{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:SX2P6YOOVHTHPDJ3UPAMZMXJZE","short_pith_number":"pith:SX2P6YOO","canonical_record":{"source":{"id":"2403.08763","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-03-13T17:58:57Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"bc0a7dd09b961d0434778597f6f5760582a445aa1ca0c9c6892eb25327c1b8f0","abstract_canon_sha256":"727811dfbd0842cb68edb01b326db42474a73d1efa9e2aec4dbd4276043d1518"},"schema_version":"1.0"},"canonical_sha256":"95f4ff61cea9e6778d3ba3c0ccb2e9c91f0982ea6987e9f71da3d51c6b07b592","source":{"kind":"arxiv","id":"2403.08763","version":4},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2403.08763","created_at":"2026-07-05T09:02:57Z"},{"alias_kind":"arxiv_version","alias_value":"2403.08763v4","created_at":"2026-07-05T09:02:57Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.08763","created_at":"2026-07-05T09:02:57Z"},{"alias_kind":"pith_short_12","alias_value":"SX2P6YOOVHTH","created_at":"2026-07-05T09:02:57Z"},{"alias_kind":"pith_short_16","alias_value":"SX2P6YOOVHTHPDJ3","created_at":"2026-07-05T09:02:57Z"},{"alias_kind":"pith_short_8","alias_value":"SX2P6YOO","created_at":"2026-07-05T09:02:57Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:SX2P6YOOVHTHPDJ3UPAMZMXJZE","target":"record","payload":{"canonical_record":{"source":{"id":"2403.08763","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-03-13T17:58:57Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"bc0a7dd09b961d0434778597f6f5760582a445aa1ca0c9c6892eb25327c1b8f0","abstract_canon_sha256":"727811dfbd0842cb68edb01b326db42474a73d1efa9e2aec4dbd4276043d1518"},"schema_version":"1.0"},"canonical_sha256":"95f4ff61cea9e6778d3ba3c0ccb2e9c91f0982ea6987e9f71da3d51c6b07b592","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:02:57.098895Z","signature_b64":"v8C0znzCbUH9LM6xffWGd79G82+DKjqaq8MoZCEmUJx8NthdiL3t8s0yOcJqB01aSeYEUZkY/le6WmdPKdX5AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"95f4ff61cea9e6778d3ba3c0ccb2e9c91f0982ea6987e9f71da3d51c6b07b592","last_reissued_at":"2026-07-05T09:02:57.098368Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:02:57.098368Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2403.08763","source_version":4,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T09:02:57Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"N3Esj8Rtmf1s481xiSAtFOIFgxUpBOF7lL1AJTpmEvVI6x4QjUT8nPMR9Vn0QLdWzfUVirF9qydRkkm57NRyDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-24T19:37:16.609127Z"},"content_sha256":"dc9759b9fd52dd3de2574d70db89aca0d0c84b3deef7bb2cc3ed228a731ef5aa","schema_version":"1.0","event_id":"sha256:dc9759b9fd52dd3de2574d70db89aca0d0c84b3deef7bb2cc3ed228a731ef5aa"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:SX2P6YOOVHTHPDJ3UPAMZMXJZE","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Simple and Scalable Strategies to Continually Pre-train Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Adam Ibrahim, Benjamin Th\\'erien, Eugene Belilovsky, Irina Rish, Kshitij Gupta, Mats L. Richter, Quentin Anthony, Timoth\\'ee Lesort","submitted_at":"2024-03-13T17:58:57Z","abstract_excerpt":"Large language models (LLMs) are routinely pre-trained on billions of tokens, only to start the process over again once new data becomes available. A much more efficient solution is to continually pre-train these models, saving significant compute compared to re-training. However, the distribution shift induced by new data typically results in degraded performance on previous data or poor adaptation to the new data. In this work, we show that a simple and scalable combination of learning rate (LR) re-warming, LR re-decaying, and replay of previous data is sufficient to match the performance of"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.08763","kind":"arxiv","version":4},"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/2403.08763/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T09:02:57Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"89HMhsy+5GWH/GB0i5VjminJHkMt4iI0EoAn6vXyVwtIWCFjpiRahxfauBRjAyrKnIR3P1IayIkZUAIDksspBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-24T19:37:16.609503Z"},"content_sha256":"0ab1a511bb949a5c8ac89549216a103e87b45ebb18ec70900ff80d175a123bd0","schema_version":"1.0","event_id":"sha256:0ab1a511bb949a5c8ac89549216a103e87b45ebb18ec70900ff80d175a123bd0"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/SX2P6YOOVHTHPDJ3UPAMZMXJZE/bundle.json","state_url":"https://pith.science/pith/SX2P6YOOVHTHPDJ3UPAMZMXJZE/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/SX2P6YOOVHTHPDJ3UPAMZMXJZE/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-24T19:37:16Z","links":{"resolver":"https://pith.science/pith/SX2P6YOOVHTHPDJ3UPAMZMXJZE","bundle":"https://pith.science/pith/SX2P6YOOVHTHPDJ3UPAMZMXJZE/bundle.json","state":"https://pith.science/pith/SX2P6YOOVHTHPDJ3UPAMZMXJZE/state.json","well_known_bundle":"https://pith.science/.well-known/pith/SX2P6YOOVHTHPDJ3UPAMZMXJZE/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:SX2P6YOOVHTHPDJ3UPAMZMXJZE","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"727811dfbd0842cb68edb01b326db42474a73d1efa9e2aec4dbd4276043d1518","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-03-13T17:58:57Z","title_canon_sha256":"bc0a7dd09b961d0434778597f6f5760582a445aa1ca0c9c6892eb25327c1b8f0"},"schema_version":"1.0","source":{"id":"2403.08763","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2403.08763","created_at":"2026-07-05T09:02:57Z"},{"alias_kind":"arxiv_version","alias_value":"2403.08763v4","created_at":"2026-07-05T09:02:57Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.08763","created_at":"2026-07-05T09:02:57Z"},{"alias_kind":"pith_short_12","alias_value":"SX2P6YOOVHTH","created_at":"2026-07-05T09:02:57Z"},{"alias_kind":"pith_short_16","alias_value":"SX2P6YOOVHTHPDJ3","created_at":"2026-07-05T09:02:57Z"},{"alias_kind":"pith_short_8","alias_value":"SX2P6YOO","created_at":"2026-07-05T09:02:57Z"}],"graph_snapshots":[{"event_id":"sha256:0ab1a511bb949a5c8ac89549216a103e87b45ebb18ec70900ff80d175a123bd0","target":"graph","created_at":"2026-07-05T09:02:57Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2403.08763/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large language models (LLMs) are routinely pre-trained on billions of tokens, only to start the process over again once new data becomes available. A much more efficient solution is to continually pre-train these models, saving significant compute compared to re-training. However, the distribution shift induced by new data typically results in degraded performance on previous data or poor adaptation to the new data. In this work, we show that a simple and scalable combination of learning rate (LR) re-warming, LR re-decaying, and replay of previous data is sufficient to match the performance of","authors_text":"Adam Ibrahim, Benjamin Th\\'erien, Eugene Belilovsky, Irina Rish, Kshitij Gupta, Mats L. Richter, Quentin Anthony, Timoth\\'ee Lesort","cross_cats":["cs.AI","cs.CL"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-03-13T17:58:57Z","title":"Simple and Scalable Strategies to Continually Pre-train Large Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.08763","kind":"arxiv","version":4},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:dc9759b9fd52dd3de2574d70db89aca0d0c84b3deef7bb2cc3ed228a731ef5aa","target":"record","created_at":"2026-07-05T09:02:57Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"727811dfbd0842cb68edb01b326db42474a73d1efa9e2aec4dbd4276043d1518","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-03-13T17:58:57Z","title_canon_sha256":"bc0a7dd09b961d0434778597f6f5760582a445aa1ca0c9c6892eb25327c1b8f0"},"schema_version":"1.0","source":{"id":"2403.08763","kind":"arxiv","version":4}},"canonical_sha256":"95f4ff61cea9e6778d3ba3c0ccb2e9c91f0982ea6987e9f71da3d51c6b07b592","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"95f4ff61cea9e6778d3ba3c0ccb2e9c91f0982ea6987e9f71da3d51c6b07b592","first_computed_at":"2026-07-05T09:02:57.098368Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T09:02:57.098368Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"v8C0znzCbUH9LM6xffWGd79G82+DKjqaq8MoZCEmUJx8NthdiL3t8s0yOcJqB01aSeYEUZkY/le6WmdPKdX5AQ==","signature_status":"signed_v1","signed_at":"2026-07-05T09:02:57.098895Z","signed_message":"canonical_sha256_bytes"},"source_id":"2403.08763","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:dc9759b9fd52dd3de2574d70db89aca0d0c84b3deef7bb2cc3ed228a731ef5aa","sha256:0ab1a511bb949a5c8ac89549216a103e87b45ebb18ec70900ff80d175a123bd0"],"state_sha256":"16b1e3737e390fac8eb7ed6859b64267ae0ba4f14d598498a0d3f89d501a0e65"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"39iYoh2NnP0UlEM/1mEp5LXPMKwKMv6uloxAdFEvBwG94IbBBDvKbYqAoesZQAYOGBZDExmVlxjct0FsTxZtCg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-24T19:37:16.611988Z","bundle_sha256":"96d0276ba2225cafd84a22a8269740828beba0f403295d48b0e12cf9d56c3ff7"}}