{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:W72U54EM6JM6AFMSCZ2WGNQYOU","short_pith_number":"pith:W72U54EM","schema_version":"1.0","canonical_sha256":"b7f54ef08cf259e015921675633618750c3cbe5fe08ead1dda4a45b694fd834f","source":{"kind":"arxiv","id":"2503.09029","version":1},"attestation_state":"computed","paper":{"title":"DAST: Difficulty-Aware Self-Training on Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Boyang Xue, Fei Mi, Hongling Xu, Hongru Wang, Kam-Fai Wong, Lifeng Shang, Qi Zhu, Qun Liu, Rui Wang, Sheng Wang, Yasheng Wang","submitted_at":"2025-03-12T03:36:45Z","abstract_excerpt":"Present Large Language Models (LLM) self-training methods always under-sample on challenging queries, leading to inadequate learning on difficult problems which limits LLMs' ability. Therefore, this work proposes a difficulty-aware self-training (DAST) framework that focuses on improving both the quantity and quality of self-generated responses on challenging queries during self-training. DAST is specified in three components: 1) sampling-based difficulty level estimation, 2) difficulty-aware data augmentation, and 3) the self-training algorithm using SFT and DPO respectively. Experiments on m"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2503.09029","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-03-12T03:36:45Z","cross_cats_sorted":[],"title_canon_sha256":"ce1df9c4aedfc5b510016fe61bdea52f45c722cbf6bbe28de375b721a58dea43","abstract_canon_sha256":"fc00dd4fc93249f7711aecafb0ac83095c2cddc021603f778dda91bfb816636f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:29:46.248564Z","signature_b64":"g/+tYTEmqN40sUheFk2KhP07Ju72a7nOm2HempLIdGEaE6KYJHEPDHyjh9qtfQTyKHYz5IooFnhwUItJBsFCAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b7f54ef08cf259e015921675633618750c3cbe5fe08ead1dda4a45b694fd834f","last_reissued_at":"2026-07-05T10:29:46.247988Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:29:46.247988Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DAST: Difficulty-Aware Self-Training on Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Boyang Xue, Fei Mi, Hongling Xu, Hongru Wang, Kam-Fai Wong, Lifeng Shang, Qi Zhu, Qun Liu, Rui Wang, Sheng Wang, Yasheng Wang","submitted_at":"2025-03-12T03:36:45Z","abstract_excerpt":"Present Large Language Models (LLM) self-training methods always under-sample on challenging queries, leading to inadequate learning on difficult problems which limits LLMs' ability. Therefore, this work proposes a difficulty-aware self-training (DAST) framework that focuses on improving both the quantity and quality of self-generated responses on challenging queries during self-training. DAST is specified in three components: 1) sampling-based difficulty level estimation, 2) difficulty-aware data augmentation, and 3) the self-training algorithm using SFT and DPO respectively. Experiments on m"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.09029","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/2503.09029/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":"2503.09029","created_at":"2026-07-05T10:29:46.248054+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.09029v1","created_at":"2026-07-05T10:29:46.248054+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.09029","created_at":"2026-07-05T10:29:46.248054+00:00"},{"alias_kind":"pith_short_12","alias_value":"W72U54EM6JM6","created_at":"2026-07-05T10:29:46.248054+00:00"},{"alias_kind":"pith_short_16","alias_value":"W72U54EM6JM6AFMS","created_at":"2026-07-05T10:29:46.248054+00:00"},{"alias_kind":"pith_short_8","alias_value":"W72U54EM","created_at":"2026-07-05T10:29:46.248054+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.04065","citing_title":"Progressive Mastery: Customized Curriculum Learning with Guided Prompting for Mathematical Reasoning","ref_index":30,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/W72U54EM6JM6AFMSCZ2WGNQYOU","json":"https://pith.science/pith/W72U54EM6JM6AFMSCZ2WGNQYOU.json","graph_json":"https://pith.science/api/pith-number/W72U54EM6JM6AFMSCZ2WGNQYOU/graph.json","events_json":"https://pith.science/api/pith-number/W72U54EM6JM6AFMSCZ2WGNQYOU/events.json","paper":"https://pith.science/paper/W72U54EM"},"agent_actions":{"view_html":"https://pith.science/pith/W72U54EM6JM6AFMSCZ2WGNQYOU","download_json":"https://pith.science/pith/W72U54EM6JM6AFMSCZ2WGNQYOU.json","view_paper":"https://pith.science/paper/W72U54EM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.09029&json=true","fetch_graph":"https://pith.science/api/pith-number/W72U54EM6JM6AFMSCZ2WGNQYOU/graph.json","fetch_events":"https://pith.science/api/pith-number/W72U54EM6JM6AFMSCZ2WGNQYOU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/W72U54EM6JM6AFMSCZ2WGNQYOU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/W72U54EM6JM6AFMSCZ2WGNQYOU/action/storage_attestation","attest_author":"https://pith.science/pith/W72U54EM6JM6AFMSCZ2WGNQYOU/action/author_attestation","sign_citation":"https://pith.science/pith/W72U54EM6JM6AFMSCZ2WGNQYOU/action/citation_signature","submit_replication":"https://pith.science/pith/W72U54EM6JM6AFMSCZ2WGNQYOU/action/replication_record"}},"created_at":"2026-07-05T10:29:46.248054+00:00","updated_at":"2026-07-05T10:29:46.248054+00:00"}