{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:TJDVYDGP6Y44V7N2NJGREJ4AJF","short_pith_number":"pith:TJDVYDGP","schema_version":"1.0","canonical_sha256":"9a475c0ccff639cafdba6a4d122780497b9c98ca64b9437ee9dc6fa8d090f5cd","source":{"kind":"arxiv","id":"2401.06118","version":4},"attestation_state":"computed","paper":{"title":"Extreme Compression of Large Language Models via Additive Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Andrei Panferov, Artem Babenko, Dan Alistarh, Denis Kuznedelev, Elias Frantar, Vage Egiazarian","submitted_at":"2024-01-11T18:54:44Z","abstract_excerpt":"The emergence of accurate open large language models (LLMs) has led to a race towards performant quantization techniques which can enable their execution on end-user devices. In this paper, we revisit the problem of \"extreme\" LLM compression-defined as targeting extremely low bit counts, such as 2 to 3 bits per parameter-from the point of view of classic methods in Multi-Codebook Quantization (MCQ). Our algorithm, called AQLM, generalizes the classic Additive Quantization (AQ) approach for information retrieval to advance the state-of-the-art in LLM compression, via two innovations: 1) learned"},"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":"2401.06118","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-11T18:54:44Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"a045a3c833b486ccc38552de7f4a853d602ddd24faf592a723b0e4ab1155fa37","abstract_canon_sha256":"3be88b68a69df916863dfd2072f712bd8a74c8c57c75264d32f9ab69a8647cd9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:05:33.267013Z","signature_b64":"Iii0ltMU1/7q+WZT6vnd2MVJRLPl47XRJMUzd0oB4tZPIgg8unFVio3hNrcbNxSQNMdlm/uOQk5lryjPsMhxAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9a475c0ccff639cafdba6a4d122780497b9c98ca64b9437ee9dc6fa8d090f5cd","last_reissued_at":"2026-07-05T09:05:33.266505Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:05:33.266505Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Extreme Compression of Large Language Models via Additive Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Andrei Panferov, Artem Babenko, Dan Alistarh, Denis Kuznedelev, Elias Frantar, Vage Egiazarian","submitted_at":"2024-01-11T18:54:44Z","abstract_excerpt":"The emergence of accurate open large language models (LLMs) has led to a race towards performant quantization techniques which can enable their execution on end-user devices. In this paper, we revisit the problem of \"extreme\" LLM compression-defined as targeting extremely low bit counts, such as 2 to 3 bits per parameter-from the point of view of classic methods in Multi-Codebook Quantization (MCQ). Our algorithm, called AQLM, generalizes the classic Additive Quantization (AQ) approach for information retrieval to advance the state-of-the-art in LLM compression, via two innovations: 1) learned"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.06118","kind":"arxiv","version":4},"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/2401.06118/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":"2401.06118","created_at":"2026-07-05T09:05:33.266566+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.06118v4","created_at":"2026-07-05T09:05:33.266566+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.06118","created_at":"2026-07-05T09:05:33.266566+00:00"},{"alias_kind":"pith_short_12","alias_value":"TJDVYDGP6Y44","created_at":"2026-07-05T09:05:33.266566+00:00"},{"alias_kind":"pith_short_16","alias_value":"TJDVYDGP6Y44V7N2","created_at":"2026-07-05T09:05:33.266566+00:00"},{"alias_kind":"pith_short_8","alias_value":"TJDVYDGP","created_at":"2026-07-05T09:05:33.266566+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":27,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.27205","citing_title":"Smaller Models, Unexpected Costs: Trade-offs in LLM Quantization for Automated Program Repair","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2606.24937","citing_title":"The Hitchhiker's Guide to Agentic AI: From Foundations to Systems","ref_index":150,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12876","citing_title":"Multi-Bitwidth Quantization for LLMs Using Additive Codebooks","ref_index":67,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04050","citing_title":"LiftQuant: Continuous Bit-Width LLM via Dimensional Lifting and Projection","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03465","citing_title":"Rethinking the Role of Tensor Decompositions in Post-Training LLM Compression","ref_index":7,"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":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24011","citing_title":"ActQuant: Sub-4-bit Action-Guided Quantization for Vision-Language-Action Models","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25203","citing_title":"Influence-Inspired Spectral Rotations for Extreme Low-Bit LLM Quantization","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04050","citing_title":"LiftQuant: Continuous Bit-Width LLM via Dimensional Lifting and Projection","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29239","citing_title":"Breaking the Rounding Trap: Securing LLMs against Quantization-Conditioned Backdoors","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26660","citing_title":"WINDQuant: Weight-Informed Neural Decision-Making for Global Mixed-Precision LLM Quantization","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29843","citing_title":"HARP: Hadamard-Preconditioned Adaptive Rotation Processor for Extreme LLM Quantization","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2512.04746","citing_title":"SignRoundV2: Toward Closing the Performance Gap in Extremely Low-Bit Post-Training Quantization for LLMs","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2602.04163","citing_title":"BPDQ: Bit-Plane Decomposition Quantization on a Variable Grid for Large Language Models","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18475","citing_title":"GAMMA: Global Bit Allocation for Mixed-Precision Models under Arbitrary Budgets","ref_index":45,"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":9,"is_internal_anchor":false},{"citing_arxiv_id":"2508.04853","citing_title":"Provable Post-Training Quantization: Theoretical Analysis of OPTQ and Qronos","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2509.22166","citing_title":"Motivating Next-Gen Accelerators with Flexible (N:M) Activation Sparsity via Benchmarking Lightweight Post-Training Sparsification Approaches","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2602.05902","citing_title":"CoreQ: Learning-Free Mismatch Correction and Successive Rounding for Quantization","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2602.22575","citing_title":"S2O: Early Stopping for Sparse Attention via Online Permutation","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2405.16406","citing_title":"SpinQuant: LLM quantization with learned rotations","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02501","citing_title":"ECG Foundation Models and Medical LLMs for Agentic Cardiovascular Intelligence at the Edge: A Review and Outlook","ref_index":99,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19167","citing_title":"LBLLM: Lightweight Binarization of Large Language Models via Three-Stage Distillation","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2403.13372","citing_title":"LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06916","citing_title":"FP4 Explore, BF16 Train: Diffusion Reinforcement Learning via Efficient Rollout Scaling","ref_index":54,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TJDVYDGP6Y44V7N2NJGREJ4AJF","json":"https://pith.science/pith/TJDVYDGP6Y44V7N2NJGREJ4AJF.json","graph_json":"https://pith.science/api/pith-number/TJDVYDGP6Y44V7N2NJGREJ4AJF/graph.json","events_json":"https://pith.science/api/pith-number/TJDVYDGP6Y44V7N2NJGREJ4AJF/events.json","paper":"https://pith.science/paper/TJDVYDGP"},"agent_actions":{"view_html":"https://pith.science/pith/TJDVYDGP6Y44V7N2NJGREJ4AJF","download_json":"https://pith.science/pith/TJDVYDGP6Y44V7N2NJGREJ4AJF.json","view_paper":"https://pith.science/paper/TJDVYDGP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.06118&json=true","fetch_graph":"https://pith.science/api/pith-number/TJDVYDGP6Y44V7N2NJGREJ4AJF/graph.json","fetch_events":"https://pith.science/api/pith-number/TJDVYDGP6Y44V7N2NJGREJ4AJF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TJDVYDGP6Y44V7N2NJGREJ4AJF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TJDVYDGP6Y44V7N2NJGREJ4AJF/action/storage_attestation","attest_author":"https://pith.science/pith/TJDVYDGP6Y44V7N2NJGREJ4AJF/action/author_attestation","sign_citation":"https://pith.science/pith/TJDVYDGP6Y44V7N2NJGREJ4AJF/action/citation_signature","submit_replication":"https://pith.science/pith/TJDVYDGP6Y44V7N2NJGREJ4AJF/action/replication_record"}},"created_at":"2026-07-05T09:05:33.266566+00:00","updated_at":"2026-07-05T09:05:33.266566+00:00"}