{"as_of":"2026-08-10T13:44:00Z","caps":{"database_statements":6,"inbound":100,"outbound":100},"context_digest":"sha256:42578e4348b7f58358a69a3aa6c2cbcba115af2956dffeaac893cb8ead3c9da0","coverage":[{"denominator":82,"lane":"reference_resolution","note":"Typed states for the displayed outbound observations.","records_observed":82,"source":"paper_references, paper_reference_links","source_observed_at":"2026-08-03T19:16:04.999697Z","state":"measured"},{"denominator":84,"lane":"standing_notices","note":"One-hop event checks from named stored sources.","records_observed":84,"source":"scholarly_work_events, retraction_status_cache","source_observed_at":"2026-08-10T06:31:04.303077+00:00","state":"measured"},{"denominator":2,"lane":"inbound_itemization","note":"Pith citing papers itemized under the disclosed page cap.","records_observed":2,"source":"paper_references, paper_reference_links","source_observed_at":"2026-05-14T21:12:06.989077Z","state":"measured"},{"denominator":1,"lane":"external_citation_measurements","note":"A source-named dated measurement, never combined with another source.","records_observed":0,"source":"arxiv_reference","source_observed_at":"2026-05-14T21:12:58.892326Z","state":"measured"}],"external_citation_measurements":[],"inbound":[{"citation":{"cited_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"cited_work":{"arxiv_id":"2512.01461","doi":null,"metadata_source":"arxiv_reference","pith_arxiv_id":"2512.01461","snapshot_observed_at":"2026-06-30T03:17:25.261573Z","title":"Stay unique, stay efficient: Preserving model personality in multi-task merging","venue":null,"work_id":"3de52971-58db-4e99-ae0c-065f7924aa63","year":2025},"citing_paper":{"arxiv_id":"2605.11679","last_updated":"2026-05-13T09:28:34Z","snapshot_observed_at":"2026-07-06T23:23:30.499023Z","submitted_at":"2026-05-12T07:38:59Z","title":"Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion","version":1},"reference_index":31,"source":"pdf_text","source_observed_at":"2026-05-13T01:03:10.263663Z"},"links":{"cited_paper":"/paper/2512.01461","citing_paper":"/paper/2605.11679"},"observation_digest":"sha256:5c89fb007e625b38e9df4fbfc2d6d6a287bbf7179447a43d91344e5228242167","observation_id":"e522c0d3-3187-4312-9b12-9b693b87bc25","resolution":{"observed_at":"2026-06-30T03:17:25.261573Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-10T06:31:04.303077+00:00","source":"crossref"},{"observed_at":"2026-08-10T06:30:57.382061+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"cited_work":{"arxiv_id":"2512.01461","doi":null,"metadata_source":"arxiv_reference","pith_arxiv_id":"2512.01461","snapshot_observed_at":"2026-06-30T03:17:25.261573Z","title":"Stay unique, stay efficient: Preserving model personality in multi-task merging","venue":null,"work_id":"3de52971-58db-4e99-ae0c-065f7924aa63","year":2025},"citing_paper":{"arxiv_id":"2605.11679","last_updated":"2026-05-13T09:28:34Z","snapshot_observed_at":"2026-07-06T23:23:30.499023Z","submitted_at":"2026-05-12T07:38:59Z","title":"Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion","version":2},"reference_index":31,"source":"pdf_text","source_observed_at":"2026-05-14T21:12:06.989077Z"},"links":{"cited_paper":"/paper/2512.01461","citing_paper":"/paper/2605.11679"},"observation_digest":"sha256:72222dcc8fd51daaa531cd447d1f9bb8bdc50255db49e61adb044f9655ad1a55","observation_id":"19ecc26c-6536-4bfb-9a4c-3a81b4f2e090","resolution":{"observed_at":"2026-06-30T03:17:25.261573Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-10T06:31:04.303077+00:00","source":"crossref"},{"observed_at":"2026-08-10T06:30:57.382061+00:00","source":"retraction_watch"}],"state":"measured"}}],"links":{"evidence":"/evidence","html":"/paper/2512.01461/citation-record","integrity":"/paper/2512.01461/integrity","json":"/paper/2512.01461/citation-record.json","paper":"/paper/2512.01461"},"outbound":[{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:50.783729Z","title":"Git re-basin: Merging models modulo permutation symmetries","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":1,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:50.783729Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:d10fa5be6cbbc3460d4f24f8568cc46da525030e17cf6c8debb8f915bb59f870","observation_id":"da399615-8fbb-4b24-b938-fe22c4bfd0ef","resolution":{"observed_at":"2026-08-03T19:15:50.783729Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2309.16609","last_updated":"2023-09-28T17:07:49Z","snapshot_observed_at":"2026-08-09T21:25:20.369782Z","submitted_at":"2023-09-28T17:07:49Z","title":"Qwen Technical Report","version":1},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2309.16609","snapshot_observed_at":"2026-08-03T19:15:50.873340Z","title":"Qwen technical report.arXiv preprint arXiv:2309.16609,","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":2,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:50.873340Z"},"links":{"cited_paper":"/paper/2309.16609","citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:c5a8f2cc05ec417f8a7ca568cf85d07c33f5d7c17c8c31f362065d89fca1f64b","observation_id":"b07ffe1c-0c09-4324-a646-bb87d8b19463","resolution":{"observed_at":"2026-08-03T19:15:50.873340Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:50.963658Z","title":"Multitask learning.Machine learning, 28(1): 41–75, 1997","venue":null,"work_id":null,"year":1997},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":3,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:50.963658Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:f74833c9368ba5c1153d14f6bfc64df6b917424b33e96f3dd70b7af36a7b69bb","observation_id":"28dd684c-b709-4a27-964d-05dac9aaed0c","resolution":{"observed_at":"2026-08-03T19:15:50.963658Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:51.095132Z","title":"Semeval-2017 task 1: Semantic textual similarity multilingual and crosslingual focused evaluation","venue":null,"work_id":null,"year":2017},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":4,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:51.095132Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:f81774964698df23f659871a475402b8bda8ba4fce32a0dac9e7ae2db6ca292f","observation_id":"e992de7c-3f6b-450a-8b90-b0b43539e11a","resolution":{"observed_at":"2026-08-03T19:15:51.095132Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:51.252755Z","title":"Fw-merging: Scaling model merging with frank-wolfe optimization","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":5,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:51.252755Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:481a42d7a590587dd6abac9a85f60cc2d6b3d4200726333b8b31ca69957781f9","observation_id":"de5ec714-ec6f-4e0f-bc99-b65a1b40f4fc","resolution":{"observed_at":"2026-08-03T19:15:51.252755Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:51.424928Z","title":"Quora question pairs, 2018","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":6,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:51.424928Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:9e522501bc6e81c5df13ef6392042cd4aa94e1843b730b5f64fadb39a1ca1057","observation_id":"336b8f46-82db-4710-9d67-3687a94639ef","resolution":{"observed_at":"2026-08-03T19:15:51.424928Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:51.567201Z","title":"Remote sensing image scene classification: Benchmark and state of the art","venue":null,"work_id":null,"year":2017},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":7,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:51.567201Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:c8f44163e69541f09a43ede93338451eb4bf5d7c3d57c886a6e9b456c6306777","observation_id":"12efaed5-dd10-4d1d-ae86-4be31b08a428","resolution":{"observed_at":"2026-08-03T19:15:51.567201Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:51.756694Z","title":"Describing textures in the wild","venue":null,"work_id":null,"year":2014},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":8,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:51.756694Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:813ce4f1c201c9777110e93382ee50b20b01b239a04d0fd366f0721a9ffdbf73","observation_id":"9fcd8eea-b3b3-4733-b9c9-78019ba3f713","resolution":{"observed_at":"2026-08-03T19:15:51.756694Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:51.928142Z","title":"Model breadcrumbs: scalable upcycling of finetuned foundation models via sparse task vectors merging","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":9,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:51.928142Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:adc54152e8b08fa32320900e0b56162cbcf8ee0c245bb093049347cd2105f436","observation_id":"b11c9aa3-2bd8-431e-b30a-b9e113337182","resolution":{"observed_at":"2026-08-03T19:15:51.928142Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:52.105687Z","title":"The mnist database of handwritten digit images for machine learning research.IEEE signal processing magazine, pages 141–142, 2012","venue":null,"work_id":null,"year":2012},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":10,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:52.105687Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:df0bff7aacb5808e85b381b1f95c9ca20a2c3e52d896c6261af833298c0b67a3","observation_id":"ddbf7887-15fd-4f7d-a1d2-ee1b2829481c","resolution":{"observed_at":"2026-08-03T19:15:52.105687Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2506.09446","last_updated":"2025-06-11T06:52:36Z","snapshot_observed_at":"2026-08-07T04:45:42.093041Z","submitted_at":"2025-06-11T06:52:36Z","title":"Harmonizing and Merging Source Models for CLIP-based Domain Generalization","version":1},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2506.09446","snapshot_observed_at":"2026-08-03T19:15:52.251953Z","title":"Harmonizing and merging source mod- els for clip-based domain generalization.arXiv preprint arXiv:2506.09446, 2025","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":11,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:52.251953Z"},"links":{"cited_paper":"/paper/2506.09446","citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:d81c9b3762a997812510062efdc71b8fbdecf18a1b9af158917a1380e303a66b","observation_id":"f6d345af-7335-4899-a85a-72d1b24c1874","resolution":{"observed_at":"2026-08-03T19:15:52.251953Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2002.06305","last_updated":"2020-02-15T02:40:10Z","snapshot_observed_at":"2026-08-07T15:45:43.952318Z","submitted_at":"2020-02-15T02:40:10Z","title":"Fine-Tuning Pretrained Language Models: Weight Initializations, Data Orders, and Early Stopping","version":1},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2002.06305","snapshot_observed_at":"2026-08-03T19:15:52.462406Z","title":"Fine-tuning pretrained language models: Weight initializations, data orders, and early stopping.arXiv:2002.06305, 2020","venue":null,"work_id":null,"year":2002},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":12,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:52.462406Z"},"links":{"cited_paper":"/paper/2002.06305","citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:a9f745c5e198d5f8f62dc0bac43ec68cc8c821ca9ee99597efca7f551e91cc95","observation_id":"21fea34f-8013-43e0-a42d-cb2833372c9d","resolution":{"observed_at":"2026-08-03T19:15:52.462406Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:52.605455Z","title":"Automatically constructing a corpus of sentential paraphrases","venue":null,"work_id":null,"year":2005},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":13,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:52.605455Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:57074a64f0cebf98773341f4422f373fd5b11041aa4e0295a0a8d89cffeb335f","observation_id":"e046c146-1076-4729-a560-dee117416d40","resolution":{"observed_at":"2026-08-03T19:15:52.605455Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:52.747149Z","title":"Task singular vectors: Reducing task interference in model merging","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":14,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:52.747149Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:cbd0c18728826aed66e8e1bae1b6ed5b76c23eb308f3e554b1621411c4473faa","observation_id":"7f29518c-dff4-4e75-94d7-1442fab13a4a","resolution":{"observed_at":"2026-08-03T19:15:52.747149Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:52.891962Z","title":"The third pascal recognizing textual entail- ment challenge","venue":null,"work_id":null,"year":2007},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":15,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:52.891962Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:1bf8fbf0a3c86f8e86ab3b208d7f0c2b254ba094f28955c55012f6b38b2af15f","observation_id":"61197950-567a-430a-b54f-b8d35421d587","resolution":{"observed_at":"2026-08-03T19:15:52.891962Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:53.067984Z","title":"Gradient reweighting: Towards imbalanced class-incremental learning","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":16,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:53.067984Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:b440eee74ef716c34a43ce055de88a9b4882005ee5b340aacb8286f93708d58b","observation_id":"a102c5cb-1c8b-4eda-a937-62edccec1e50","resolution":{"observed_at":"2026-08-03T19:15:53.067984Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:53.289663Z","title":null,"venue":null,"work_id":null,"year":2019},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":17,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:53.289663Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:b3905c8d392399458f98824b1b9796ededaf999d6ab87c814234bdf5dc318784","observation_id":"bf17e717-a55c-4d3b-8fa1-4f7368dc1168","resolution":{"observed_at":"2026-08-03T19:15:53.289663Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:53.468921Z","title":"Measuring massive multitask language understanding","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":18,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:53.468921Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:475f196c93499fc1825dbec67a9f63915298548c92c74c4ff094048a03a868d9","observation_id":"8f2e8975-5c84-4179-aad9-895c5c93b1a4","resolution":{"observed_at":"2026-08-03T19:15:53.468921Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:53.682709Z","title":"Emr-merging: Tuning-free high- performance model merging","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":19,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:53.682709Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:1fe67bd4183ef60b93baf361b350582e0f93b973154846c9a306efdf92118baa","observation_id":"b357f422-841c-42f5-8dd6-50e3593779be","resolution":{"observed_at":"2026-08-03T19:15:53.682709Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:53.885946Z","title":"Multi-granular spatio-temporal token merging for training-free acceleration of video llms","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":20,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:53.885946Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:02b0dbbbb0c03cd82802231e4b5035022852a85b5761d6955dce9029b7c87c14","observation_id":"d3b1d99c-8781-4863-8a2e-0550319ebfef","resolution":{"observed_at":"2026-08-03T19:15:53.885946Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:54.107569Z","title":"Editing models with task arithmetic","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":21,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:54.107569Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:2bf68e11c57f42c6cc0ee424eccc96cd963d0a4b538f60c4d0ece63249efcf84","observation_id":"c362a8e0-5616-4ef7-9792-5103f5a6c4bc","resolution":{"observed_at":"2026-08-03T19:15:54.107569Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:54.303434Z","title":"Pytorch.Programming with TensorFlow: so- lution for edge computing applications, pages 87–104, 2021","venue":null,"work_id":null,"year":2021},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":22,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:54.303434Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:2fa6225d4b14928360e5e0b0572975f1f3a170ff0ef82e7331127482eca54f5e","observation_id":"d0af8a09-7bed-4837-922c-46280612bd72","resolution":{"observed_at":"2026-08-03T19:15:54.303434Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:54.493367Z","title":"Dataless knowledge fusion by merging weights of language models","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":23,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:54.493367Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:4f95d01993b43fff5d7ee918128839ea5443f8c669d5d47ce9b47127cbd4fe7f","observation_id":"bd2ce9cb-5cdf-4b88-9cf2-7eb6549eac0c","resolution":{"observed_at":"2026-08-03T19:15:54.493367Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:54.780062Z","title":"Repair: Renormalizing permuted activations for interpolation repair","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":24,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:54.780062Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:04b81352d86c90fa5d8a5ee998d2d6185081004d7006d6a09dec0bcff6e82c98","observation_id":"9a456770-13d9-4b6f-bece-974294769e1e","resolution":{"observed_at":"2026-08-03T19:15:54.780062Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:54.938422Z","title":"Task vector quantization for memory- efficient model merging","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":25,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:54.938422Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:c043f2c8ffbdca88353b6a1cb6618758e830e61c30d59508d0bf6203cfb9c6b9","observation_id":"df19c80f-66dc-40ad-abb3-a13669cfc5ee","resolution":{"observed_at":"2026-08-03T19:15:54.938422Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:55.125850Z","title":"3d object representations for fine-grained categorization","venue":null,"work_id":null,"year":2013},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":26,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:55.125850Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:5d1776068451b16e86e0b8726a2e2ad536bddac5b53f6924243387cc15e1faad","observation_id":"f3544bb3-a802-40fa-bb4c-c797cdc0f16e","resolution":{"observed_at":"2026-08-03T19:15:55.125850Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:55.215038Z","title":"Singular value decompo- sition.Numerical analysis for statisticians, pages 129–142,","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":27,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:55.215038Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:3fb1a481ceee6c45c654a3a3cdb6b7a1e411865e63826b64220c2b791821a57d","observation_id":"7052d903-a01a-474d-b5a8-9058ffcca397","resolution":{"observed_at":"2026-08-03T19:15:55.215038Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:55.302483Z","title":"Mitigating parameter interference in model merging via sharpness-aware fine-tuning","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":28,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:55.302483Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:881f1b244e3176941c2a0934b134b1577ad446d153fda147055564cda787e812","observation_id":"05e070d1-56fd-4993-a159-eabed10295ab","resolution":{"observed_at":"2026-08-03T19:15:55.302483Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:55.383126Z","title":"Map: Low-compute model merging with amortized pareto fronts via quadratic approximation","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":29,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:55.383126Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:38698faccbb6fa1c0431dfbf30baf63758ca0687adab7ac0a80f16d09561254b","observation_id":"034a1d56-e3f8-4bc8-bbcf-13b1e5d92356","resolution":{"observed_at":"2026-08-03T19:15:55.383126Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:55.407891Z","title":"Model merging in pre-training of large language models","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":30,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:55.407891Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:f8a1e323a5ba99aa39b64ee6edd30c1a3ec36627cb139edd66aac3c974330f42","observation_id":"52a0132b-2b31-4174-840b-1bfec81a6453","resolution":{"observed_at":"2026-08-03T19:15:55.407891Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:55.473559Z","title":"Truthfulqa: Measuring how models mimic human falsehoods","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":31,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:55.473559Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:7fe4189d1c286201c43b03c54e91b61efdb2d719ef76a2c051d1eddcabe341aa","observation_id":"1e4bced3-f827-4dee-9c27-6dcdab66b936","resolution":{"observed_at":"2026-08-03T19:15:55.473559Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2502.10743","last_updated":"2025-05-23T02:52:05Z","snapshot_observed_at":"2026-08-07T18:16:03.824830Z","submitted_at":"2025-02-15T09:47:50Z","title":"1bit-Merging: Dynamic Quantized Merging for Large Language Models","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2502.10743","snapshot_observed_at":"2026-08-03T19:15:55.592279Z","title":"1bit-merging: Dynamic quantized merging for large language models","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":32,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:55.592279Z"},"links":{"cited_paper":"/paper/2502.10743","citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:fc1b37d399e693f5f02ec590c3976481a0a882c9db6c6e993c2daac6dec42925","observation_id":"92ec3d1a-a429-409a-8b34-d11440dd8b04","resolution":{"observed_at":"2026-08-03T19:15:55.592279Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"1907.11692","last_updated":"2019-07-26T17:48:29Z","snapshot_observed_at":"2026-07-31T22:31:37.910868Z","submitted_at":"2019-07-26T17:48:29Z","title":"RoBERTa: A Robustly Optimized BERT Pretraining Approach","version":1},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"1907.11692","snapshot_observed_at":"2026-08-03T19:15:55.722729Z","title":"Roberta: A robustly optimized bert pretraining approach.arXiv:1907.11692, 2019","venue":null,"work_id":null,"year":1907},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":33,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:55.722729Z"},"links":{"cited_paper":"/paper/1907.11692","citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:0113c066225f5f130de7f0fcae941b6e0b574d152ff4de8a369ac93d52509434","observation_id":"9c8dc007-c3d1-458d-bd2c-f43bb78300bf","resolution":{"observed_at":"2026-08-03T19:15:55.722729Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:55.843459Z","title":"Twin-merging: Dynamic integration of modular expertise in model merging","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":34,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:55.843459Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:c55ebdd0df77d2844ec3e71ef998fd8c0c5cecea8418d27b705da7f1aab47495","observation_id":"d5af9a1b-a5b7-4cf3-8ce2-718930b74d08","resolution":{"observed_at":"2026-08-03T19:15:55.843459Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:55.993993Z","title":"No task left behind: Isotropic model merging with common and task-specific subspaces","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":35,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:55.993993Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:f464aace45271a86cc1508aed9e4550f73730e3f02a3b1854e604086c2db1bf9","observation_id":"a11c0d9d-1620-497e-bd79-55f369567d98","resolution":{"observed_at":"2026-08-03T19:15:55.993993Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:56.131288Z","title":"Merging models with fisher-weighted averaging","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":36,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:56.131288Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:223d0f5eee2a95c7e7dccbd7f302b5f8d016db87d3e11361caf8281c484ccabc","observation_id":"498a2ceb-3d32-492d-ae8f-88f4761b69d2","resolution":{"observed_at":"2026-08-03T19:15:56.131288Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2306.03745","last_updated":"2024-05-13T16:20:29Z","snapshot_observed_at":"2026-07-06T15:39:14.565475Z","submitted_at":"2023-06-06T15:04:31Z","title":"Soft Merging of Experts with Adaptive Routing","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2306.03745","snapshot_observed_at":"2026-08-03T19:15:56.279870Z","title":"Soft merging of experts with adaptive routing.arXiv:2306.03745,","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":37,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:56.279870Z"},"links":{"cited_paper":"/paper/2306.03745","citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:2254298e5032ec88844615c143cf22d21992437058dfff8d60bd23f8c356a1ae","observation_id":"c03961da-b987-40f1-bdb4-91d93cbc3706","resolution":{"observed_at":"2026-08-03T19:15:56.279870Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:56.365931Z","title":"Reading digits in natural images with unsupervised feature learning","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":38,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:56.365931Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:d071f716d40bad11c0a327272a2abd0ba4e7e962b5c6f2338b31fa96c8384af7","observation_id":"d8879641-fd29-461b-b695-df044e42c017","resolution":{"observed_at":"2026-08-03T19:15:56.365931Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:56.476943Z","title":"Dawin: Training-free dynamic weight interpolation for robust adaptation","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":39,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:56.476943Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:28ea734ea2df79742ab2950b49add9a4f4e5956dac62eb59fee5e69c036a9638","observation_id":"729470e4-0a3d-43f8-b2ef-70fcd5d70d4a","resolution":{"observed_at":"2026-08-03T19:15:56.476943Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:56.604743Z","title":"Accurate and efficient low-rank model merging in core space","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":40,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:56.604743Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:8998d0ed2a57acaa5c9283c3737bc22f702e22d2ab7329b16ea167283e1e6735","observation_id":"a3a5bdce-b087-418c-b5f2-92d9a423c79b","resolution":{"observed_at":"2026-08-03T19:15:56.604743Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:56.685292Z","title":"Bbq: A hand-built bias benchmark for question answering","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":41,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:56.685292Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:683845644f9d759f300fea151d9620e92dc1e280dce456ddd0a0ed7555e40d93","observation_id":"3476cbb8-6728-4f84-8c35-0be6f294bf3c","resolution":{"observed_at":"2026-08-03T19:15:56.685292Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:56.744947Z","title":"Less is more: Efficient model merging with binary task switch","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":42,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:56.744947Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:d9a38877f8279e47fd9e3bb846e5c34e4cd89397742f2f2312156c6f12f1d17e","observation_id":"bb995de4-9539-408f-bfdf-8733495f60c9","resolution":{"observed_at":"2026-08-03T19:15:56.744947Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:56.862391Z","title":"Mingle: Mixtures of null- space gated low-rank experts for test-time continual model merging","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":43,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:56.862391Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:c357ccb8e5d12b3051a4f36266b7067947ac9359f485b583e394a7a4df5727d5","observation_id":"c0fbedc3-a5b0-42a7-99ff-81e9cbcbe405","resolution":{"observed_at":"2026-08-03T19:15:56.862391Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:56.948299Z","title":"Language models are unsuper- vised multitask learners.OpenAI blog, page 9, 2019","venue":null,"work_id":null,"year":2019},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":44,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:56.948299Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:889b6f9f83ff0bf4357a71463f5edcf59a2f7a5d51668fd603ca2c734f8aead3","observation_id":"ff65d78c-2bee-4ad8-889f-6c85b7262052","resolution":{"observed_at":"2026-08-03T19:15:56.948299Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:57.071460Z","title":"Learning transferable visual models from natural language supervision","venue":null,"work_id":null,"year":2021},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":45,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:57.071460Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:da3d8f1f4b2eb8acf8ead18f348baa1477c6b7b145371d445b20f9e27456c96c","observation_id":"cae3866d-f1c6-4bc6-b408-f0faaadf5531","resolution":{"observed_at":"2026-08-03T19:15:57.071460Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:57.194051Z","title":"Squad: 100,000+ questions for machine comprehen- sion of text","venue":null,"work_id":null,"year":2016},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":46,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:57.194051Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:f961990e2ffe1b5511c29b50337b87b613d9de4877418eb2792d1a25f3901ba4","observation_id":"1239da18-ec36-4d92-b49a-137665d5485f","resolution":{"observed_at":"2026-08-03T19:15:57.194051Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:15:57.670437Z","title":"Recursive deep models for semantic compositionality over a sentiment treebank","venue":null,"work_id":null,"year":2013},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":47,"source":"pdf_text","source_observed_at":"2026-08-03T19:15:57.670437Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:34bacf82716a1df5156d27b1c473ac006f94352babb81eb037fb5907b0d556f8","observation_id":"4ede5230-228f-4063-8972-68ede2ebd725","resolution":{"observed_at":"2026-08-03T19:15:57.670437Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:00.424362Z","title":"The german traffic sign recognition benchmark: a multi-class classification competition","venue":null,"work_id":null,"year":2011},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":48,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:00.424362Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:2a4eb1c2f11acb97c4ac4cae3960c62f9b646627a02e7c176e159b713745d8c8","observation_id":"66839cdb-60ed-45e8-a490-1e2813a7af8c","resolution":{"observed_at":"2026-08-03T19:16:00.424362Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:00.610131Z","title":"Model merging with svd to tie the knots","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":49,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:00.610131Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:d13ab842b04e93f57f34f6f588297e2e4a29257452244e1766c5d2af63db1edb","observation_id":"6c41d9eb-5b8d-470f-a2a8-86f12c8bc214","resolution":{"observed_at":"2026-08-03T19:16:00.610131Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:00.740601Z","title":"Cat merging: A training-free approach for resolving conflicts in model merging","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":50,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:00.740601Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:f7d9a3dd328feea8942240a3692cd64bccc3b9387279573a7422b90393678052","observation_id":"b0d80eb1-baa0-43da-865e-fd4491aae670","resolution":{"observed_at":"2026-08-03T19:16:00.740601Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:00.934327Z","title":"Towards minimizing feature drift in model merging: Layer-wise task vector fusion for adaptive knowledge integration.arXiv preprint arXiv:2505.23859,","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":51,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:00.934327Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:fc18759d3a746b24326109258430acf2e3fbf78d736e3d922d3d849fcab99996","observation_id":"942444c8-e662-4279-88a3-4195dc3c13f3","resolution":{"observed_at":"2026-08-03T19:16:00.934327Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:01.093238Z","title":"Merging multi-task models via weight- ensembling mixture of experts","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":52,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:01.093238Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:a24a2849e42ccf2b5aa707a7996e94c5f0e1b75f0296e90eb21a3d7340215f1a","observation_id":"c71481e9-568d-47cb-991c-56ef88ffa65a","resolution":{"observed_at":"2026-08-03T19:16:01.093238Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2307.09288","last_updated":"2023-07-19T17:08:59Z","snapshot_observed_at":"2026-08-07T12:56:43.323460Z","submitted_at":"2023-07-18T14:31:57Z","title":"Llama 2: Open Foundation and Fine-Tuned Chat Models","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2307.09288","snapshot_observed_at":"2026-08-03T19:16:01.278500Z","title":"Llama 2: Open foundation and fine-tuned chat models","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":53,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:01.278500Z"},"links":{"cited_paper":"/paper/2307.09288","citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:ad12d9283cf8c6cf22089a5a41ba309e7d65e83d45b0428e8ece6a776c9b2cf2","observation_id":"34a8f182-8956-474d-b488-b29f84db6d57","resolution":{"observed_at":"2026-08-03T19:16:01.278500Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:01.486664Z","title":"Glue: A multi-task benchmark and analysis platform for natural language under- standing","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":54,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:01.486664Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:6b7549c943a6835a4cbe64d7d0fac42389e72b3d38acfec0acf4dec5e43fe830","observation_id":"0cd60987-ede8-4b61-ab43-71b38f175ec0","resolution":{"observed_at":"2026-08-03T19:16:01.486664Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:01.696081Z","title":"Localizing task infor- mation for improved model merging and compression","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":55,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:01.696081Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:ec78653fb7a79d2724378e972ae501cc23d01fc7129b8ef81897e3c7e8e2bd93","observation_id":"8150e4dd-4c6e-4f5f-869b-0df38ca90866","resolution":{"observed_at":"2026-08-03T19:16:01.696081Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:01.794790Z","title":"Neural network acceptability judgments","venue":null,"work_id":null,"year":2019},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":56,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:01.794790Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:42f56f9c5ca586c000c95a2dc02a5ab6fc795d8a02a77be0f3a4a433bd1ce65c","observation_id":"51ee11c9-64d8-49d0-896a-aff217c76fd7","resolution":{"observed_at":"2026-08-03T19:16:01.794790Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:01.935010Z","title":"A broad-coverage challenge corpus for sentence understanding through inference","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":57,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:01.935010Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:ef624966ddac4b5c77a03501034ee234cde8cc4259dcd4b1a876b921a4fd1de7","observation_id":"c30a9d63-26a5-46d9-8e7b-75a06c9735f5","resolution":{"observed_at":"2026-08-03T19:16:01.935010Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"1910.03771","last_updated":"2020-07-14T03:42:34Z","snapshot_observed_at":"2026-07-06T08:27:58.343233Z","submitted_at":"2019-10-09T03:23:22Z","title":"HuggingFace's Transformers: State-of-the-art Natural Language Processing","version":5},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"1910.03771","snapshot_observed_at":"2026-08-03T19:16:02.135379Z","title":"Huggingface’s transformers: State-of-the-art natural language processing","venue":null,"work_id":null,"year":1910},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":58,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:02.135379Z"},"links":{"cited_paper":"/paper/1910.03771","citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:aa5366f247a0a15bf89e24702bbe17b135cb8d349f0867a30f3989460af566ee","observation_id":"174a7012-9e9d-42ff-bbec-9149ee20cd49","resolution":{"observed_at":"2026-08-03T19:16:02.135379Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:02.337074Z","title":"Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing in- ference time","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":59,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:02.337074Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:0e35cf821cd5b055351ca71a2c2bccdbf036583b3f4f52c2d1b3876831f50328","observation_id":"f240e85c-23a1-46eb-a660-daebd06cbf52","resolution":{"observed_at":"2026-08-03T19:16:02.337074Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:02.523080Z","title":"Importance-based token merging for efficient image and video generation","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":60,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:02.523080Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:9088dd16df89dfcba3f46dc4cb2067f4eec9a06e523d9d7301a9e44585c6781a","observation_id":"c2d1c4e4-f8f6-4354-a062-694e60ca61ea","resolution":{"observed_at":"2026-08-03T19:16:02.523080Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:02.757883Z","title":"Sun database: Large-scale scene recog- nition from abbey to zoo","venue":null,"work_id":null,"year":2010},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":61,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:02.757883Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:009014d30c275e00207402386a2a8cb2865378eeb61a757d39718ee27c12d3ba","observation_id":"213a6f0b-0eb5-4db6-80a5-c72eca9ced60","resolution":{"observed_at":"2026-08-03T19:16:02.757883Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:02.851393Z","title":"Ties-merging: Resolving interference when merging models","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":62,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:02.851393Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:53f6fcd896023d1e356f04879c966ece7285fc063b69d1f0a0ebb7a6ebfb1c68","observation_id":"391f56a8-5815-4c73-a63d-cd86b227854e","resolution":{"observed_at":"2026-08-03T19:16:02.851393Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.006217Z","title":"Calm: Consensus-aware localized merging for multi-task learning","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":63,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.006217Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:5e682af6a2309b9570f61a7a5947bc58885ce259cf975ba24af70629f4a20b21","observation_id":"2b213b1d-01f6-4ef9-968e-152d905452f2","resolution":{"observed_at":"2026-08-03T19:16:03.006217Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.094983Z","title":"Adamerging: Adap- tive model merging for multi-task learning","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":64,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.094983Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:0ff60700d802c298e9530b7efb76d10b334110204b2fc31b51078ea8a665f517","observation_id":"3622a0a5-e684-444d-b872-5a7c0238e7ce","resolution":{"observed_at":"2026-08-03T19:16:03.094983Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.188200Z","title":"Continual model merg- ing without data: Dual projections for balancing stability and plasticity","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":65,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.188200Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:bb53377a58e00bc8bfd598b26674d01c8b0bc8819ccfbea642c2d0405e403688","observation_id":"67bf4f89-086c-440e-81db-4a6f8af8f7f0","resolution":{"observed_at":"2026-08-03T19:16:03.188200Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.244174Z","title":"Mix data or merge models? balancing the helpfulness, honesty, and harmlessness of large language model via model merging","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":66,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.244174Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:27b29ffe266128f7fea3d6bb80249a3dfbdbc279e04a523b0679e4ca1ebd5bf6","observation_id":"4396b531-c3f7-4eeb-8c30-25a36dfc386c","resolution":{"observed_at":"2026-08-03T19:16:03.244174Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2312.16240","last_updated":"2023-12-25T09:32:28Z","snapshot_observed_at":"2026-07-06T17:08:17.073060Z","submitted_at":"2023-12-25T09:32:28Z","title":"Merging Vision Transformers from Different Tasks and Domains","version":1},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2312.16240","snapshot_observed_at":"2026-08-03T19:16:03.288029Z","title":"Merg- ing vision transformers from different tasks and domains","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":67,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.288029Z"},"links":{"cited_paper":"/paper/2312.16240","citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:473c2ac85ed5631fb7c85cc4ec16811728b066375957e3b266c74d2694029159","observation_id":"be94f38f-ed2a-4a70-b9de-248fdc155acf","resolution":{"observed_at":"2026-08-03T19:16:03.288029Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.380629Z","title":"Language models are super mario: Absorbing abilities from homologous models as a free lunch","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":68,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.380629Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:ab617ca7e39026d7fd1ee8f9d3fe9aa375865af7696c81eaddc407740ed664a4","observation_id":"348c52c8-9e30-4b41-8a50-24c8f186c4d4","resolution":{"observed_at":"2026-08-03T19:16:03.380629Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.483210Z","title":"Robustmerge: Parameter-efficient model merging for mllms with direction robustness","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":69,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.483210Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:c5619817dee3a8d6fe6543f681308e6fcf431ca29387fe7b9ad9aaaa9fab5eed","observation_id":"7977c72d-a9ab-4e34-bd37-56107e8bc9b8","resolution":{"observed_at":"2026-08-03T19:16:03.483210Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.635239Z","title":"An overview of multi-task learn- ing.National Science Review, pages 30–43, 2018","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":70,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.635239Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:c8cc43846c4b4c1f17d82a1102c7212df49249a9fd375aa229a9807bca2a2597","observation_id":"9e70b426-2ff4-490b-99b5-ee44a0eeda04","resolution":{"observed_at":"2026-08-03T19:16:03.635239Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.727047Z","title":"A survey on multi-task learning","venue":null,"work_id":null,"year":2021},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":71,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.727047Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:fb6285e0c569edfca0dac81a3503c3cfe2e59772902dce3ee41f18a1ca201726","observation_id":"1b00b4b1-9c16-4e57-bb62-7c766c93ea30","resolution":{"observed_at":"2026-08-03T19:16:03.727047Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.810134Z","title":"Beyond training: Dynamic token merging for zero-shot video understanding","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":72,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.810134Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:13f9b82481b0807e40907074f2444c97ca6d61e2822e65796e4f8ec93c7593cd","observation_id":"e2ac7eb7-36e6-482f-9c40-d6155dbbcff3","resolution":{"observed_at":"2026-08-03T19:16:03.810134Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.870473Z","title":"Merging loras like playing lego: Pushing the modularity of lora to extremes through rank-wise clustering","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":73,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.870473Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:9f5b1d87ef18c9e06d419f4a79af5d425a1dc846f37cccf097b783c5748dac00","observation_id":"b2795d49-68cf-499e-a7f1-89948d08566e","resolution":{"observed_at":"2026-08-03T19:16:03.870473Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:03.988962Z","title":"Free-merging: Fourier transform for efficient model merging","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":74,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:03.988962Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:39d8f8b95eefb10522396b31f787191cd910ee2ebb3fc97880ab8fedab957d26","observation_id":"37368d30-7f3f-467c-b41c-a382c461aa48","resolution":{"observed_at":"2026-08-03T19:16:03.988962Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:04.061770Z","title":"Aim: Adaptive inference of multi-modal llms via token merging and pruning","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":75,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:04.061770Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:e8de785f13769148b2ab15f48519077b661c180a3529193634c2a4dfa0467938","observation_id":"5487cd89-048c-4cf7-b44c-597ac5832bb2","resolution":{"observed_at":"2026-08-03T19:16:04.061770Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:04.245221Z","title":"Metagpt: Merging large language models using model exclusive task arithmetic","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":76,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:04.245221Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:f45979be961f1d76e9d32e9d1c332a9af7a3ae8b6f3a00810333ef6506fd012d","observation_id":"2556686e-7b97-4b14-8979-60530e573964","resolution":{"observed_at":"2026-08-03T19:16:04.245221Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:04.347407Z","title":"Hm3: Hierarchical multi-objective model merging for pretrained models","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":77,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:04.347407Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:45ec7d4ef9c01c833cc3558652c611285ae244cfe6d6ff525809a6f6e98e4c3d","observation_id":"b8094bb2-fba1-41df-a3e1-caa9fa85152a","resolution":{"observed_at":"2026-08-03T19:16:04.347407Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:04.534741Z","title":"Remedy: Recipe merging dynam- ics in large vision-language models","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":78,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:04.534741Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:e674857b85de0cde1fa74052feb6c18b85534e24eea93761f3bf10638f601b2c","observation_id":"7d5aa789-7c83-4f32-9390-a487750bfb51","resolution":{"observed_at":"2026-08-03T19:16:04.534741Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:04.715412Z","title":null,"venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":79,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:04.715412Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:71b8d06e11e45ad61c202cbba7472056017cea6fb17de5ba4d3ed45f57182945","observation_id":"b6b1bc46-69fc-4454-a575-24f6745b782f","resolution":{"observed_at":"2026-08-03T19:16:04.715412Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:04.812594Z","title":"Baselines for seen tasks • Individual Modelsrefer to task-specific models before merging","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":80,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:04.812594Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:7e9efb8bd6fece8a1777ff973baa974eb1b63ac7b1874df5dfa2400c9f3c672c","observation_id":"cbd15223-86e5-4f2b-89b9-df7d0cda9260","resolution":{"observed_at":"2026-08-03T19:16:04.812594Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:04.883595Z","title":"Datasets For visual classification tasks, we employ classification accu- racy as the evaluation metric","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":81,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:04.883595Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:1591fb44c98e0cb8057906e7966ef255354bc8756f0e19ef2494e11cfe5cf337","observation_id":"3e61ff90-deff-4d4c-ae97-874850397e29","resolution":{"observed_at":"2026-08-03T19:16:04.883595Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-03T19:16:04.999697Z","title":"More Backbones In addition to the backbones evaluated in the main paper, we also assess the performance of various methods on ViT-B/16, ViT-L/14, and GPT-2 backbones","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging","version":2},"reference_index":82,"source":"pdf_text","source_observed_at":"2026-08-03T19:16:04.999697Z"},"links":{"citing_paper":"/paper/2512.01461"},"observation_digest":"sha256:ef2c75039628b115d3f0b9fdeb0f5b113d0ec3552e9a7d0b61fef08f902e561c","observation_id":"d3158880-0e22-4ba3-b934-41709a028a37","resolution":{"observed_at":"2026-08-03T19:16:04.999697Z","resolver_source":null,"status":"malformed_identifier"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}}],"paper":{"arxiv_id":"2512.01461","last_updated":"2026-06-29T13:53:53Z","latest_version":2,"primary_category":"cs.LG","snapshot_observed_at":"2026-08-10T09:51:10.747278Z","submitted_at":"2025-12-01T09:47:17Z","title":"Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging"},"reference_resolution":{"displayed":82,"state_counts":{"malformed_identifier":1,"metadata_mismatch":0,"parse_uncertain":0,"unresolved":81,"verified_exact":0,"verified_fuzzy":0},"total_outbound_references":82},"refusal":"A citation records a reference. It does not transfer a finding from one paper to another.","schema":"pith.paper-citation-record.v1","standing_sources":[{"observed_at":"2026-08-10T06:31:04.303077+00:00","source":"crossref"},{"observed_at":"2026-08-10T06:30:57.382061+00:00","source":"retraction_watch"}],"thesis":"As of 10 August 2026, this Paper Citation Record lists 82 of 82 outbound references and 2 inbound Pith citation observations for arXiv:2512.01461."}