{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:BDBY4RNLPE2FSWKRB4OLHTAW6O","short_pith_number":"pith:BDBY4RNL","schema_version":"1.0","canonical_sha256":"08c38e45ab79345959510f1cb3cc16f3a0cdef58b51ecc2ca1e0256a80fdcd0b","source":{"kind":"arxiv","id":"1902.08153","version":3},"attestation_state":"computed","paper":{"title":"Learned Step Size Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Deepika Bablani, Dharmendra S. Modha, Jeffrey L. McKinstry, Rathinakumar Appuswamy, Steven K. Esser","submitted_at":"2019-02-21T17:31:32Z","abstract_excerpt":"Deep networks run with low precision operations at inference time offer power and space advantages over high precision alternatives, but need to overcome the challenge of maintaining high accuracy as precision decreases. Here, we present a method for training such networks, Learned Step Size Quantization, that achieves the highest accuracy to date on the ImageNet dataset when using models, from a variety of architectures, with weights and activations quantized to 2-, 3- or 4-bits of precision, and that can train 3-bit models that reach full precision baseline accuracy. Our approach builds upon"},"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":"1902.08153","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2019-02-21T17:31:32Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"a597708cff5278e1a025d3b17270984276b77f7d0a58713452c05255ec61ca79","abstract_canon_sha256":"6c4a135afa3125a63648bac7bc5d70736f7ddad6b8cd7a80362af021cf72ca1c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T01:00:59.405559Z","signature_b64":"AlmspCp5lb0rAlaJxGdjbUQ0c6bgq34vfWRRwPnhnjYdJyU4xXHFu6pR6828xrfK0NaqpYZAhr/Sx6L+N6/+BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"08c38e45ab79345959510f1cb3cc16f3a0cdef58b51ecc2ca1e0256a80fdcd0b","last_reissued_at":"2026-07-05T01:00:59.405086Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T01:00:59.405086Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learned Step Size Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Deepika Bablani, Dharmendra S. Modha, Jeffrey L. McKinstry, Rathinakumar Appuswamy, Steven K. Esser","submitted_at":"2019-02-21T17:31:32Z","abstract_excerpt":"Deep networks run with low precision operations at inference time offer power and space advantages over high precision alternatives, but need to overcome the challenge of maintaining high accuracy as precision decreases. Here, we present a method for training such networks, Learned Step Size Quantization, that achieves the highest accuracy to date on the ImageNet dataset when using models, from a variety of architectures, with weights and activations quantized to 2-, 3- or 4-bits of precision, and that can train 3-bit models that reach full precision baseline accuracy. Our approach builds upon"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1902.08153","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/1902.08153/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":"1902.08153","created_at":"2026-07-05T01:00:59.405140+00:00"},{"alias_kind":"arxiv_version","alias_value":"1902.08153v3","created_at":"2026-07-05T01:00:59.405140+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1902.08153","created_at":"2026-07-05T01:00:59.405140+00:00"},{"alias_kind":"pith_short_12","alias_value":"BDBY4RNLPE2F","created_at":"2026-07-05T01:00:59.405140+00:00"},{"alias_kind":"pith_short_16","alias_value":"BDBY4RNLPE2FSWKR","created_at":"2026-07-05T01:00:59.405140+00:00"},{"alias_kind":"pith_short_8","alias_value":"BDBY4RNL","created_at":"2026-07-05T01:00:59.405140+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":24,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25087","citing_title":"Neural Network Quantization by Learning Low-Loss Subspaces","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09864","citing_title":"Alignment Collapse Under KV Cache Quantization: Diagnosis and Mitigation","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31456","citing_title":"Zero-Shot Quantization for Object Detectors using Off-the-Shelf Generative Models","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24754","citing_title":"Motion-Compensated Weight Compression","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.27759","citing_title":"Layerwise Progressive Freezing: A Training Scaffold for Depth-Scalable Binary Networks","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2306.00978","citing_title":"AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2408.00923","citing_title":"Reclaiming Residual Knowledge: A Novel Paradigm to Low-Bit Quantization","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2503.03088","citing_title":"AHCQ-SAM: Toward Accurate and Hardware-Compatible Post-Training Segment Anything Model Quantization","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20706","citing_title":"Llamas on the Web: Memory-Efficient, Performance-Portable, and Multi-Precision LLM Inference with WebGPU","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16423","citing_title":"Nonlinear Bipolar Compensation: Handling Outliers in Post-Training Quantization","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17160","citing_title":"When Bits Break Recourse: Counterfactual-Faithful Quantization","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16901","citing_title":"CAR-SAM: Cross-Attention Reconstruction for Post-Training Quantization of the Segment Anything Model","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18556","citing_title":"GSQ: Highly-Accurate Low-Precision Scalar Quantization for LLMs via Gumbel-Softmax Sampling","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2509.03472","citing_title":"DPQuant: Efficient and Differentially-Private Model Training via Dynamic Quantization Scheduling","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10886","citing_title":"LoKA: Low-precision Kernel Applications for Recommendation Models At Scale","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03957","citing_title":"BWTA: Accurate and Efficient Binarized Transformer by Algorithm-Hardware Co-design","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04988","citing_title":"Prune-Quantize-Distill: An Ordered Pipeline for Efficient Neural Network Compression","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2208.07339","citing_title":"LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale","ref_index":131,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10886","citing_title":"LoKA: Low-precision Kernel Applications for Recommendation Models At Scale","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04738","citing_title":"OSAQ: Outlier Self-Absorption for Accurate Low-bit LLM Quantization","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22906","citing_title":"Network Edge Inference for Large Language Models: Principles, Techniques, and Opportunities","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01330","citing_title":"Colinearity Decay: Training Quantization-Friendly ViTs with Outlier Decay","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18556","citing_title":"GSQ: Highly-Accurate Low-Precision Scalar Quantization for LLMs via Gumbel-Softmax Sampling","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04738","citing_title":"OSAQ: Outlier Self-Absorption for Accurate Low-bit LLM Quantization","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BDBY4RNLPE2FSWKRB4OLHTAW6O","json":"https://pith.science/pith/BDBY4RNLPE2FSWKRB4OLHTAW6O.json","graph_json":"https://pith.science/api/pith-number/BDBY4RNLPE2FSWKRB4OLHTAW6O/graph.json","events_json":"https://pith.science/api/pith-number/BDBY4RNLPE2FSWKRB4OLHTAW6O/events.json","paper":"https://pith.science/paper/BDBY4RNL"},"agent_actions":{"view_html":"https://pith.science/pith/BDBY4RNLPE2FSWKRB4OLHTAW6O","download_json":"https://pith.science/pith/BDBY4RNLPE2FSWKRB4OLHTAW6O.json","view_paper":"https://pith.science/paper/BDBY4RNL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1902.08153&json=true","fetch_graph":"https://pith.science/api/pith-number/BDBY4RNLPE2FSWKRB4OLHTAW6O/graph.json","fetch_events":"https://pith.science/api/pith-number/BDBY4RNLPE2FSWKRB4OLHTAW6O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BDBY4RNLPE2FSWKRB4OLHTAW6O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BDBY4RNLPE2FSWKRB4OLHTAW6O/action/storage_attestation","attest_author":"https://pith.science/pith/BDBY4RNLPE2FSWKRB4OLHTAW6O/action/author_attestation","sign_citation":"https://pith.science/pith/BDBY4RNLPE2FSWKRB4OLHTAW6O/action/citation_signature","submit_replication":"https://pith.science/pith/BDBY4RNLPE2FSWKRB4OLHTAW6O/action/replication_record"}},"created_at":"2026-07-05T01:00:59.405140+00:00","updated_at":"2026-07-05T01:00:59.405140+00:00"}