{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:54JYBCNVDPVK44V37HLIOOBIGQ","short_pith_number":"pith:54JYBCNV","schema_version":"1.0","canonical_sha256":"ef138089b51beaae72bbf9d68738283401e004dd2218e82eab27fffcec199e43","source":{"kind":"arxiv","id":"2506.12379","version":1},"attestation_state":"computed","paper":{"title":"Training-free LLM Merging for Multi-task Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Hongzhi Yin, Shanshan Ye, Wanyu Wang, Xiangyu Zhao, Xian Wu, Yefeng Zheng, Yejing Wang, Yi Chang, Zichuan Fu","submitted_at":"2025-06-14T07:21:11Z","abstract_excerpt":"Large Language Models (LLMs) have demonstrated exceptional capabilities across diverse natural language processing (NLP) tasks. The release of open-source LLMs like LLaMA and Qwen has triggered the development of numerous fine-tuned models tailored for various tasks and languages. In this paper, we explore an important question: is it possible to combine these specialized models to create a unified model with multi-task capabilities. We introduces Hierarchical Iterative Merging (Hi-Merging), a training-free method for unifying different specialized LLMs into a single model. Specifically, Hi-Me"},"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":"2506.12379","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-14T07:21:11Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"b61f5bfb90bea2ab0996e663e95252f92ee9052b76e12e9155138a841a8f3512","abstract_canon_sha256":"9629d2cd4ad8b06eee598d39ba3c860849fb021317a3c327deb65941b1a963c7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:21:52.359028Z","signature_b64":"syhsMVBbmINyilfl6cZ+7TgXTkRf5KS+6BoTtnQBxM64j6p7FBoXRYmYcJbRmLMZtFJR9c8rvNIlacrTy70qDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ef138089b51beaae72bbf9d68738283401e004dd2218e82eab27fffcec199e43","last_reissued_at":"2026-07-05T11:21:52.358564Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:21:52.358564Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Training-free LLM Merging for Multi-task Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Hongzhi Yin, Shanshan Ye, Wanyu Wang, Xiangyu Zhao, Xian Wu, Yefeng Zheng, Yejing Wang, Yi Chang, Zichuan Fu","submitted_at":"2025-06-14T07:21:11Z","abstract_excerpt":"Large Language Models (LLMs) have demonstrated exceptional capabilities across diverse natural language processing (NLP) tasks. The release of open-source LLMs like LLaMA and Qwen has triggered the development of numerous fine-tuned models tailored for various tasks and languages. In this paper, we explore an important question: is it possible to combine these specialized models to create a unified model with multi-task capabilities. We introduces Hierarchical Iterative Merging (Hi-Merging), a training-free method for unifying different specialized LLMs into a single model. Specifically, Hi-Me"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.12379","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/2506.12379/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":"2506.12379","created_at":"2026-07-05T11:21:52.358626+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.12379v1","created_at":"2026-07-05T11:21:52.358626+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.12379","created_at":"2026-07-05T11:21:52.358626+00:00"},{"alias_kind":"pith_short_12","alias_value":"54JYBCNVDPVK","created_at":"2026-07-05T11:21:52.358626+00:00"},{"alias_kind":"pith_short_16","alias_value":"54JYBCNVDPVK44V3","created_at":"2026-07-05T11:21:52.358626+00:00"},{"alias_kind":"pith_short_8","alias_value":"54JYBCNV","created_at":"2026-07-05T11:21:52.358626+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.19409","citing_title":"Unlocking the Potential of Continual Model Merging: An ODE Perspective","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19409","citing_title":"Unlocking the Potential of Continual Model Merging: An ODE Perspective","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19409","citing_title":"Unlocking the Potential of Continual Model Merging: An ODE Perspective","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/54JYBCNVDPVK44V37HLIOOBIGQ","json":"https://pith.science/pith/54JYBCNVDPVK44V37HLIOOBIGQ.json","graph_json":"https://pith.science/api/pith-number/54JYBCNVDPVK44V37HLIOOBIGQ/graph.json","events_json":"https://pith.science/api/pith-number/54JYBCNVDPVK44V37HLIOOBIGQ/events.json","paper":"https://pith.science/paper/54JYBCNV"},"agent_actions":{"view_html":"https://pith.science/pith/54JYBCNVDPVK44V37HLIOOBIGQ","download_json":"https://pith.science/pith/54JYBCNVDPVK44V37HLIOOBIGQ.json","view_paper":"https://pith.science/paper/54JYBCNV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.12379&json=true","fetch_graph":"https://pith.science/api/pith-number/54JYBCNVDPVK44V37HLIOOBIGQ/graph.json","fetch_events":"https://pith.science/api/pith-number/54JYBCNVDPVK44V37HLIOOBIGQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/54JYBCNVDPVK44V37HLIOOBIGQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/54JYBCNVDPVK44V37HLIOOBIGQ/action/storage_attestation","attest_author":"https://pith.science/pith/54JYBCNVDPVK44V37HLIOOBIGQ/action/author_attestation","sign_citation":"https://pith.science/pith/54JYBCNVDPVK44V37HLIOOBIGQ/action/citation_signature","submit_replication":"https://pith.science/pith/54JYBCNVDPVK44V37HLIOOBIGQ/action/replication_record"}},"created_at":"2026-07-05T11:21:52.358626+00:00","updated_at":"2026-07-05T11:21:52.358626+00:00"}