{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:67G7AXIDSFBUU6NXZQZ6O4SQDH","short_pith_number":"pith:67G7AXID","schema_version":"1.0","canonical_sha256":"f7cdf05d0391434a79b7cc33e7725019d95409c0b344a67b917c5e2ea8e1cdaa","source":{"kind":"arxiv","id":"2606.04272","version":1},"attestation_state":"computed","paper":{"title":"RL Excursions during Pre-Training: Re-examining Policy Optimization for LLM training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Clara Mohri, David Alvarez-Melis, Rachit Bansal, Sham Kakade, Tian Qin","submitted_at":"2026-06-02T22:55:18Z","abstract_excerpt":"The standard LLM training pipeline applies reinforcement learning (RL) only after pre-training and supervised fine-tuning (SFT). We question this status quo by training a LLM from scratch and applying RL, SFT, and SFT followed by RL directly to intermediate pre-training checkpoints. We find that RL is effective very early, and often matches the full SFT$\\to$RL pipeline early as well. Through experiments on harder problems, we find that targeted pre-training data composition is a strong lever for RL effectiveness, even more so than model scale. Beyond reasoning accuracy, applying RL directly to"},"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":"2606.04272","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-02T22:55:18Z","cross_cats_sorted":[],"title_canon_sha256":"e0a11feb5d8b1cd8d5371e64ca145ef1936b3d87c36083152e984c159318b28e","abstract_canon_sha256":"af776880cb8ce2c3e4806d9cfd2fe45edaa416f8a331f745c8443fa6169eb135"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-04T01:09:01.201399Z","signature_b64":"p0SQkLnV9sNV/oiN8HPcY8E3oI084IQ9cCTH3vmxTkNWl17NvdhTQQdtK1Bk9OX5OowCnvD+yK1okWaHrUJrCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f7cdf05d0391434a79b7cc33e7725019d95409c0b344a67b917c5e2ea8e1cdaa","last_reissued_at":"2026-06-04T01:09:01.200946Z","signature_status":"signed_v1","first_computed_at":"2026-06-04T01:09:01.200946Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RL Excursions during Pre-Training: Re-examining Policy Optimization for LLM training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Clara Mohri, David Alvarez-Melis, Rachit Bansal, Sham Kakade, Tian Qin","submitted_at":"2026-06-02T22:55:18Z","abstract_excerpt":"The standard LLM training pipeline applies reinforcement learning (RL) only after pre-training and supervised fine-tuning (SFT). We question this status quo by training a LLM from scratch and applying RL, SFT, and SFT followed by RL directly to intermediate pre-training checkpoints. We find that RL is effective very early, and often matches the full SFT$\\to$RL pipeline early as well. Through experiments on harder problems, we find that targeted pre-training data composition is a strong lever for RL effectiveness, even more so than model scale. Beyond reasoning accuracy, applying RL directly to"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.04272","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/2606.04272/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":"2606.04272","created_at":"2026-06-04T01:09:01.201016+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.04272v1","created_at":"2026-06-04T01:09:01.201016+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.04272","created_at":"2026-06-04T01:09:01.201016+00:00"},{"alias_kind":"pith_short_12","alias_value":"67G7AXIDSFBU","created_at":"2026-06-04T01:09:01.201016+00:00"},{"alias_kind":"pith_short_16","alias_value":"67G7AXIDSFBUU6NX","created_at":"2026-06-04T01:09:01.201016+00:00"},{"alias_kind":"pith_short_8","alias_value":"67G7AXID","created_at":"2026-06-04T01:09:01.201016+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/67G7AXIDSFBUU6NXZQZ6O4SQDH","json":"https://pith.science/pith/67G7AXIDSFBUU6NXZQZ6O4SQDH.json","graph_json":"https://pith.science/api/pith-number/67G7AXIDSFBUU6NXZQZ6O4SQDH/graph.json","events_json":"https://pith.science/api/pith-number/67G7AXIDSFBUU6NXZQZ6O4SQDH/events.json","paper":"https://pith.science/paper/67G7AXID"},"agent_actions":{"view_html":"https://pith.science/pith/67G7AXIDSFBUU6NXZQZ6O4SQDH","download_json":"https://pith.science/pith/67G7AXIDSFBUU6NXZQZ6O4SQDH.json","view_paper":"https://pith.science/paper/67G7AXID","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.04272&json=true","fetch_graph":"https://pith.science/api/pith-number/67G7AXIDSFBUU6NXZQZ6O4SQDH/graph.json","fetch_events":"https://pith.science/api/pith-number/67G7AXIDSFBUU6NXZQZ6O4SQDH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/67G7AXIDSFBUU6NXZQZ6O4SQDH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/67G7AXIDSFBUU6NXZQZ6O4SQDH/action/storage_attestation","attest_author":"https://pith.science/pith/67G7AXIDSFBUU6NXZQZ6O4SQDH/action/author_attestation","sign_citation":"https://pith.science/pith/67G7AXIDSFBUU6NXZQZ6O4SQDH/action/citation_signature","submit_replication":"https://pith.science/pith/67G7AXIDSFBUU6NXZQZ6O4SQDH/action/replication_record"}},"created_at":"2026-06-04T01:09:01.201016+00:00","updated_at":"2026-06-04T01:09:01.201016+00:00"}