{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:DA5PMZWEDSJIBAILJERAP22J73","short_pith_number":"pith:DA5PMZWE","canonical_record":{"source":{"id":"2605.12705","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T20:08:00Z","cross_cats_sorted":[],"title_canon_sha256":"6fe29006452258f23bebdc91ad610ebd316c92593f87cbf3ba80d94371051193","abstract_canon_sha256":"e8c0907e9d5fb802a936c88d8ea0be29e73c356d09aaa643f42975db6acbeb5c"},"schema_version":"1.0"},"canonical_sha256":"183af666c41c9280810b492207eb49fefd2f347f6fbf777d118b30fac1521edf","source":{"kind":"arxiv","id":"2605.12705","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12705","created_at":"2026-05-18T03:09:49Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12705v1","created_at":"2026-05-18T03:09:49Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12705","created_at":"2026-05-18T03:09:49Z"},{"alias_kind":"pith_short_12","alias_value":"DA5PMZWEDSJI","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"DA5PMZWEDSJIBAIL","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"DA5PMZWE","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:DA5PMZWEDSJIBAILJERAP22J73","target":"record","payload":{"canonical_record":{"source":{"id":"2605.12705","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T20:08:00Z","cross_cats_sorted":[],"title_canon_sha256":"6fe29006452258f23bebdc91ad610ebd316c92593f87cbf3ba80d94371051193","abstract_canon_sha256":"e8c0907e9d5fb802a936c88d8ea0be29e73c356d09aaa643f42975db6acbeb5c"},"schema_version":"1.0"},"canonical_sha256":"183af666c41c9280810b492207eb49fefd2f347f6fbf777d118b30fac1521edf","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:09:49.628343Z","signature_b64":"qEKSDc3zDV2/i8ORi2ycnFGBa9mKXbmml//qu+mXiGxDZMK4N3cfxw0nnJGXTTmDWBCgxkZ6RzXeZA7CLVE4Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"183af666c41c9280810b492207eb49fefd2f347f6fbf777d118b30fac1521edf","last_reissued_at":"2026-05-18T03:09:49.627455Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:09:49.627455Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.12705","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:49Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"vBUPkU+VI/XSI7iNVg+RZpC+lRuZyqhMLPHobXY3WNPGaRdyjKhxGpG+hV6NWRRAhyOFqG+Pl8xiAHk7YYbFBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T03:09:39.852340Z"},"content_sha256":"48e3c848d4913bd7b2b0814dce8867c1cb8c6fd2e64dace99a1cd1d618ba3754","schema_version":"1.0","event_id":"sha256:48e3c848d4913bd7b2b0814dce8867c1cb8c6fd2e64dace99a1cd1d618ba3754"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:DA5PMZWEDSJIBAILJERAP22J73","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Early Data Exposure Improves Robustness to Subsequent Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Mixing some target data into pretraining improves retention of that capability after later fine-tuning on new tasks.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aditi Raghunathan, Gaurav R. Ghosal, Jacob Mitchell Springer, Lawrence Feng, Ziqian Zhong","submitted_at":"2026-05-12T20:08:00Z","abstract_excerpt":"How can we train models whose post-trained capabilities survive subsequent fine-tuning? Rather than focusing on downstream interventions to mitigate forgetting of upstream capabilities, we study how upstream training choices - that is, the manner in which a capability is acquired - shape how robustly that capability is retained. We investigate this question in a controlled three-stage language-model pipeline: pretraining, post-training to acquire a target capability, and downstream fine-tuning on a new objective. Across 135M and 1B models, two post-training domains, and two downstream fine-tun"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"early exposure - mixing post-training data into pretraining - consistently improves the frontier between retained upstream performance and downstream performance. In compute-matched experiments, where the target data must be allocated between pretraining and post-training, we find that the optimum lies at neither extreme.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the controlled three-stage pipeline and specific model sizes/tasks used here capture the dynamics that matter in larger-scale, real-world training where data distributions and objectives are more complex and less cleanly separated.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Early mixing of post-training data into pretraining improves retention of acquired capabilities after subsequent fine-tuning in language models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Mixing some target data into pretraining improves retention of that capability after later fine-tuning on new tasks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"49e4a956cd8d03ee3867551d08e960145732db5c04d665042db2fec67d66b0cf"},"source":{"id":"2605.12705","kind":"arxiv","version":1},"verdict":{"id":"5e444599-3348-46fb-8ccf-109ac0cf67b1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:40:28.418904Z","strongest_claim":"early exposure - mixing post-training data into pretraining - consistently improves the frontier between retained upstream performance and downstream performance. In compute-matched experiments, where the target data must be allocated between pretraining and post-training, we find that the optimum lies at neither extreme.","one_line_summary":"Early mixing of post-training data into pretraining improves retention of acquired capabilities after subsequent fine-tuning in language models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the controlled three-stage pipeline and specific model sizes/tasks used here capture the dynamics that matter in larger-scale, real-world training where data distributions and objectives are more complex and less cleanly separated.","pith_extraction_headline":"Mixing some target data into pretraining improves retention of that capability after later fine-tuning on new tasks."},"references":{"count":25,"sample":[{"doi":"","year":null,"title":"SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model","work_id":"8472f581-14d4-40f8-8189-62ed8b470c4e","ref_index":1,"cited_arxiv_id":"2502.02737","is_internal_anchor":true},{"doi":"","year":null,"title":"Louis Bethune, David Grangier, Dan Busbridge, Eleonora Gualdoni, Marco Cuturi, and Pierre Ablin","work_id":"253db970-0a8b-4c85-85b9-b762b6ce3086","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"arXiv preprint arXiv:2502.06042 , year=","work_id":"9617c4ed-1474-439a-9d08-e0612fa954ae","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Lora learns less and forgets less","work_id":"ca00d37c-5297-4fa7-b692-52260eb614f2","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Scaling laws for predicting downstream performance in llms, 2025","work_id":"10184b86-cdee-4047-93c1-53dd0f8c9a18","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":25,"snapshot_sha256":"03edf1716805044acdfc9f8e36e50bd3febc707645515dfd849d5d2aa47dfdc4","internal_anchors":7},"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":"5e444599-3348-46fb-8ccf-109ac0cf67b1"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:49Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"61FLpL9oGaj8rhsWkC/Fezje5Pt9xRoVGR+Ofoz/ytUoD9Sa6nR5uZl+Zc5+ggFtE45JXcvYlvMcL968Y09bDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T03:09:39.853224Z"},"content_sha256":"6cdbf473cb3f68479a1916cf611729c9d8546b08c4b3469cd46b9d61f9a115c3","schema_version":"1.0","event_id":"sha256:6cdbf473cb3f68479a1916cf611729c9d8546b08c4b3469cd46b9d61f9a115c3"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/DA5PMZWEDSJIBAILJERAP22J73/bundle.json","state_url":"https://pith.science/pith/DA5PMZWEDSJIBAILJERAP22J73/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/DA5PMZWEDSJIBAILJERAP22J73/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":"2026-05-28T03:09:39Z","links":{"resolver":"https://pith.science/pith/DA5PMZWEDSJIBAILJERAP22J73","bundle":"https://pith.science/pith/DA5PMZWEDSJIBAILJERAP22J73/bundle.json","state":"https://pith.science/pith/DA5PMZWEDSJIBAILJERAP22J73/state.json","well_known_bundle":"https://pith.science/.well-known/pith/DA5PMZWEDSJIBAILJERAP22J73/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:DA5PMZWEDSJIBAILJERAP22J73","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":"e8c0907e9d5fb802a936c88d8ea0be29e73c356d09aaa643f42975db6acbeb5c","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T20:08:00Z","title_canon_sha256":"6fe29006452258f23bebdc91ad610ebd316c92593f87cbf3ba80d94371051193"},"schema_version":"1.0","source":{"id":"2605.12705","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12705","created_at":"2026-05-18T03:09:49Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12705v1","created_at":"2026-05-18T03:09:49Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12705","created_at":"2026-05-18T03:09:49Z"},{"alias_kind":"pith_short_12","alias_value":"DA5PMZWEDSJI","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"DA5PMZWEDSJIBAIL","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"DA5PMZWE","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:6cdbf473cb3f68479a1916cf611729c9d8546b08c4b3469cd46b9d61f9a115c3","target":"graph","created_at":"2026-05-18T03:09:49Z","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":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"early exposure - mixing post-training data into pretraining - consistently improves the frontier between retained upstream performance and downstream performance. In compute-matched experiments, where the target data must be allocated between pretraining and post-training, we find that the optimum lies at neither extreme."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the controlled three-stage pipeline and specific model sizes/tasks used here capture the dynamics that matter in larger-scale, real-world training where data distributions and objectives are more complex and less cleanly separated."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Early mixing of post-training data into pretraining improves retention of acquired capabilities after subsequent fine-tuning in language models."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Mixing some target data into pretraining improves retention of that capability after later fine-tuning on new tasks."}],"snapshot_sha256":"49e4a956cd8d03ee3867551d08e960145732db5c04d665042db2fec67d66b0cf"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"How can we train models whose post-trained capabilities survive subsequent fine-tuning? Rather than focusing on downstream interventions to mitigate forgetting of upstream capabilities, we study how upstream training choices - that is, the manner in which a capability is acquired - shape how robustly that capability is retained. We investigate this question in a controlled three-stage language-model pipeline: pretraining, post-training to acquire a target capability, and downstream fine-tuning on a new objective. Across 135M and 1B models, two post-training domains, and two downstream fine-tun","authors_text":"Aditi Raghunathan, Gaurav R. Ghosal, Jacob Mitchell Springer, Lawrence Feng, Ziqian Zhong","cross_cats":[],"headline":"Mixing some target data into pretraining improves retention of that capability after later fine-tuning on new tasks.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T20:08:00Z","title":"Early Data Exposure Improves Robustness to Subsequent Fine-Tuning"},"references":{"count":25,"internal_anchors":7,"resolved_work":25,"sample":[{"cited_arxiv_id":"2502.02737","doi":"","is_internal_anchor":true,"ref_index":1,"title":"SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model","work_id":"8472f581-14d4-40f8-8189-62ed8b470c4e","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Louis Bethune, David Grangier, Dan Busbridge, Eleonora Gualdoni, Marco Cuturi, and Pierre Ablin","work_id":"253db970-0a8b-4c85-85b9-b762b6ce3086","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"arXiv preprint arXiv:2502.06042 , year=","work_id":"9617c4ed-1474-439a-9d08-e0612fa954ae","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Lora learns less and forgets less","work_id":"ca00d37c-5297-4fa7-b692-52260eb614f2","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Scaling laws for predicting downstream performance in llms, 2025","work_id":"10184b86-cdee-4047-93c1-53dd0f8c9a18","year":null}],"snapshot_sha256":"03edf1716805044acdfc9f8e36e50bd3febc707645515dfd849d5d2aa47dfdc4"},"source":{"id":"2605.12705","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:40:28.418904Z","id":"5e444599-3348-46fb-8ccf-109ac0cf67b1","model_set":{"reader":"grok-4.3"},"one_line_summary":"Early mixing of post-training data into pretraining improves retention of acquired capabilities after subsequent fine-tuning in language models.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Mixing some target data into pretraining improves retention of that capability after later fine-tuning on new tasks.","strongest_claim":"early exposure - mixing post-training data into pretraining - consistently improves the frontier between retained upstream performance and downstream performance. In compute-matched experiments, where the target data must be allocated between pretraining and post-training, we find that the optimum lies at neither extreme.","weakest_assumption":"That the controlled three-stage pipeline and specific model sizes/tasks used here capture the dynamics that matter in larger-scale, real-world training where data distributions and objectives are more complex and less cleanly separated."}},"verdict_id":"5e444599-3348-46fb-8ccf-109ac0cf67b1"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:48e3c848d4913bd7b2b0814dce8867c1cb8c6fd2e64dace99a1cd1d618ba3754","target":"record","created_at":"2026-05-18T03:09:49Z","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":"e8c0907e9d5fb802a936c88d8ea0be29e73c356d09aaa643f42975db6acbeb5c","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-12T20:08:00Z","title_canon_sha256":"6fe29006452258f23bebdc91ad610ebd316c92593f87cbf3ba80d94371051193"},"schema_version":"1.0","source":{"id":"2605.12705","kind":"arxiv","version":1}},"canonical_sha256":"183af666c41c9280810b492207eb49fefd2f347f6fbf777d118b30fac1521edf","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"183af666c41c9280810b492207eb49fefd2f347f6fbf777d118b30fac1521edf","first_computed_at":"2026-05-18T03:09:49.627455Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:09:49.627455Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"qEKSDc3zDV2/i8ORi2ycnFGBa9mKXbmml//qu+mXiGxDZMK4N3cfxw0nnJGXTTmDWBCgxkZ6RzXeZA7CLVE4Aw==","signature_status":"signed_v1","signed_at":"2026-05-18T03:09:49.628343Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.12705","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:48e3c848d4913bd7b2b0814dce8867c1cb8c6fd2e64dace99a1cd1d618ba3754","sha256:6cdbf473cb3f68479a1916cf611729c9d8546b08c4b3469cd46b9d61f9a115c3"],"state_sha256":"5292fb205244e1d11a00973fa86dba62e715edbe38b2505096aafa559a4cd4f8"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"+sWAqNWkEi5isjuu7Ltjs9uyFb0x6i3PtrfnwLJ3qn4Y/izzl0PuUhsRxy655j8/uu9H08xYdi/yLF/wV8UJDA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-28T03:09:39.856664Z","bundle_sha256":"3860b8a37b53a4e11f37984079e69cf4d325e173ca991d703a442e83eb005be8"}}