{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:F6SJB5MRXTZOUKZNP2M2HWWKI4","short_pith_number":"pith:F6SJB5MR","schema_version":"1.0","canonical_sha256":"2fa490f591bcf2ea2b2d7e99a3daca473031888bd2f3df351d0be9e68684b4dd","source":{"kind":"arxiv","id":"2311.03099","version":3},"attestation_state":"computed","paper":{"title":"Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bowen Yu, Fei Huang, Haiyang Yu, Le Yu, Yongbin Li","submitted_at":"2023-11-06T13:43:07Z","abstract_excerpt":"In this paper, we unveil that Language Models (LMs) can acquire new capabilities by assimilating parameters from homologous models without retraining or GPUs. We first introduce DARE to set most delta parameters (i.e., the disparity between fine-tuned and pre-trained parameters) to zeros without affecting the abilities of Supervised Fine-Tuning (SFT) LMs, which randomly Drops delta parameters with a ratio $p$ And REscales the remaining ones by $1 / (1 - p)$ to approximate the original embeddings. Then, we use DARE as a versatile plug-in to sparsify delta parameters of multiple SFT homologous m"},"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":"2311.03099","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-11-06T13:43:07Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"c33a0898692b185d348fe348d800fa2a5c2a78ec6aec534ea138a06f225ffe5d","abstract_canon_sha256":"342713a513884c7f4c8834058a05df90854e930d0a23c9e0bbcaa95996d5471b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:31:06.657698Z","signature_b64":"lrwNu7oOshKk964bFCsbY9cXKpyuAKN7CBu7erzVfPkf3CRvfKerBd5gcpOZRG+QKyXUK/aSsCgrHMcysuhNBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2fa490f591bcf2ea2b2d7e99a3daca473031888bd2f3df351d0be9e68684b4dd","last_reissued_at":"2026-07-05T08:31:06.657216Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:31:06.657216Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bowen Yu, Fei Huang, Haiyang Yu, Le Yu, Yongbin Li","submitted_at":"2023-11-06T13:43:07Z","abstract_excerpt":"In this paper, we unveil that Language Models (LMs) can acquire new capabilities by assimilating parameters from homologous models without retraining or GPUs. We first introduce DARE to set most delta parameters (i.e., the disparity between fine-tuned and pre-trained parameters) to zeros without affecting the abilities of Supervised Fine-Tuning (SFT) LMs, which randomly Drops delta parameters with a ratio $p$ And REscales the remaining ones by $1 / (1 - p)$ to approximate the original embeddings. Then, we use DARE as a versatile plug-in to sparsify delta parameters of multiple SFT homologous m"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2311.03099","kind":"arxiv","version":3},"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/2311.03099/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":"2311.03099","created_at":"2026-07-05T08:31:06.657271+00:00"},{"alias_kind":"arxiv_version","alias_value":"2311.03099v3","created_at":"2026-07-05T08:31:06.657271+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2311.03099","created_at":"2026-07-05T08:31:06.657271+00:00"},{"alias_kind":"pith_short_12","alias_value":"F6SJB5MRXTZO","created_at":"2026-07-05T08:31:06.657271+00:00"},{"alias_kind":"pith_short_16","alias_value":"F6SJB5MRXTZOUKZN","created_at":"2026-07-05T08:31:06.657271+00:00"},{"alias_kind":"pith_short_8","alias_value":"F6SJB5MR","created_at":"2026-07-05T08:31:06.657271+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.02252","citing_title":"ResMerge: Residual-based Spectral Merging of Large Language Models","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28649","citing_title":"Interpretability-Guided Layer Selection over Subspace Projection: SAEs as Stethoscopes, Not Scalpels, for Raw Task Vector Model Editing","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29498","citing_title":"Mask the Target: A Plug-and-Play Regularizer Against LoRA Forgetting","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2603.06610","citing_title":"CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2411.02813","citing_title":"Sparse Orthogonal Parameters Tuning for Continual Learning","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2604.01674","citing_title":"Can Heterogeneous Language Models Be Fused?","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20296","citing_title":"Spectral Unforgetting: Post-Hoc Recovery of Damaged Capabilities Without Retraining","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12326","citing_title":"Black-Box Optimization of Mixed Binary-Continuous Variables: Challenges and Opportunities in Evolutionary Model Merging","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13694","citing_title":"Weight Patching: Toward Source-Level Mechanistic Localization in LLMs","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18473","citing_title":"Train Separately, Merge Together: Modular Post-Training with Mixture-of-Experts","ref_index":46,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/F6SJB5MRXTZOUKZNP2M2HWWKI4","json":"https://pith.science/pith/F6SJB5MRXTZOUKZNP2M2HWWKI4.json","graph_json":"https://pith.science/api/pith-number/F6SJB5MRXTZOUKZNP2M2HWWKI4/graph.json","events_json":"https://pith.science/api/pith-number/F6SJB5MRXTZOUKZNP2M2HWWKI4/events.json","paper":"https://pith.science/paper/F6SJB5MR"},"agent_actions":{"view_html":"https://pith.science/pith/F6SJB5MRXTZOUKZNP2M2HWWKI4","download_json":"https://pith.science/pith/F6SJB5MRXTZOUKZNP2M2HWWKI4.json","view_paper":"https://pith.science/paper/F6SJB5MR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2311.03099&json=true","fetch_graph":"https://pith.science/api/pith-number/F6SJB5MRXTZOUKZNP2M2HWWKI4/graph.json","fetch_events":"https://pith.science/api/pith-number/F6SJB5MRXTZOUKZNP2M2HWWKI4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/F6SJB5MRXTZOUKZNP2M2HWWKI4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/F6SJB5MRXTZOUKZNP2M2HWWKI4/action/storage_attestation","attest_author":"https://pith.science/pith/F6SJB5MRXTZOUKZNP2M2HWWKI4/action/author_attestation","sign_citation":"https://pith.science/pith/F6SJB5MRXTZOUKZNP2M2HWWKI4/action/citation_signature","submit_replication":"https://pith.science/pith/F6SJB5MRXTZOUKZNP2M2HWWKI4/action/replication_record"}},"created_at":"2026-07-05T08:31:06.657271+00:00","updated_at":"2026-07-05T08:31:06.657271+00:00"}