_short_8","alias_value":"4MOA4IJZ","created_at":"2026-07-05T02:18:51.695545+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2112.11447","citing_title":"Multi-Modality Distillation via Learning the teacher's modality-level Gram Matrix","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14037","citing_title":"Self-Pruned Key-Value Attention: Learning When to Write by Predicting Future Utility","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2208.07339","citing_title":"LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale","ref_index":132,"is_internal_anchor":false},{"citing_arxiv_id":"2006.04768","citing_title":"Linformer: Self-Attention with Linear Complexity","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04493","citing_title":"SLaB: Sparse-Lowrank-Binary Decomposition for Efficient Large Language Models","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4MOA4IJZPKR4DQFC6M6Z2MTS6N","json":"https://pith.science/pith/4MOA4IJZPKR4DQFC6M6Z2MTS6N.json","graph_json":"https://pith.science/api/pith-number/4MOA4IJZPKR4DQFC6M6Z2MTS6N/graph.json","events_json":"https://pith.science/api/pith-number/4MOA4IJZPKR4DQFC6M6Z2MTS6N/events.json","paper":"https://pith.science/paper/4MOA4IJZ"},"agent_actions":{"view_html":"https://pith.science/pith/4MOA4IJZPKR4DQFC6M6Z2MTS6N","download_json":"https://pith.science/pith/4MOA4IJZPKR4DQFC6M6Z2MTS6N.json","view_paper":"https://pith.science/paper/4MOA4IJZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2004.07320&json=true","fetch_graph":"https://pith.science/api/pith-number/4MOA4IJZPKR4DQFC6M6Z2MTS6N/graph.json","fetch_events":"https://pith.science/api/pith-number/4MOA4IJZPKR4DQFC6M6Z2MTS6N/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4MOA4IJZPKR4DQFC6M6Z2MTS6N/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4MOA4IJZPKR4DQFC6M6Z2MTS6N/action/storage_attestation","attest_author":"https://pith.science/pith/4MOA4IJZPKR4DQFC6M6Z2MTS6N/action/author_attestation","sign_citation":"https://pith.science/pith/4MOA4IJZPKR4DQFC6M6Z2MTS6N/action/citation_signature","submit_replication":"https://pith.science/pith/4MOA4IJZPKR4DQFC6M6Z2MTS6N/action/replication_record"}},"created_at":"2026-07-05T02:18:51.695545+00:00","updated_at":"2026-07-05T02:18:51.695545+00:00"}