{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:Q7ASRHN6SCKAWFSKM2XTKE43EL","short_pith_number":"pith:Q7ASRHN6","canonical_record":{"source":{"id":"2509.22075","kind":"arxiv","version":6},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-09-26T08:55:09Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"d6e17c26284705680c617022401148f8074164ae5ff3afd6c70469945f8f3af8","abstract_canon_sha256":"bc82722b17729c3ca381508862d7ffaa450d2d34c2fcc91ee4412e4685255600"},"schema_version":"1.0"},"canonical_sha256":"87c1289dbe90940b164a66af35139b22eb037a47630948cdb0bd9f968d26af09","source":{"kind":"arxiv","id":"2509.22075","version":6},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2509.22075","created_at":"2026-06-23T01:11:57Z"},{"alias_kind":"arxiv_version","alias_value":"2509.22075v6","created_at":"2026-06-23T01:11:57Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.22075","created_at":"2026-06-23T01:11:57Z"},{"alias_kind":"pith_short_12","alias_value":"Q7ASRHN6SCKA","created_at":"2026-06-23T01:11:57Z"},{"alias_kind":"pith_short_16","alias_value":"Q7ASRHN6SCKAWFSK","created_at":"2026-06-23T01:11:57Z"},{"alias_kind":"pith_short_8","alias_value":"Q7ASRHN6","created_at":"2026-06-23T01:11:57Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:Q7ASRHN6SCKAWFSKM2XTKE43EL","target":"record","payload":{"canonical_record":{"source":{"id":"2509.22075","kind":"arxiv","version":6},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-09-26T08:55:09Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"d6e17c26284705680c617022401148f8074164ae5ff3afd6c70469945f8f3af8","abstract_canon_sha256":"bc82722b17729c3ca381508862d7ffaa450d2d34c2fcc91ee4412e4685255600"},"schema_version":"1.0"},"canonical_sha256":"87c1289dbe90940b164a66af35139b22eb037a47630948cdb0bd9f968d26af09","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T01:11:57.156794Z","signature_b64":"MSPKpzwjX3rDf60mHlWMh13+Z/O8cdTkSd//EEqG5HU0zq69OH/1bCJO5NZ9zaCPWudNdI55gCRQrzYb3CWGDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"87c1289dbe90940b164a66af35139b22eb037a47630948cdb0bd9f968d26af09","last_reissued_at":"2026-06-23T01:11:57.156174Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T01:11:57.156174Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2509.22075","source_version":6,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-23T01:11:57Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"56nnJmBB0mdKEUg5IAg6ve9L7ZwLqlUzjy1WScHGBfKzh1V4cxrAp+DpOn2Musv/aQoQY/7j8iQ1xhES+RsmCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-30T18:53:38.381087Z"},"content_sha256":"f11073877094f1625735f6893dc2cd239e8d798206bdc04adce14c0ff177c08b","schema_version":"1.0","event_id":"sha256:f11073877094f1625735f6893dc2cd239e8d798206bdc04adce14c0ff177c08b"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:Q7ASRHN6SCKAWFSKM2XTKE43EL","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"CoSpaDi: Compressing LLMs via Calibration-Guided Sparse Dictionary Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"CoSpaDi replaces low-rank factorization with a sparse dictionary model that better preserves LLM accuracy at 20-40 percent compression.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Ammar Ali, Denis Makhov, Dmitriy Shopkhoev, Magauiya Zhussip, Stamatios Lefkimmiatis","submitted_at":"2025-09-26T08:55:09Z","abstract_excerpt":"Post-training LLM compression often relies on low-rank approximations, which force all columns of a projection matrix to share a single low-dimensional subspace. We propose CoSpaDi, a training-free compression framework that replaces this single-subspace assumption with a union-of-subspaces model via sparse dictionary learning. CoSpaDi factorizes each weight matrix into a dense dictionary and column-sparse coefficients, allowing different columns to select different subsets of dictionary atoms at the same storage budget. To preserve model behavior, we use calibration activations to transform f"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across Llama and Qwen model families, CoSpaDi consistently improves the accuracy-compression and perplexity-compression trade-offs over state-of-the-art SVD-based baselines and strong structured pruning baselines at 20-40% compression ratios.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that minimizing functional reconstruction error on a small calibration set will produce a factorization whose downstream task performance remains close to the original model without any fine-tuning or further adaptation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CoSpaDi introduces a training-free sparse dictionary learning framework for post-training LLM compression that optimizes functional reconstruction error via activation-derived orthonormalization and achieves improved accuracy-compression trade-offs over SVD and pruning baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"CoSpaDi replaces low-rank factorization with a sparse dictionary model that better preserves LLM accuracy at 20-40 percent compression.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f7e2648c9a5527e0cfc6f00a30dade4184bacec6357a9e3215bb0202c79939f6"},"source":{"id":"2509.22075","kind":"arxiv","version":6},"verdict":{"id":"5e9533e4-fa16-4a29-9e32-8f1fd8b16999","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-18T13:14:45.608935Z","strongest_claim":"Across Llama and Qwen model families, CoSpaDi consistently improves the accuracy-compression and perplexity-compression trade-offs over state-of-the-art SVD-based baselines and strong structured pruning baselines at 20-40% compression ratios.","one_line_summary":"CoSpaDi introduces a training-free sparse dictionary learning framework for post-training LLM compression that optimizes functional reconstruction error via activation-derived orthonormalization and achieves improved accuracy-compression trade-offs over SVD and pruning baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that minimizing functional reconstruction error on a small calibration set will produce a factorization whose downstream task performance remains close to the original model without any fine-tuning or further adaptation.","pith_extraction_headline":"CoSpaDi replaces low-rank factorization with a sparse dictionary model that better preserves LLM accuracy at 20-40 percent compression."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2509.22075/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"},"verdict_id":"5e9533e4-fa16-4a29-9e32-8f1fd8b16999"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-23T01:11:57Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"sCrO/N3LAiPt7Sbi1m5vNy0QowiRrryAs/kOREIF/DuPg6Ghs8iDHTpF455H0eeL+TxDU0jWpS/CVKn/0PZLCA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-30T18:53:38.381582Z"},"content_sha256":"64d0cf59d234c158175ee346db05f06c20d80c07ea13d2e3636eee3e51c633f2","schema_version":"1.0","event_id":"sha256:64d0cf59d234c158175ee346db05f06c20d80c07ea13d2e3636eee3e51c633f2"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/Q7ASRHN6SCKAWFSKM2XTKE43EL/bundle.json","state_url":"https://pith.science/pith/Q7ASRHN6SCKAWFSKM2XTKE43EL/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/Q7ASRHN6SCKAWFSKM2XTKE43EL/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-30T18:53:38Z","links":{"resolver":"https://pith.science/pith/Q7ASRHN6SCKAWFSKM2XTKE43EL","bundle":"https://pith.science/pith/Q7ASRHN6SCKAWFSKM2XTKE43EL/bundle.json","state":"https://pith.science/pith/Q7ASRHN6SCKAWFSKM2XTKE43EL/state.json","well_known_bundle":"https://pith.science/.well-known/pith/Q7ASRHN6SCKAWFSKM2XTKE43EL/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:Q7ASRHN6SCKAWFSKM2XTKE43EL","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"bc82722b17729c3ca381508862d7ffaa450d2d34c2fcc91ee4412e4685255600","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-09-26T08:55:09Z","title_canon_sha256":"d6e17c26284705680c617022401148f8074164ae5ff3afd6c70469945f8f3af8"},"schema_version":"1.0","source":{"id":"2509.22075","kind":"arxiv","version":6}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2509.22075","created_at":"2026-06-23T01:11:57Z"},{"alias_kind":"arxiv_version","alias_value":"2509.22075v6","created_at":"2026-06-23T01:11:57Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.22075","created_at":"2026-06-23T01:11:57Z"},{"alias_kind":"pith_short_12","alias_value":"Q7ASRHN6SCKA","created_at":"2026-06-23T01:11:57Z"},{"alias_kind":"pith_short_16","alias_value":"Q7ASRHN6SCKAWFSK","created_at":"2026-06-23T01:11:57Z"},{"alias_kind":"pith_short_8","alias_value":"Q7ASRHN6","created_at":"2026-06-23T01:11:57Z"}],"graph_snapshots":[{"event_id":"sha256:64d0cf59d234c158175ee346db05f06c20d80c07ea13d2e3636eee3e51c633f2","target":"graph","created_at":"2026-06-23T01:11:57Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Across Llama and Qwen model families, CoSpaDi consistently improves the accuracy-compression and perplexity-compression trade-offs over state-of-the-art SVD-based baselines and strong structured pruning baselines at 20-40% compression ratios."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The assumption that minimizing functional reconstruction error on a small calibration set will produce a factorization whose downstream task performance remains close to the original model without any fine-tuning or further adaptation."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"CoSpaDi introduces a training-free sparse dictionary learning framework for post-training LLM compression that optimizes functional reconstruction error via activation-derived orthonormalization and achieves improved accuracy-compression trade-offs over SVD and pruning baselines."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"CoSpaDi replaces low-rank factorization with a sparse dictionary model that better preserves LLM accuracy at 20-40 percent compression."}],"snapshot_sha256":"f7e2648c9a5527e0cfc6f00a30dade4184bacec6357a9e3215bb0202c79939f6"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2509.22075/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Post-training LLM compression often relies on low-rank approximations, which force all columns of a projection matrix to share a single low-dimensional subspace. We propose CoSpaDi, a training-free compression framework that replaces this single-subspace assumption with a union-of-subspaces model via sparse dictionary learning. CoSpaDi factorizes each weight matrix into a dense dictionary and column-sparse coefficients, allowing different columns to select different subsets of dictionary atoms at the same storage budget. To preserve model behavior, we use calibration activations to transform f","authors_text":"Ammar Ali, Denis Makhov, Dmitriy Shopkhoev, Magauiya Zhussip, Stamatios Lefkimmiatis","cross_cats":["cs.AI"],"headline":"CoSpaDi replaces low-rank factorization with a sparse dictionary model that better preserves LLM accuracy at 20-40 percent compression.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-09-26T08:55:09Z","title":"CoSpaDi: Compressing LLMs via Calibration-Guided Sparse Dictionary Learning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.22075","kind":"arxiv","version":6},"verdict":{"created_at":"2026-05-18T13:14:45.608935Z","id":"5e9533e4-fa16-4a29-9e32-8f1fd8b16999","model_set":{"reader":"grok-4.3"},"one_line_summary":"CoSpaDi introduces a training-free sparse dictionary learning framework for post-training LLM compression that optimizes functional reconstruction error via activation-derived orthonormalization and achieves improved accuracy-compression trade-offs over SVD and pruning baselines.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"CoSpaDi replaces low-rank factorization with a sparse dictionary model that better preserves LLM accuracy at 20-40 percent compression.","strongest_claim":"Across Llama and Qwen model families, CoSpaDi consistently improves the accuracy-compression and perplexity-compression trade-offs over state-of-the-art SVD-based baselines and strong structured pruning baselines at 20-40% compression ratios.","weakest_assumption":"The assumption that minimizing functional reconstruction error on a small calibration set will produce a factorization whose downstream task performance remains close to the original model without any fine-tuning or further adaptation."}},"verdict_id":"5e9533e4-fa16-4a29-9e32-8f1fd8b16999"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:f11073877094f1625735f6893dc2cd239e8d798206bdc04adce14c0ff177c08b","target":"record","created_at":"2026-06-23T01:11:57Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"bc82722b17729c3ca381508862d7ffaa450d2d34c2fcc91ee4412e4685255600","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-09-26T08:55:09Z","title_canon_sha256":"d6e17c26284705680c617022401148f8074164ae5ff3afd6c70469945f8f3af8"},"schema_version":"1.0","source":{"id":"2509.22075","kind":"arxiv","version":6}},"canonical_sha256":"87c1289dbe90940b164a66af35139b22eb037a47630948cdb0bd9f968d26af09","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"87c1289dbe90940b164a66af35139b22eb037a47630948cdb0bd9f968d26af09","first_computed_at":"2026-06-23T01:11:57.156174Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-23T01:11:57.156174Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"MSPKpzwjX3rDf60mHlWMh13+Z/O8cdTkSd//EEqG5HU0zq69OH/1bCJO5NZ9zaCPWudNdI55gCRQrzYb3CWGDA==","signature_status":"signed_v1","signed_at":"2026-06-23T01:11:57.156794Z","signed_message":"canonical_sha256_bytes"},"source_id":"2509.22075","source_kind":"arxiv","source_version":6}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:f11073877094f1625735f6893dc2cd239e8d798206bdc04adce14c0ff177c08b","sha256:64d0cf59d234c158175ee346db05f06c20d80c07ea13d2e3636eee3e51c633f2"],"state_sha256":"747019df65c8a6c36dd5f7e516fdd88259bceadbf17d4dcbdf7db456a1b44799"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"3fvJ6dTf8qD4r7+3NNDT5ompIVyMnqkPkb0sGBDVFvSoSd9EXnq2zL9Pfsqu0nIVGizodZAwC3DpGK3Eo5KLDg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-30T18:53:38.383816Z","bundle_sha256":"a2d879f26cae55efaf642c7c63afdb85a669b6e8046aaeca16b563d4df312241"}}