{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:SLOQYERK24JWE3YJNJ5HQEU5OZ","short_pith_number":"pith:SLOQYERK","schema_version":"1.0","canonical_sha256":"92dd0c122ad713626f096a7a78129d7671e678d3b3bb4b3ad61802b1d364326b","source":{"kind":"arxiv","id":"2407.14679","version":2},"attestation_state":"computed","paper":{"title":"Compact Language Models via Pruning and Knowledge Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bryan Catanzaro, Jan Kautz, Marcin Chochowski, Mohammad Shoeybi, Mostofa Patwary, Pavlo Molchanov, Raviraj Joshi, Saurav Muralidharan, Sharath Turuvekere Sreenivas","submitted_at":"2024-07-19T21:47:57Z","abstract_excerpt":"Large language models (LLMs) targeting different deployment scales and sizes are currently produced by training each variant from scratch; this is extremely compute-intensive. In this paper, we investigate if pruning an existing LLM and then re-training it with a fraction (<3%) of the original training data can be a suitable alternative to repeated, full retraining. To this end, we develop a set of practical and effective compression best practices for LLMs that combine depth, width, attention and MLP pruning with knowledge distillation-based retraining; we arrive at these best practices throu"},"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":"2407.14679","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-19T21:47:57Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"c867a888aaa20b33c3f81dcef656f7e7d6d711d081f3b8ef5284ba319f5335e0","abstract_canon_sha256":"82f1ae514c88363647c8b000e1faf3386dd97387ef611a4b2cb6612222c207a1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:30:49.437017Z","signature_b64":"qARTv8XZoC1KEd/S0XmzbM/7WaCAUDp5bgpOtDL6ZKhuwKve0wKFGBGwWJRsNIXP9LtoZQ5zCHqGr+VD8GZqCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"92dd0c122ad713626f096a7a78129d7671e678d3b3bb4b3ad61802b1d364326b","last_reissued_at":"2026-07-05T09:30:49.436523Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:30:49.436523Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Compact Language Models via Pruning and Knowledge Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bryan Catanzaro, Jan Kautz, Marcin Chochowski, Mohammad Shoeybi, Mostofa Patwary, Pavlo Molchanov, Raviraj Joshi, Saurav Muralidharan, Sharath Turuvekere Sreenivas","submitted_at":"2024-07-19T21:47:57Z","abstract_excerpt":"Large language models (LLMs) targeting different deployment scales and sizes are currently produced by training each variant from scratch; this is extremely compute-intensive. In this paper, we investigate if pruning an existing LLM and then re-training it with a fraction (<3%) of the original training data can be a suitable alternative to repeated, full retraining. To this end, we develop a set of practical and effective compression best practices for LLMs that combine depth, width, attention and MLP pruning with knowledge distillation-based retraining; we arrive at these best practices throu"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.14679","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/2407.14679/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":"2407.14679","created_at":"2026-07-05T09:30:49.436583+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.14679v2","created_at":"2026-07-05T09:30:49.436583+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.14679","created_at":"2026-07-05T09:30:49.436583+00:00"},{"alias_kind":"pith_short_12","alias_value":"SLOQYERK24JW","created_at":"2026-07-05T09:30:49.436583+00:00"},{"alias_kind":"pith_short_16","alias_value":"SLOQYERK24JWE3YJ","created_at":"2026-07-05T09:30:49.436583+00:00"},{"alias_kind":"pith_short_8","alias_value":"SLOQYERK","created_at":"2026-07-05T09:30:49.436583+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24747","citing_title":"Scaling Laws for Task-Specific LLM Distillation","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04063","citing_title":"LLM Compression with Jointly Optimizing Architectural and Quantization choices","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28207","citing_title":"Pruning and Distilling Mixture-of-Experts into Dense Language Models","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07604","citing_title":"Contribution Weights: A Geometrical Analysis of Self-Attention Transformers","ref_index":100,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08738","citing_title":"SlimQwen: Exploring the Pruning and Distillation in Large MoE Model Pre-training","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07721","citing_title":"Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18860","citing_title":"Spectral structural distortion reveals redundant neurons in neural networks","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2512.20856","citing_title":"NVIDIA Nemotron 3: Efficient and Open Intelligence","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2512.22671","citing_title":"Fragile Knowledge, Robust Instruction-Following: The Width Pruning Dichotomy in Llama-3.2","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2602.20816","citing_title":"Don't Ignore the Tail: Decoupling top-K Probabilities for Efficient Language Model Distillation","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2601.08584","citing_title":"Ministral 3","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11999","citing_title":"The Illusion of Power Capping in LLM Decode: A Phase-Aware Energy Characterisation Across Attention Architectures","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08738","citing_title":"SlimQwen: Exploring the Pruning and Distillation in Large MoE Model Pre-training","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07182","citing_title":"Star Elastic: Many-in-One Reasoning LLMs with Efficient Budget Control","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07721","citing_title":"Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models","ref_index":30,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SLOQYERK24JWE3YJNJ5HQEU5OZ","json":"https://pith.science/pith/SLOQYERK24JWE3YJNJ5HQEU5OZ.json","graph_json":"https://pith.science/api/pith-number/SLOQYERK24JWE3YJNJ5HQEU5OZ/graph.json","events_json":"https://pith.science/api/pith-number/SLOQYERK24JWE3YJNJ5HQEU5OZ/events.json","paper":"https://pith.science/paper/SLOQYERK"},"agent_actions":{"view_html":"https://pith.science/pith/SLOQYERK24JWE3YJNJ5HQEU5OZ","download_json":"https://pith.science/pith/SLOQYERK24JWE3YJNJ5HQEU5OZ.json","view_paper":"https://pith.science/paper/SLOQYERK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.14679&json=true","fetch_graph":"https://pith.science/api/pith-number/SLOQYERK24JWE3YJNJ5HQEU5OZ/graph.json","fetch_events":"https://pith.science/api/pith-number/SLOQYERK24JWE3YJNJ5HQEU5OZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SLOQYERK24JWE3YJNJ5HQEU5OZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SLOQYERK24JWE3YJNJ5HQEU5OZ/action/storage_attestation","attest_author":"https://pith.science/pith/SLOQYERK24JWE3YJNJ5HQEU5OZ/action/author_attestation","sign_citation":"https://pith.science/pith/SLOQYERK24JWE3YJNJ5HQEU5OZ/action/citation_signature","submit_replication":"https://pith.science/pith/SLOQYERK24JWE3YJNJ5HQEU5OZ/action/replication_record"}},"created_at":"2026-07-05T09:30:49.436583+00:00","updated_at":"2026-07-05T09:30:49.436583+00:00"}