{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:FPXWJIWM4S3I72FUWEYZMAKBGM","short_pith_number":"pith:FPXWJIWM","schema_version":"1.0","canonical_sha256":"2bef64a2cce4b68fe8b4b131960141331c6744c37cfb816710b4ab4258935c16","source":{"kind":"arxiv","id":"2109.12948","version":1},"attestation_state":"computed","paper":{"title":"Understanding and Overcoming the Challenges of Efficient Transformer Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Markus Nagel, Tijmen Blankevoort, Yelysei Bondarenko","submitted_at":"2021-09-27T10:57:18Z","abstract_excerpt":"Transformer-based architectures have become the de-facto standard models for a wide range of Natural Language Processing tasks. However, their memory footprint and high latency are prohibitive for efficient deployment and inference on resource-limited devices. In this work, we explore quantization for transformers. We show that transformers have unique quantization challenges -- namely, high dynamic activation ranges that are difficult to represent with a low bit fixed-point format. We establish that these activations contain structured outliers in the residual connections that encourage speci"},"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":"2109.12948","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2021-09-27T10:57:18Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"a354470eae49614b22c5218a72469e93a37c021a015b040b87d8aabbadb031ec","abstract_canon_sha256":"224766e4f21327ebaaa363d2ed92d9b0764871161f3dbbb94c94945b98b28478"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:17:39.183663Z","signature_b64":"lVGHhRqBQHFX/egUgm+gPWXC646drJait6kXR5ehT4Rls72OtOiOxCMFI2iEWhzTFWEyycXdcwfjyeSxMJJlBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2bef64a2cce4b68fe8b4b131960141331c6744c37cfb816710b4ab4258935c16","last_reissued_at":"2026-07-05T03:17:39.183196Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:17:39.183196Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding and Overcoming the Challenges of Efficient Transformer Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Markus Nagel, Tijmen Blankevoort, Yelysei Bondarenko","submitted_at":"2021-09-27T10:57:18Z","abstract_excerpt":"Transformer-based architectures have become the de-facto standard models for a wide range of Natural Language Processing tasks. However, their memory footprint and high latency are prohibitive for efficient deployment and inference on resource-limited devices. In this work, we explore quantization for transformers. We show that transformers have unique quantization challenges -- namely, high dynamic activation ranges that are difficult to represent with a low bit fixed-point format. We establish that these activations contain structured outliers in the residual connections that encourage speci"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2109.12948","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/2109.12948/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":"2109.12948","created_at":"2026-07-05T03:17:39.183253+00:00"},{"alias_kind":"arxiv_version","alias_value":"2109.12948v1","created_at":"2026-07-05T03:17:39.183253+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2109.12948","created_at":"2026-07-05T03:17:39.183253+00:00"},{"alias_kind":"pith_short_12","alias_value":"FPXWJIWM4S3I","created_at":"2026-07-05T03:17:39.183253+00:00"},{"alias_kind":"pith_short_16","alias_value":"FPXWJIWM4S3I72FU","created_at":"2026-07-05T03:17:39.183253+00:00"},{"alias_kind":"pith_short_8","alias_value":"FPXWJIWM","created_at":"2026-07-05T03:17:39.183253+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.01127","citing_title":"$\\text{Log}_\\text{b}$Quant: Quantizing Language Models in Logarithmic Space","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2402.17762","citing_title":"Massive Activations in Large Language Models","ref_index":107,"is_internal_anchor":false},{"citing_arxiv_id":"2208.07339","citing_title":"LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale","ref_index":120,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04701","citing_title":"MUXQ: Mixed-to-Uniform Precision MatriX Quantization via Low-Rank Outlier Decomposition","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FPXWJIWM4S3I72FUWEYZMAKBGM","json":"https://pith.science/pith/FPXWJIWM4S3I72FUWEYZMAKBGM.json","graph_json":"https://pith.science/api/pith-number/FPXWJIWM4S3I72FUWEYZMAKBGM/graph.json","events_json":"https://pith.science/api/pith-number/FPXWJIWM4S3I72FUWEYZMAKBGM/events.json","paper":"https://pith.science/paper/FPXWJIWM"},"agent_actions":{"view_html":"https://pith.science/pith/FPXWJIWM4S3I72FUWEYZMAKBGM","download_json":"https://pith.science/pith/FPXWJIWM4S3I72FUWEYZMAKBGM.json","view_paper":"https://pith.science/paper/FPXWJIWM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2109.12948&json=true","fetch_graph":"https://pith.science/api/pith-number/FPXWJIWM4S3I72FUWEYZMAKBGM/graph.json","fetch_events":"https://pith.science/api/pith-number/FPXWJIWM4S3I72FUWEYZMAKBGM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FPXWJIWM4S3I72FUWEYZMAKBGM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FPXWJIWM4S3I72FUWEYZMAKBGM/action/storage_attestation","attest_author":"https://pith.science/pith/FPXWJIWM4S3I72FUWEYZMAKBGM/action/author_attestation","sign_citation":"https://pith.science/pith/FPXWJIWM4S3I72FUWEYZMAKBGM/action/citation_signature","submit_replication":"https://pith.science/pith/FPXWJIWM4S3I72FUWEYZMAKBGM/action/replication_record"}},"created_at":"2026-07-05T03:17:39.183253+00:00","updated_at":"2026-07-05T03:17:39.183253+00:00"}