{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:FWTSF2R3RPH5CNDJLD5FAGIKCS","short_pith_number":"pith:FWTSF2R3","schema_version":"1.0","canonical_sha256":"2da722ea3b8bcfd1346958fa50190a14ab4e320c490443eb90b9e045cd567d07","source":{"kind":"arxiv","id":"2408.11743","version":1},"attestation_state":"computed","paper":{"title":"MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dan Alistarh, Elias Frantar, Jiale Chen, Roberto L. Castro, Torsten Hoefler","submitted_at":"2024-08-21T16:10:41Z","abstract_excerpt":"As inference on Large Language Models (LLMs) emerges as an important workload in machine learning applications, weight quantization has become a standard technique for efficient GPU deployment. Quantization not only reduces model size, but has also been shown to yield substantial speedups for single-user inference, due to reduced memory movement, with low accuracy impact. Yet, it remains open whether speedups are achievable also in \\emph{batched} settings with multiple parallel clients, which are highly relevant for practical serving. It is unclear whether GPU kernels can be designed to remain"},"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":"2408.11743","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-08-21T16:10:41Z","cross_cats_sorted":[],"title_canon_sha256":"61f30ebd3afbe30c25d7c97071b42dc648e38fc8f76f8b2f91f13bad00335bb5","abstract_canon_sha256":"c6c567b1a3fc336aa8b960d8a97de9ac03adda5775b5e8b22370453e6f25d5ea"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:57:48.383149Z","signature_b64":"b0Fc4CECzM5A6ZsELdolmpI6b8hA2hnDUgeRxpjNdqBYw8Mb2A4wIpz7JftHGRCOthNmnrwhh663HeEB1YjqDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2da722ea3b8bcfd1346958fa50190a14ab4e320c490443eb90b9e045cd567d07","last_reissued_at":"2026-07-05T08:57:48.382696Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:57:48.382696Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dan Alistarh, Elias Frantar, Jiale Chen, Roberto L. Castro, Torsten Hoefler","submitted_at":"2024-08-21T16:10:41Z","abstract_excerpt":"As inference on Large Language Models (LLMs) emerges as an important workload in machine learning applications, weight quantization has become a standard technique for efficient GPU deployment. Quantization not only reduces model size, but has also been shown to yield substantial speedups for single-user inference, due to reduced memory movement, with low accuracy impact. Yet, it remains open whether speedups are achievable also in \\emph{batched} settings with multiple parallel clients, which are highly relevant for practical serving. It is unclear whether GPU kernels can be designed to remain"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.11743","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/2408.11743/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":"2408.11743","created_at":"2026-07-05T08:57:48.382752+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.11743v1","created_at":"2026-07-05T08:57:48.382752+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.11743","created_at":"2026-07-05T08:57:48.382752+00:00"},{"alias_kind":"pith_short_12","alias_value":"FWTSF2R3RPH5","created_at":"2026-07-05T08:57:48.382752+00:00"},{"alias_kind":"pith_short_16","alias_value":"FWTSF2R3RPH5CNDJ","created_at":"2026-07-05T08:57:48.382752+00:00"},{"alias_kind":"pith_short_8","alias_value":"FWTSF2R3","created_at":"2026-07-05T08:57:48.382752+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23419","citing_title":"GRINQH: Graded Input-based Quantization Hierarchy for Efficient LLM Generation","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11357","citing_title":"TileFuse: A Fused Mixed-Precision Kernel Library for Efficient Quantized LLM Inference on AMD NPUs","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11357","citing_title":"TileFuse: A Fused Mixed-Precision Kernel Library for Efficient Quantized LLM Inference on AMD NPUs","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09682","citing_title":"AutoMegaKernel: A Statically-Checked Agent Harness for Self-Retargeting Megakernel Synthesis","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08094","citing_title":"vla.cpp: A Unified Inference Runtime for Vision-Language-Action Models","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2602.13595","citing_title":"The Quantization Trap: Breaking Linear Scaling Laws in Multi-Hop Reasoning","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13915","citing_title":"Multi-Scale Dequant: Eliminating Dequantization Bottleneck via Activation Decomposition for Efficient LLM Inference","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02404","citing_title":"Statistically-Lossless Quantization of Large Language Models","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13806","citing_title":"Robust Ultra Low-Bit Post-Training Quantization via Stable Diagonal Curvature Estimate","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21026","citing_title":"MCAP: Deployment-Time Layer Profiling for Memory-Constrained LLM Inference","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20079","citing_title":"On the Quantization Robustness of Diffusion Language Models in Coding Benchmarks","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FWTSF2R3RPH5CNDJLD5FAGIKCS","json":"https://pith.science/pith/FWTSF2R3RPH5CNDJLD5FAGIKCS.json","graph_json":"https://pith.science/api/pith-number/FWTSF2R3RPH5CNDJLD5FAGIKCS/graph.json","events_json":"https://pith.science/api/pith-number/FWTSF2R3RPH5CNDJLD5FAGIKCS/events.json","paper":"https://pith.science/paper/FWTSF2R3"},"agent_actions":{"view_html":"https://pith.science/pith/FWTSF2R3RPH5CNDJLD5FAGIKCS","download_json":"https://pith.science/pith/FWTSF2R3RPH5CNDJLD5FAGIKCS.json","view_paper":"https://pith.science/paper/FWTSF2R3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.11743&json=true","fetch_graph":"https://pith.science/api/pith-number/FWTSF2R3RPH5CNDJLD5FAGIKCS/graph.json","fetch_events":"https://pith.science/api/pith-number/FWTSF2R3RPH5CNDJLD5FAGIKCS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FWTSF2R3RPH5CNDJLD5FAGIKCS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FWTSF2R3RPH5CNDJLD5FAGIKCS/action/storage_attestation","attest_author":"https://pith.science/pith/FWTSF2R3RPH5CNDJLD5FAGIKCS/action/author_attestation","sign_citation":"https://pith.science/pith/FWTSF2R3RPH5CNDJLD5FAGIKCS/action/citation_signature","submit_replication":"https://pith.science/pith/FWTSF2R3RPH5CNDJLD5FAGIKCS/action/replication_record"}},"created_at":"2026-07-05T08:57:48.382752+00:00","updated_at":"2026-07-05T08:57:48.382752+00:00"}