{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6RMQDXRDVIVQ7UEFO6M73Y7AWW","short_pith_number":"pith:6RMQDXRD","schema_version":"1.0","canonical_sha256":"f45901de23aa2b0fd0857799fde3e0b5ba023f7be0c66b8138533dcc8ed036b5","source":{"kind":"arxiv","id":"2508.12116","version":1},"attestation_state":"computed","paper":{"title":"DynamixSFT: Dynamic Mixture Optimization of Instruction Tuning Collections","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Haebin Shin, Lei Ji, Qi Chen, Xiao Liu, Yeyun Gong, Zhiwei Yu","submitted_at":"2025-08-16T18:01:39Z","abstract_excerpt":"As numerous instruction-tuning datasets continue to emerge during the post-training stage, dynamically balancing and optimizing their mixtures has become a critical challenge. To address this, we propose DynamixSFT, a dynamic and automated method for instruction-tuning dataset mixture optimization. We formulate the problem as a multi-armed bandit setup and introduce a Prior-scaled Boltzmann Exploration that softly anchors the updated sampling distribution to the original dataset proportions, thereby preserving the inherent diversity and coverage of the collection. Sampling probabilities are up"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2508.12116","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-16T18:01:39Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"9d926d418da2fcfe92f25bddb13df477899b5a33c237356b38b5a7a46bd4d717","abstract_canon_sha256":"7c47a2678d5fe9fb5c13bc872f60ac6ad1806cd0eac23dbbf3311f58a8cf4e5d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:55:00.460863Z","signature_b64":"a4rVqytXXLlTGRN5AC3p1ROuKq/1DN2ZDkkqvkV5pg9i66LQ6xV5ZyFZ8dTM/CKqM9grm+902kYoyYowT+VuDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f45901de23aa2b0fd0857799fde3e0b5ba023f7be0c66b8138533dcc8ed036b5","last_reissued_at":"2026-07-05T11:55:00.460285Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:55:00.460285Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DynamixSFT: Dynamic Mixture Optimization of Instruction Tuning Collections","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Haebin Shin, Lei Ji, Qi Chen, Xiao Liu, Yeyun Gong, Zhiwei Yu","submitted_at":"2025-08-16T18:01:39Z","abstract_excerpt":"As numerous instruction-tuning datasets continue to emerge during the post-training stage, dynamically balancing and optimizing their mixtures has become a critical challenge. To address this, we propose DynamixSFT, a dynamic and automated method for instruction-tuning dataset mixture optimization. We formulate the problem as a multi-armed bandit setup and introduce a Prior-scaled Boltzmann Exploration that softly anchors the updated sampling distribution to the original dataset proportions, thereby preserving the inherent diversity and coverage of the collection. Sampling probabilities are up"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.12116","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2508.12116/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2508.12116","created_at":"2026-07-05T11:55:00.460360+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.12116v1","created_at":"2026-07-05T11:55:00.460360+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.12116","created_at":"2026-07-05T11:55:00.460360+00:00"},{"alias_kind":"pith_short_12","alias_value":"6RMQDXRDVIVQ","created_at":"2026-07-05T11:55:00.460360+00:00"},{"alias_kind":"pith_short_16","alias_value":"6RMQDXRDVIVQ7UEF","created_at":"2026-07-05T11:55:00.460360+00:00"},{"alias_kind":"pith_short_8","alias_value":"6RMQDXRD","created_at":"2026-07-05T11:55:00.460360+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2608.05250","citing_title":"Beyond Full-Model Rollback: AuroSFT for Adapter-State Multi-Task Fine-Tuning","ref_index":27,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6RMQDXRDVIVQ7UEFO6M73Y7AWW","json":"https://pith.science/pith/6RMQDXRDVIVQ7UEFO6M73Y7AWW.json","graph_json":"https://pith.science/api/pith-number/6RMQDXRDVIVQ7UEFO6M73Y7AWW/graph.json","events_json":"https://pith.science/api/pith-number/6RMQDXRDVIVQ7UEFO6M73Y7AWW/events.json","paper":"https://pith.science/paper/6RMQDXRD"},"agent_actions":{"view_html":"https://pith.science/pith/6RMQDXRDVIVQ7UEFO6M73Y7AWW","download_json":"https://pith.science/pith/6RMQDXRDVIVQ7UEFO6M73Y7AWW.json","view_paper":"https://pith.science/paper/6RMQDXRD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.12116&json=true","fetch_graph":"https://pith.science/api/pith-number/6RMQDXRDVIVQ7UEFO6M73Y7AWW/graph.json","fetch_events":"https://pith.science/api/pith-number/6RMQDXRDVIVQ7UEFO6M73Y7AWW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6RMQDXRDVIVQ7UEFO6M73Y7AWW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6RMQDXRDVIVQ7UEFO6M73Y7AWW/action/storage_attestation","attest_author":"https://pith.science/pith/6RMQDXRDVIVQ7UEFO6M73Y7AWW/action/author_attestation","sign_citation":"https://pith.science/pith/6RMQDXRDVIVQ7UEFO6M73Y7AWW/action/citation_signature","submit_replication":"https://pith.science/pith/6RMQDXRDVIVQ7UEFO6M73Y7AWW/action/replication_record"}},"created_at":"2026-07-05T11:55:00.460360+00:00","updated_at":"2026-07-05T11:55:00.460360+00:00"}