{"as_of":"2026-08-09T11:39:00Z","caps":{"database_statements":6,"inbound":100,"outbound":100},"context_digest":"sha256:ac391b6ab3175cbc38f9d2dda16c47970d4ed3529b191543b61041f3ac5570c6","coverage":[{"denominator":72,"lane":"reference_resolution","note":"Typed states for the displayed outbound observations.","records_observed":72,"source":"paper_references, paper_reference_links","source_observed_at":"2026-08-08T17:11:02.195801Z","state":"measured"},{"denominator":72,"lane":"standing_notices","note":"One-hop event checks from named stored sources.","records_observed":72,"source":"scholarly_work_events, retraction_status_cache","source_observed_at":"2026-08-09T06:31:02.800959+00:00","state":"measured"},{"denominator":0,"lane":"inbound_itemization","note":"Pith citing papers itemized under the disclosed page cap.","records_observed":0,"source":"paper_references, paper_reference_links","source_observed_at":null,"state":"measured"},{"denominator":1,"lane":"external_citation_measurements","note":"A source-named dated measurement, never combined with another source.","records_observed":0,"source":"cited_works","source_observed_at":null,"state":"measured"}],"external_citation_measurements":[],"inbound":[],"links":{"evidence":"/evidence","html":"/paper/2608.05250/citation-record","integrity":"/paper/2608.05250/integrity","json":"/paper/2608.05250/citation-record.json","paper":"/paper/2608.05250"},"outbound":[{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:01.917747Z","title":"Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":1,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.917747Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:663174565178f60fb1944d66853bc3b448a4e6c8d6abb909cd3e312eab7886df","observation_id":"7fa26ea9-22d6-410b-9502-ee123d636ba5","resolution":{"observed_at":"2026-08-08T17:11:01.917747Z","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":"2026-08-08T17:11:01.924506Z","title":"Classification Problem Solving","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":2,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.924506Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:d7800f0252e730ddcafc6df84e5fc01310a1b4231bd9ce233696478c6e37789c","observation_id":"26b7a2cf-c658-48c5-a3e3-b6130722bcb6","resolution":{"observed_at":"2026-08-08T17:11:01.924506Z","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":"2026-08-08T17:11:01.928468Z","title":", title =","venue":null,"work_id":null,"year":1980},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":3,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.928468Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:f11c3e8ab7f1081b7612836259d8621f4b0d1a402bdb12dd2b23e07c36988f44","observation_id":"65bbd1af-60db-42b6-9d27-cc161d761027","resolution":{"observed_at":"2026-08-08T17:11:01.928468Z","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":"2026-08-08T17:11:01.932393Z","title":"New Ways to Make Microcircuits Smaller---Duplicate Entry","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":4,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.932393Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:0b9fc0b62a0038cb774b1010101cd5a0f5444e110e96c3e9c86aed436dfa3420","observation_id":"66dc00e3-3dd3-4967-a658-ccefad01e13b","resolution":{"observed_at":"2026-08-08T17:11:01.932393Z","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":"2026-08-08T17:11:01.936335Z","title":"Clancey and Glenn Rennels , abstract =","venue":null,"work_id":null,"year":1984},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":5,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.936335Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:c31b40836d7bd755e023513e1298216aaabc1bc991e2241dba7251c518e6d6e0","observation_id":"5d7a9da4-fb10-4d20-9f15-31774e18542f","resolution":{"observed_at":"2026-08-08T17:11:01.936335Z","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":"2026-08-08T17:11:01.940398Z","title":"and Rennels, Glenn R","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":6,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.940398Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:fcf0e295ad6331fcd300607e8cb7e0660a4f8be61ea29ea547bbd31ed694a429","observation_id":"80393bd7-7560-4cd6-8120-c8e328eebb5e","resolution":{"observed_at":"2026-08-08T17:11:01.940398Z","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":"2026-08-08T17:11:01.944452Z","title":"Poligon: A System for Parallel Problem Solving","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":7,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.944452Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:0bdc9b618187c1e8794763c806ad52f544667c3ef0c2d51f51628e0c6f16184c","observation_id":"b8ab3240-9004-4a7d-9180-1f601e3e5dee","resolution":{"observed_at":"2026-08-08T17:11:01.944452Z","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":"2026-08-08T17:11:01.948295Z","title":"Transfer of Rule-Based Expertise through a Tutorial Dialogue","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":8,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.948295Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:5b142ac9d37cae8c481519e6dd41c030847800f0268f1d1153763c29315d114e","observation_id":"b20c1548-50fe-4842-addb-481953e7b243","resolution":{"observed_at":"2026-08-08T17:11:01.948295Z","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":"2026-08-08T17:11:01.952718Z","title":"The Engineering of Qualitative Models","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":9,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.952718Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:0e0f6062c721023b8d5d967425e45ff5ce0fee720f1fd8a38efd6a1a4497aa44","observation_id":"73235bfc-83af-487e-ab9b-5b7a3c431339","resolution":{"observed_at":"2026-08-08T17:11:01.952718Z","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":"2026-08-08T17:11:01.956350Z","title":"2023 , eprint=","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":10,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.956350Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:024cbf55b201eb173661b9d48eb85b24ad5e152f19c5863df3fc0a2b06f4fb18","observation_id":"a02cbb07-5581-4f4f-9ab1-d3ce589d5f58","resolution":{"observed_at":"2026-08-08T17:11:01.956350Z","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":"2026-08-08T17:11:01.960384Z","title":"Pluto: The 'Other' Red Planet","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":11,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.960384Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:57c952eb962d1c9e4fefeaf1be25de7b2b07f298ee407f70cc9646ea7677aa3e","observation_id":"a97d2755-0259-48a5-9c36-fea325e50c59","resolution":{"observed_at":"2026-08-08T17:11:01.960384Z","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":"2026-08-08T17:11:01.964264Z","title":"Attention is All you Need , url =","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":12,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.964264Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:fd7ca0d4c082e74ebbdf3f4ef861777a84dd164acff4ef3c392b77c75a5059ed","observation_id":"612f8ce0-c1b9-47d7-9ff1-4054a0f224e9","resolution":{"observed_at":"2026-08-08T17:11:01.964264Z","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":"2026-08-08T17:11:01.967985Z","title":"2020 , eprint=","venue":null,"work_id":null,"year":2020},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":13,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.967985Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:adb6104eaffb05308638bd2dd012320294dd8d65465a85af6e894fab6f1078ed","observation_id":"db56a1f1-39fb-43f8-8ba4-a38fd9416905","resolution":{"observed_at":"2026-08-08T17:11:01.967985Z","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":"2026-08-08T17:11:01.971761Z","title":"arXiv preprint arXiv:2504.07139 , year=","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":14,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.971761Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:d85eb37319d01e7ec34a9cd1df35c86ab6547fdba520af31cdf585add41036ce","observation_id":"f3f2728d-3bae-47cf-b602-d14e3e122139","resolution":{"observed_at":"2026-08-08T17:11:01.971761Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2406.11704","last_updated":"2024-08-06T22:37:06Z","snapshot_observed_at":"2026-08-01T16:04:10.873571Z","submitted_at":"2024-06-17T16:25:04Z","title":"Nemotron-4 340B Technical Report","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2406.11704","snapshot_observed_at":"2026-08-08T17:11:01.975473Z","title":"arXiv preprint arXiv:2406.11704 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":15,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.975473Z"},"links":{"cited_paper":"/paper/2406.11704","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:b623edd7f01442c01cfaf991e0bcf569584df9c84a441c54d515bbd658110174","observation_id":"56a918ba-39a8-4fb5-85d2-72fffde59e0d","resolution":{"observed_at":"2026-08-08T17:11:01.975473Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2409.12186","last_updated":"2024-11-12T13:24:25Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2024-09-18T17:57:57Z","title":"Qwen2.5-Coder Technical Report","version":3},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2409.12186","snapshot_observed_at":"2026-08-08T17:11:01.980152Z","title":"5-coder technical report , author=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":16,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.980152Z"},"links":{"cited_paper":"/paper/2409.12186","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:61ec63c5733abb7df665765a196b32cde2ccddc3a13bddcd00db14e94c7b8cd0","observation_id":"71c3b695-b9e4-4f82-8139-f22506109253","resolution":{"observed_at":"2026-08-08T17:11:01.980152Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2407.21783","last_updated":"2024-11-23T23:27:33Z","snapshot_observed_at":"2026-07-06T18:55:11.576666Z","submitted_at":"2024-07-31T17:54:27Z","title":"The Llama 3 Herd of Models","version":3},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2407.21783","snapshot_observed_at":"2026-08-08T17:11:01.984366Z","title":"arXiv preprint arXiv:2407.21783 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":17,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.984366Z"},"links":{"cited_paper":"/paper/2407.21783","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:76f8e0509cf4890f0d60f96c05f91e4aea4a31599b459c54349054702afa76d8","observation_id":"7aa580fa-f4cc-4517-8dbc-f6a5b98b976c","resolution":{"observed_at":"2026-08-08T17:11:01.984366Z","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":"2026-08-08T17:11:01.988073Z","title":"arXiv preprint arXiv:2512.13607 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":18,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.988073Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:64c27b4b727680f3edc6a55b34385c45456bbfd754e695ae95602ff121ba4a2d","observation_id":"bfe18244-8f27-40ad-805f-9879e8c9226f","resolution":{"observed_at":"2026-08-08T17:11:01.988073Z","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":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:03.082067Z","title":"An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-Tuning , year=","venue":null,"work_id":"ddb68e83-e73a-41d6-b1d3-a9550e1ec4b0","year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":19,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.991714Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:13dc57ce435abee3e71074dca229de68f1c8a1ece3fc0865adb22a1f475bd268","observation_id":"51d201ae-1133-4d3c-a8f8-7791e2b0faa2","resolution":{"observed_at":"2026-08-08T17:11:03.086310Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2506.10910","last_updated":"2025-06-12T17:22:37Z","snapshot_observed_at":"2026-08-08T22:49:29.831091Z","submitted_at":"2025-06-12T17:22:37Z","title":"Magistral","version":1},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2506.10910","snapshot_observed_at":"2026-08-08T17:11:01.995426Z","title":"arXiv preprint arXiv:2506.10910 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":20,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.995426Z"},"links":{"cited_paper":"/paper/2506.10910","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:55e329f61e6c23370507bbd1afd25be4eecc43dec6745873ef9c7463c16f22e3","observation_id":"63f9a6ff-2722-4a98-bdb1-3722d656b668","resolution":{"observed_at":"2026-08-08T17:11:01.995426Z","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":"2026-08-08T17:11:01.999388Z","title":"OLM o: Accelerating the Science of Language Models","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":21,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:01.999388Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:6e359d3b49b9ef76f700420616b9866b657cdb911fefa6cf8c5c6976f0f348a8","observation_id":"2f317759-341d-4707-b2e2-42c57b7a81f2","resolution":{"observed_at":"2026-08-08T17:11:01.999388Z","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":"2026-08-08T17:11:02.003544Z","title":"Smith and Hannaneh Hajishirzi , booktitle=","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":22,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.003544Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:29de7802e1d18afc453e526d0fe240d0a721a0cd4d0d6f35cc630d688927ee0e","observation_id":"96a67b40-b7cc-410c-8037-0784854f92ff","resolution":{"observed_at":"2026-08-08T17:11:02.003544Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2512.13961","last_updated":"2026-04-14T15:12:44Z","snapshot_observed_at":"2026-08-07T08:18:31.274999Z","submitted_at":"2025-12-15T23:41:48Z","title":"Olmo 3","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2512.13961","snapshot_observed_at":"2026-08-08T17:11:02.007167Z","title":"arXiv preprint arXiv:2512.13961 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":23,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.007167Z"},"links":{"cited_paper":"/paper/2512.13961","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:fac9ee1725b34c0f44f8b283ae814bf1ac51e1fbf3eedbbbbea30001a29d91e6","observation_id":"b61a751c-399a-4277-8c2f-bbc7ddefb95a","resolution":{"observed_at":"2026-08-08T17:11:02.007167Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2412.19437","last_updated":"2025-02-18T17:26:38Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2024-12-27T04:03:16Z","title":"DeepSeek-V3 Technical Report","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2412.19437","snapshot_observed_at":"2026-08-08T17:11:02.011121Z","title":"arXiv preprint arXiv:2412.19437 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":24,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.011121Z"},"links":{"cited_paper":"/paper/2412.19437","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:c2e0e0f6f8d952e07a5574b61055439d309c6471b69699f2697e8e3c46b17976","observation_id":"15f43113-0a55-4b26-98f9-083d31e4a338","resolution":{"observed_at":"2026-08-08T17:11:02.011121Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2501.12948","last_updated":"2026-01-04T03:57:36Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2025-01-22T15:19:35Z","title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2501.12948","snapshot_observed_at":"2026-08-08T17:11:02.015146Z","title":"arXiv preprint arXiv:2501.12948 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":25,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.015146Z"},"links":{"cited_paper":"/paper/2501.12948","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:cf67c71d81a206cb9128b9542b2f1acf5ec60dd4a549f0fe3cf316de8669cccb","observation_id":"0d58dd51-2d8a-4357-874d-23b8d0dad506","resolution":{"observed_at":"2026-08-08T17:11:02.015146Z","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":"2026-08-08T17:11:02.019052Z","title":"2025 , eprint=","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":26,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.019052Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:1b1c38e7cc96788138147a5583876c8d37dcd4c5f462c4ff1a67d6087f531870","observation_id":"2464088d-5454-427f-9193-3e1483b0c35a","resolution":{"observed_at":"2026-08-08T17:11:02.019052Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2508.12116","last_updated":"2026-07-03T00:43:32Z","snapshot_observed_at":"2026-08-09T06:28:40.149447Z","submitted_at":"2025-08-16T18:01:39Z","title":"DynamixSFT: Dynamic Mixture Optimization of Instruction Tuning Collections","version":2},"cited_work":{"arxiv_id":"2508.12116","doi":null,"metadata_source":"pith","pith_arxiv_id":"2508.12116","snapshot_observed_at":"2026-08-08T17:11:02.665026Z","title":"DynamixSFT: Dynamic Mixture Optimization of Instruction Tuning Collections","venue":"cs.LG","work_id":"fde9a47b-c400-47e3-a8ef-8d20d41734d7","year":2025},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":27,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.022936Z"},"links":{"cited_paper":"/paper/2508.12116","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:85a2064181174c28ee42d266a4bb4319dea6d54f9d28013bffc47d9baf3aadde","observation_id":"1ec2a1a8-c8c9-4d27-afe9-3f40f7de60cb","resolution":{"observed_at":"2026-08-08T17:11:02.669339Z","resolver_source":"local_arxiv","status":"metadata_mismatch"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:03.049790Z","title":"2024 , eprint=","venue":null,"work_id":"ef8c7651-7f97-45e1-af8e-1e528acf8e36","year":2024},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":28,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.027202Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:87a3f3e511b4d1d86e5b0cfb011cb9a3e32e0d3e353740e9ff709a32869744bd","observation_id":"273aad41-1666-4d91-9c19-9c8f799432fe","resolution":{"observed_at":"2026-08-08T17:11:03.053512Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:03.038325Z","title":", booktitle =","venue":null,"work_id":"6ba60c77-cc62-4a99-b9aa-64b3e3a41cc5","year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":29,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.031868Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:d0b6a0d4d40b4b9dd87cd53eda0b05c53ea114fa2e0388cb06fbff1c67fbaab4","observation_id":"6ed44052-af84-4559-99af-d3aa00f0ec60","resolution":{"observed_at":"2026-08-08T17:11:03.042096Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:03.027009Z","title":"Journal of Machine Learning Research , year =","venue":null,"work_id":"d2e6b71a-fd42-4bf5-be87-9320669ca1f0","year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":30,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.035923Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:432510c298ddbdb2ebd5fdc8bd9f5174cf885288b6e937826d25a17793c5082a","observation_id":"9583d764-883e-4022-aa68-937a08b70579","resolution":{"observed_at":"2026-08-08T17:11:03.031093Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":"10.1016/s0893-6080(98)00010-0","metadata_source":"doi_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.297421Z","title":"Automatic early stopping using cross validation: quantifying the criteria , journal =","venue":null,"work_id":"05dcad5a-e9cb-4769-9555-8aa1e8ea0789","year":1998},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":31,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.039785Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:73a17614be4e3f5edaf6168e8cdd510a697c6462cef4df0ab7230fb10f18fce7","observation_id":"6591fbcc-deeb-4f5b-b4d0-fec4f9aa51f2","resolution":{"observed_at":"2026-08-08T17:11:02.301427Z","resolver_source":"doi","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.044038Z","title":"Revisiting Scaling Laws for Language Models: The Role of Data Quality and Training Strategies","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":32,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.044038Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:011bf114eacc6831666867f3a14cf1e4c7a3ba1e3351ab692b0c23b1c3fbbf39","observation_id":"c180e444-a76f-4c92-8cc2-b401525f67da","resolution":{"observed_at":"2026-08-08T17:11:02.044038Z","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":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:03.014559Z","title":"2025 , eprint=","venue":null,"work_id":"bada2f83-510c-47de-8453-0082685f7f8e","year":2025},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":33,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.048314Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:14016bd51da3c2619e4d9f58ea7902f589850a72556ad3eb891e051ac8e04de4","observation_id":"176b6445-575f-4342-b517-b591c940cc8c","resolution":{"observed_at":"2026-08-08T17:11:03.019032Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:03.002959Z","title":"2026 , eprint=","venue":null,"work_id":"a1768f2d-8e2e-42c1-ba19-87230d568108","year":2026},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":34,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.052073Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:199194f6496bbb0b2d0bb83f7a0228e5b8a7060321031608915eff6d22ce28c1","observation_id":"1744dbbf-a95a-4f2e-a47a-22f6bb99c0af","resolution":{"observed_at":"2026-08-08T17:11:03.006807Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.055894Z","title":"The Thirteenth International Conference on Learning Representations , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":35,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.055894Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:b8cde6e89934bc3bd5ce9deb6978ac7334348866123e1d718bc67dab2b3457e3","observation_id":"095f0c91-3102-4b46-ae10-1bed3c8f5705","resolution":{"observed_at":"2026-08-08T17:11:02.055894Z","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":"10.18653/v1/2025.naacl-long.80","metadata_source":"doi_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.276527Z","title":"Dynamic Data Mixing Maximizes Instruction Tuning for Mixture-of-Experts","venue":null,"work_id":"a8cb507c-da6a-4e26-a396-dd368fdff987","year":2025},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":36,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.059974Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:0c698335069e10eca8a4543efab6ece21639525505e732b32557c33c00cc9584","observation_id":"0baa6942-2abd-45e9-abbe-e09149c70d4e","resolution":{"observed_at":"2026-08-08T17:11:02.282262Z","resolver_source":"doi","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.064468Z","title":"How Abilities in Large Language Models are Affected by Supervised Fine-tuning Data Composition","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":37,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.064468Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:9ac3c79b5600e91dd43c7252c8207a03415df9be7698d51b88827c5494b2335d","observation_id":"66a37c46-daed-4216-b382-17d2c676a840","resolution":{"observed_at":"2026-08-08T17:11:02.064468Z","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":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.985270Z","title":"2025 , eprint=","venue":null,"work_id":"f6b5cf81-1b5b-4c5d-ae65-e55d27b5e93e","year":2025},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":38,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.068522Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:77ccf8ccb6ffa53bd5c7be4a69c767dc997986a93e7765ace7129dd771e73ec8","observation_id":"ff985bd7-3d3b-418c-a9fe-d879a1191172","resolution":{"observed_at":"2026-08-08T17:11:02.989256Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.972428Z","title":"2026 , url=","venue":null,"work_id":"16b1769d-f2d6-4d17-9ce7-df9d86f2420f","year":2026},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":39,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.072317Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:c4f8fee8b6daba01d76f470f98156a73bcfcc4b5939b73b748b17cfb4e58a1a4","observation_id":"0abe4e0d-af11-4f41-a531-3eac5ba860a0","resolution":{"observed_at":"2026-08-08T17:11:02.976157Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.075922Z","title":"From Quantity to Quality: Boosting LLM Performance with Self-Guided Data Selection for Instruction Tuning","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":40,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.075922Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:23fd2224beb29039a144f21d4e5169da2b0f0051fd03568c4d8475408f1dee32","observation_id":"cb812d02-a546-4f08-bfbe-d200173fc496","resolution":{"observed_at":"2026-08-08T17:11:02.075922Z","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":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.961431Z","title":"Advances in Neural Information Processing Systems , editor =","venue":null,"work_id":"a0d7c28a-eb48-4503-8d86-bed0991c5938","year":2020},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":41,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.080336Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:6606049cbea10b3ccbf8415c253cf2e0a7091467c3881e07c7b9b8c55b4ec227","observation_id":"7d7be7d2-82ff-4b47-86af-e1e6c83404cc","resolution":{"observed_at":"2026-08-08T17:11:02.965105Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.949996Z","title":"Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing , pages=","venue":null,"work_id":"00ef6f84-87e3-48c4-a80e-e85a4457e0d5","year":2022},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":42,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.084959Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:79b07f4ea124e49950ff65dab08a09a7e2569ef6ce93fc2cea54bf5ca1dadf53","observation_id":"ef436195-50ef-43d6-a346-46c4aa52fbb9","resolution":{"observed_at":"2026-08-08T17:11:02.953961Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.088698Z","title":"C ommonsense QA : A Question Answering Challenge Targeting Commonsense Knowledge","venue":null,"work_id":null,"year":2019},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":43,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.088698Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:36603480e86c8843f88442a11c20ca9e84bd356082a04910ff9bb0c38680ae63","observation_id":"97dd98aa-8cf1-4c5c-ba55-d683f409f3a5","resolution":{"observed_at":"2026-08-08T17:11:02.088698Z","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":"2026-08-08T17:11:02.092641Z","title":"Can a Suit of Armor Conduct Electricity? A New Dataset for Open Book Question Answering","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":44,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.092641Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:b18a5ba352b20b95cb1a13e6549606d6fbbf854ac3c7add686887cb4fcce6905","observation_id":"b246ed59-5f14-4d4b-884b-1fe8fce69efc","resolution":{"observed_at":"2026-08-08T17:11:02.092641Z","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":"2026-08-08T17:11:02.096257Z","title":"Program Induction by Rationale Generation: Learning to Solve and Explain Algebraic Word Problems","venue":null,"work_id":null,"year":2017},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":45,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.096257Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:5cbd46c945da2adee5d5df3c535c855a1f33d000e99a641e67ec8dfb35d6ac4e","observation_id":"ffe7e759-940c-4be9-b117-2d9680453395","resolution":{"observed_at":"2026-08-08T17:11:02.096257Z","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":"2026-08-08T17:11:02.099872Z","title":"2021 , eprint=","venue":null,"work_id":null,"year":2021},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":46,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.099872Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:52ebfb9708f0b573f22aef77a9bed5a77d8f6eada7fe5eb48a47d8d9af2f1ac3","observation_id":"851b3790-a830-4098-9d6d-95fa5f633b1e","resolution":{"observed_at":"2026-08-08T17:11:02.099872Z","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":"2026-08-08T17:11:02.103333Z","title":"and Gardner, Matt","venue":null,"work_id":null,"year":2017},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":47,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.103333Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:de7bd90e6f311a33d6ae98b908be2bb9501ff881b077d677800bf8877b7ebf61","observation_id":"b6a2511f-8222-45b0-812f-6ca696536dfe","resolution":{"observed_at":"2026-08-08T17:11:02.103333Z","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":"2026-08-08T17:11:02.107030Z","title":"2018 , eprint=","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":48,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.107030Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:69803527ae7361121de8f33f2e5d889622cb042fae47f19b62739bbe3ac0a8bd","observation_id":"902d7c09-649c-490a-ab60-b44648bc0619","resolution":{"observed_at":"2026-08-08T17:11:02.107030Z","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":"2026-08-08T17:11:02.110927Z","title":"H ella S wag: Can a Machine Really Finish Your Sentence?","venue":null,"work_id":null,"year":2019},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":49,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.110927Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:0310a5901c4b2eb53ecca507efd5b66a67673175ff7d6b7e09c63967b4f9a107","observation_id":"575b65b2-470e-4897-9452-0aa3da73b693","resolution":{"observed_at":"2026-08-08T17:11:02.110927Z","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":"2026-08-08T17:11:02.114583Z","title":"Proceedings of the AAAI Conference on Artificial Intelligence , volume=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":50,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.114583Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:a6939e970261f77dfe899a543abc75dcd4bbaaaf04fe9f6c81bf9ebf93c32604","observation_id":"34cba346-132b-4cc2-beb4-eaec0fec7fdc","resolution":{"observed_at":"2026-08-08T17:11:02.114583Z","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":"2026-08-08T17:11:02.117975Z","title":"B ool Q : Exploring the Surprising Difficulty of Natural Yes/No Questions","venue":null,"work_id":null,"year":2019},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":51,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.117975Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:cd98acb4b5acf1a8a82d84a74fc9b8c684ab54f95c9a0be71b67f0c644d7bd2a","observation_id":"5a610b19-f81d-477c-b00a-57241a0a8892","resolution":{"observed_at":"2026-08-08T17:11:02.117975Z","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":"2026-08-08T17:11:02.121362Z","title":"Proceedings of the Conference on Health, Inference, and Learning , pages =","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":52,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.121362Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:6813f8442d0c68313daaa7a5feb73525a2eb556630cddb6ac1c01684a168a40d","observation_id":"9440e8ea-68af-44d3-83ef-5587541a58bc","resolution":{"observed_at":"2026-08-08T17:11:02.121362Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2411.15124","last_updated":"2025-04-14T22:39:09Z","snapshot_observed_at":"2026-08-08T12:58:42.430328Z","submitted_at":"2024-11-22T18:44:04Z","title":"Tulu 3: Pushing Frontiers in Open Language Model Post-Training","version":5},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2411.15124","snapshot_observed_at":"2026-08-08T17:11:02.125832Z","title":"arXiv preprint arXiv:2411.15124 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":53,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.125832Z"},"links":{"cited_paper":"/paper/2411.15124","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:293001ff73c76dd7c8a3fbf6c77f5d2fbf0acdad3caa7f255d1404f8d6e3b285","observation_id":"cf6e7922-31ca-4789-8784-ade051d97fbd","resolution":{"observed_at":"2026-08-08T17:11:02.125832Z","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":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.913295Z","title":"Proceedings of the 2022 conference on empirical methods in natural language processing , pages=","venue":null,"work_id":"c554c673-bc1c-4302-a77e-9ea689303a9c","year":2022},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":54,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.129808Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:42f01af422377a04b6fc83d2c9992c6f92bd1502a8145c1205473bf2801927dc","observation_id":"9b0beaef-6838-46d2-8f2b-831ca045f6ac","resolution":{"observed_at":"2026-08-08T17:11:02.917434Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2508.11953","last_updated":"2025-08-16T07:28:39Z","snapshot_observed_at":"2026-08-05T19:39:29.759828Z","submitted_at":"2025-08-16T07:28:39Z","title":"Data Mixing Optimization for Supervised Fine-Tuning of Large Language Models","version":1},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2508.11953","snapshot_observed_at":"2026-08-08T17:11:02.133177Z","title":"arXiv preprint arXiv:2508.11953 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":55,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.133177Z"},"links":{"cited_paper":"/paper/2508.11953","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:e5e9765543f92796183bb2558127d83a4d19327b6ec04c5c1e0796d76f0028c9","observation_id":"222a1d89-64c8-40b8-97dc-ec19b59e88fe","resolution":{"observed_at":"2026-08-08T17:11:02.133177Z","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":"2026-08-08T17:11:02.136783Z","title":"International conference on machine learning , pages=","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":56,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.136783Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:c8acddffd253a587047bfd71601a58f82ed67f2a3a2a6319427cfe8cd8e5fb62","observation_id":"ed7522e3-e34d-4adf-b026-b1d0bff25497","resolution":{"observed_at":"2026-08-08T17:11:02.136783Z","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":"2026-08-08T17:11:02.140289Z","title":"Advances in neural information processing systems , volume=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":57,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.140289Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:3a89cdc1040ff90bb6858cdbc76761c27168772ae84075a9910f81d311c23686","observation_id":"ff6d394a-78c2-4a3f-83aa-5e5d3916c931","resolution":{"observed_at":"2026-08-08T17:11:02.140289Z","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":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.888435Z","title":"Advances in neural information processing systems , volume=","venue":null,"work_id":"0dbb1e87-ecf4-454e-9b03-2827ab262d9d","year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":58,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.143758Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:71f5cba7a139e48d8e4b49d845ae14dc0fb56dd233f90a69ca391119860c35fb","observation_id":"5a82404f-7951-4552-85b8-971ab7801c3e","resolution":{"observed_at":"2026-08-08T17:11:02.893010Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.147245Z","title":"Advances in Neural Information Processing Systems , volume=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":59,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.147245Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:0046db3909d161c9c0e8abaede6593d0196380b9961299043b06a7983cd78d3c","observation_id":"829e4c4a-7b91-425d-ab22-aaf48cd68d26","resolution":{"observed_at":"2026-08-08T17:11:02.147245Z","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":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.869739Z","title":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages=","venue":null,"work_id":"db34e554-26a8-416e-8fd7-c7bcf8aafa38","year":2024},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":60,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.150697Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:edfc4d15cf471024dbe9f191e64a5b2ba23dd8cf0a36ca37679a96442584f3e2","observation_id":"ebbc110b-d718-43fd-9b49-163849057606","resolution":{"observed_at":"2026-08-08T17:11:02.874722Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2410.05248","last_updated":"2026-04-18T22:35:58Z","snapshot_observed_at":"2026-07-06T19:29:09.714725Z","submitted_at":"2024-10-07T17:52:21Z","title":"SFTMix: Elevating Language Model Instruction Tuning with Mixup Recipe","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2410.05248","snapshot_observed_at":"2026-08-08T17:11:02.153955Z","title":"arXiv preprint arXiv:2410.05248 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":61,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.153955Z"},"links":{"cited_paper":"/paper/2410.05248","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:09568d7765a6daa5c123a160919767e04ea11e5f0933d0498ea09eec105000a3","observation_id":"cbc188b6-4755-47e8-8a31-01a7ef8f67ec","resolution":{"observed_at":"2026-08-08T17:11:02.153955Z","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":"2603.21606","doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.617947Z","title":"arXiv preprint arXiv:2603.21606 , year=","venue":null,"work_id":"5da491a4-28f0-4916-aeac-c4933ddefcd6","year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":62,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.157668Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:1a11e2abdbd19392ad15e6e6668931548420f6256345c2b4c293ed849efec2be","observation_id":"6c3142d9-9c37-4bad-b1a6-fea1d8088ebc","resolution":{"observed_at":"2026-08-08T17:11:02.623407Z","resolver_source":"raw_fallback","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.161150Z","title":"arXiv preprint arXiv:2505.18738 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":63,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.161150Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:1910b447490fa8d7d356fd32677a6a3370b9970d551a562023e6727b2091761a","observation_id":"3e21ff3b-aaee-48db-b830-59de57f52ae3","resolution":{"observed_at":"2026-08-08T17:11:02.161150Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2106.09685","last_updated":"2021-10-16T18:40:34Z","snapshot_observed_at":"2026-08-07T07:43:16.294957Z","submitted_at":"2021-06-17T17:37:18Z","title":"LoRA: Low-Rank Adaptation of Large Language Models","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2106.09685","snapshot_observed_at":"2026-08-08T17:11:02.164466Z","title":"arXiv preprint arXiv:2106.09685 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":64,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.164466Z"},"links":{"cited_paper":"/paper/2106.09685","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:4773c1f3751f372b90680255ccc0331c661304071e9a6486666bb64866f35e71","observation_id":"37a8c7e3-a17d-40f7-b4ae-336e2dcde4b2","resolution":{"observed_at":"2026-08-08T17:11:02.164466Z","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":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-08-08T17:11:02.857146Z","title":"Proceedings of the 36th International Conference on Machine Learning , year=","venue":null,"work_id":"7e21d774-7d93-4220-b634-34fd96ef84a4","year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":65,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.168576Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:2f07134043aeddda986013e6527e95cf2743eb14c8e2ea58baecedf8fc78b41b","observation_id":"f0628d89-d5e9-4bdf-846c-3e47b31f96a0","resolution":{"observed_at":"2026-08-08T17:11:02.861214Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2101.00190","last_updated":"2021-01-01T08:00:36Z","snapshot_observed_at":"2026-07-06T10:29:18.734092Z","submitted_at":"2021-01-01T08:00:36Z","title":"Prefix-Tuning: Optimizing Continuous Prompts for Generation","version":1},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2101.00190","snapshot_observed_at":"2026-08-08T17:11:02.172024Z","title":"arXiv preprint arXiv:2101.00190 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":66,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.172024Z"},"links":{"cited_paper":"/paper/2101.00190","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:63428f5777b25d0b19034923bb5531e86247245e2f57431a63176e55124cfa68","observation_id":"a34dbec1-d63d-4b7c-a627-7db9e58f1e33","resolution":{"observed_at":"2026-08-08T17:11:02.172024Z","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":"2026-08-08T17:11:02.176825Z","title":"Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , year=","venue":null,"work_id":null,"year":2021},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":67,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.176825Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:17d6cb828dfe9d66e8c51fa97a893b5923c76d50b07323142d9b3cf91f9c6888","observation_id":"a9929ffa-4503-4b52-8aa6-f7e6a9ef0135","resolution":{"observed_at":"2026-08-08T17:11:02.176825Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2305.14314","last_updated":"2023-05-23T17:50:33Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2023-05-23T17:50:33Z","title":"QLoRA: Efficient Finetuning of Quantized LLMs","version":1},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2305.14314","snapshot_observed_at":"2026-08-08T17:11:02.181412Z","title":"arXiv preprint arXiv:2305.14314 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":68,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.181412Z"},"links":{"cited_paper":"/paper/2305.14314","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:2b2094d2845c6788c68061f1fc388323d7d38d7c6c9e36fcd1431481297a2153","observation_id":"801ec374-b896-46de-9b5f-e2fbe2c4e803","resolution":{"observed_at":"2026-08-08T17:11:02.181412Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2303.10512","last_updated":"2023-12-20T20:56:14Z","snapshot_observed_at":"2026-07-06T15:05:16.471478Z","submitted_at":"2023-03-18T22:36:25Z","title":"AdaLoRA: Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2303.10512","snapshot_observed_at":"2026-08-08T17:11:02.185238Z","title":"arXiv preprint arXiv:2303.10512 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":69,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.185238Z"},"links":{"cited_paper":"/paper/2303.10512","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:80b803d2c64a1c6d99ccabc5e7c9d1d7dc4983579161cb576f12627e8a587cb9","observation_id":"ba074b6f-cdc0-4e93-bc70-8f533cff1713","resolution":{"observed_at":"2026-08-08T17:11:02.185238Z","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":"2026-08-08T17:11:02.188994Z","title":"arXiv preprint arXiv:2308.10792 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":70,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.188994Z"},"links":{"citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:83c1d8d0c8fb22895efb586056d8bf82e61f8575deacd469f5a55c12dab729bd","observation_id":"cd21da5b-ecb2-4f2f-a5a7-220bb84837e4","resolution":{"observed_at":"2026-08-08T17:11:02.188994Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2402.09353","last_updated":"2024-07-09T05:59:16Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2024-02-14T17:59:34Z","title":"DoRA: Weight-Decomposed Low-Rank Adaptation","version":6},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2402.09353","snapshot_observed_at":"2026-08-08T17:11:02.192352Z","title":"arXiv preprint arXiv:2402.09353 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":71,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.192352Z"},"links":{"cited_paper":"/paper/2402.09353","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:6fa8e7174712c90a61c13211069013338f34c8f9c4826f7506c1daca53ac00fd","observation_id":"3cd40e0c-24eb-451e-9d74-b021d0a861be","resolution":{"observed_at":"2026-08-08T17:11:02.192352Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2407.05000","last_updated":"2024-07-16T07:32:23Z","snapshot_observed_at":"2026-07-06T18:42:18.289966Z","submitted_at":"2024-07-06T08:37:21Z","title":"LoRA-GA: Low-Rank Adaptation with Gradient Approximation","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2407.05000","snapshot_observed_at":"2026-08-08T17:11:02.195801Z","title":"arXiv preprint arXiv:2407.05000 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","version":1},"reference_index":72,"source":"arxiv_source","source_observed_at":"2026-08-08T17:11:02.195801Z"},"links":{"cited_paper":"/paper/2407.05000","citing_paper":"/paper/2608.05250"},"observation_digest":"sha256:55aff4f474187ee1ee325756c3999e8ef8a3c1475a513a7ff17594c3bc0702ce","observation_id":"613486fe-4142-4316-9292-2f82f01213ad","resolution":{"observed_at":"2026-08-08T17:11:02.195801Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}}],"paper":{"arxiv_id":"2608.05250","last_updated":"2026-08-05T15:57:43Z","latest_version":1,"primary_category":"cs.LG","snapshot_observed_at":"2026-08-09T11:10:59.714636Z","submitted_at":"2026-08-05T15:57:43Z","title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning"},"reference_resolution":{"displayed":72,"state_counts":{"malformed_identifier":0,"metadata_mismatch":1,"parse_uncertain":0,"unresolved":54,"verified_exact":3,"verified_fuzzy":14},"total_outbound_references":72},"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":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"thesis":"As of 9 August 2026, this Paper Citation Record lists 72 of 72 outbound references and 0 inbound Pith citation observations for arXiv:2608.05250."}