{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:XAEMDIAUA7KOZHOIRNT2ZBHMCL","short_pith_number":"pith:XAEMDIAU","schema_version":"1.0","canonical_sha256":"b808c1a01407d4ec9dc88b67ac84ec12da1303e402a75fbb5b0f2a9dc4bd8a48","source":{"kind":"arxiv","id":"2402.04396","version":2},"attestation_state":"computed","paper":{"title":"QuIP#: Even Better LLM Quantization with Hadamard Incoherence and Lattice Codebooks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Albert Tseng, Christopher De Sa, Jerry Chee, Qingyao Sun, Volodymyr Kuleshov","submitted_at":"2024-02-06T20:52:12Z","abstract_excerpt":"Post-training quantization (PTQ) reduces the memory footprint of LLMs by quantizing their weights to low-precision. In this work, we introduce QuIP#, a weight-only PTQ method that achieves state-of-the-art results in extreme compression regimes ($\\le$ 4 bits per weight) using three novel techniques. First, QuIP# improves QuIP's (Chee et al., 2023) incoherence processing by using the randomized Hadamard transform, which is faster and has better theoretical properties. Second, QuIP# uses vector quantization to take advantage of the ball-shaped sub-Gaussian distribution that incoherent weights po"},"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":"2402.04396","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-06T20:52:12Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"a8576837819e7c257c4494d170422e9bbd0ab64dc0076d55e8354031ec6ae712","abstract_canon_sha256":"af1cfcff7e770efae1a689a173d81b9c0502e8e8b0f54c0a12f5d32e65c4178f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:27:03.282410Z","signature_b64":"8fcdukow5WkrwY25A4MetqZWW1cdQTTEcZTRB5FTyeROvTK98ExJDuxIx2PBcqvhk+z1Dz/kPuO8eM40xsumBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b808c1a01407d4ec9dc88b67ac84ec12da1303e402a75fbb5b0f2a9dc4bd8a48","last_reissued_at":"2026-07-05T08:27:03.281971Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:27:03.281971Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"QuIP#: Even Better LLM Quantization with Hadamard Incoherence and Lattice Codebooks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Albert Tseng, Christopher De Sa, Jerry Chee, Qingyao Sun, Volodymyr Kuleshov","submitted_at":"2024-02-06T20:52:12Z","abstract_excerpt":"Post-training quantization (PTQ) reduces the memory footprint of LLMs by quantizing their weights to low-precision. In this work, we introduce QuIP#, a weight-only PTQ method that achieves state-of-the-art results in extreme compression regimes ($\\le$ 4 bits per weight) using three novel techniques. First, QuIP# improves QuIP's (Chee et al., 2023) incoherence processing by using the randomized Hadamard transform, which is faster and has better theoretical properties. Second, QuIP# uses vector quantization to take advantage of the ball-shaped sub-Gaussian distribution that incoherent weights po"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.04396","kind":"arxiv","version":2},"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/2402.04396/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":"2402.04396","created_at":"2026-07-05T08:27:03.282029+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.04396v2","created_at":"2026-07-05T08:27:03.282029+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.04396","created_at":"2026-07-05T08:27:03.282029+00:00"},{"alias_kind":"pith_short_12","alias_value":"XAEMDIAUA7KO","created_at":"2026-07-05T08:27:03.282029+00:00"},{"alias_kind":"pith_short_16","alias_value":"XAEMDIAUA7KOZHOI","created_at":"2026-07-05T08:27:03.282029+00:00"},{"alias_kind":"pith_short_8","alias_value":"XAEMDIAU","created_at":"2026-07-05T08:27:03.282029+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":35,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23419","citing_title":"GRINQH: Graded Input-based Quantization Hierarchy for Efficient LLM Generation","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2606.23406","citing_title":"HyperQuant: A Rate-Distortion-Optimal Quantization Pipeline for Large Language and Diffusion Models","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06547","citing_title":"FAIR-Calib: Frontier-Aware Instability-Reweighted Calibration for Post-Training Quantization of Diffusion Large Language Models","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05429","citing_title":"Minimizing the Hidden Cost of Scales: Graph-Guided Ultra-Low-Bit Quantization for Large Language Models","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04050","citing_title":"LiftQuant: Continuous Bit-Width LLM via Dimensional Lifting and Projection","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02823","citing_title":"Qift: Shift-Friendly No-Zero W2 Post-Training Quantization for Rotated W2A4/KV4 LLM Inference","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00206","citing_title":"Quantized Reasoning Models Think They Need to Think Longer, but They Do Not","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24019","citing_title":"MGVQ: Synergizing Multi-dimensional Sensitivity-Aware and Gradient-Hessian Fusion for Vector Quantization","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13768","citing_title":"High-Rate Quantized Matrix Multiplication II","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14844","citing_title":"XFP: Quality-Targeted Adaptive Codebook Quantization with Sparse Outlier Separation for LLM Inference","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24144","citing_title":"EVA: Accelerating LLM Decoding via an Efficient Vector Quantization Architecture","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25203","citing_title":"Influence-Inspired Spectral Rotations for Extreme Low-Bit LLM Quantization","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04050","citing_title":"LiftQuant: Continuous Bit-Width LLM via Dimensional Lifting and Projection","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00573","citing_title":"LASER: Loss-Aware Singular-value Decomposition and Rank Allocation for Efficient Low-Precision Vision-Language Models","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00535","citing_title":"DREAM-S: Speculative Decoding with Searchable Drafting and Target-Aware Refinement for Multimodal Generation","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23078","citing_title":"GEMQ: Global Expert-Level Mixed-Precision Quantization for MoE LLMs","ref_index":25,"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":63,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18800","citing_title":"Theory-optimal Quantization Based on Flatness","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2407.08608","citing_title":"FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18475","citing_title":"GAMMA: Global Bit Allocation for Mixed-Precision Models under Arbitrary Budgets","ref_index":56,"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":30,"is_internal_anchor":false},{"citing_arxiv_id":"2506.12040","citing_title":"BTC-LLM: Efficient Sub-1-Bit LLM Quantization via Learnable Transformation and Binary Codebook","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2508.04853","citing_title":"Provable Post-Training Quantization: Theoretical Analysis of OPTQ and Qronos","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2402.17764","citing_title":"The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2601.17187","citing_title":"High-Rate Quantized Matrix Multiplication I","ref_index":35,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XAEMDIAUA7KOZHOIRNT2ZBHMCL","json":"https://pith.science/pith/XAEMDIAUA7KOZHOIRNT2ZBHMCL.json","graph_json":"https://pith.science/api/pith-number/XAEMDIAUA7KOZHOIRNT2ZBHMCL/graph.json","events_json":"https://pith.science/api/pith-number/XAEMDIAUA7KOZHOIRNT2ZBHMCL/events.json","paper":"https://pith.science/paper/XAEMDIAU"},"agent_actions":{"view_html":"https://pith.science/pith/XAEMDIAUA7KOZHOIRNT2ZBHMCL","download_json":"https://pith.science/pith/XAEMDIAUA7KOZHOIRNT2ZBHMCL.json","view_paper":"https://pith.science/paper/XAEMDIAU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.04396&json=true","fetch_graph":"https://pith.science/api/pith-number/XAEMDIAUA7KOZHOIRNT2ZBHMCL/graph.json","fetch_events":"https://pith.science/api/pith-number/XAEMDIAUA7KOZHOIRNT2ZBHMCL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XAEMDIAUA7KOZHOIRNT2ZBHMCL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XAEMDIAUA7KOZHOIRNT2ZBHMCL/action/storage_attestation","attest_author":"https://pith.science/pith/XAEMDIAUA7KOZHOIRNT2ZBHMCL/action/author_attestation","sign_citation":"https://pith.science/pith/XAEMDIAUA7KOZHOIRNT2ZBHMCL/action/citation_signature","submit_replication":"https://pith.science/pith/XAEMDIAUA7KOZHOIRNT2ZBHMCL/action/replication_record"}},"created_at":"2026-07-05T08:27:03.282029+00:00","updated_at":"2026-07-05T08:27:03.282029+00:00"}