{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:YF6H7X47LYDRIUZRYUEIT4YCNQ","short_pith_number":"pith:YF6H7X47","canonical_record":{"source":{"id":"2602.11618","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-12T06:14:34Z","cross_cats_sorted":["q-bio.QM"],"title_canon_sha256":"51df8ba47e3cab93408e9849060252bde98832c80bfd97c054fec22b4e4baefa","abstract_canon_sha256":"453216340a058ee8fad0d5a80c53395ccace2e0f5523edaea5845ee9bc11fd9b"},"schema_version":"1.0"},"canonical_sha256":"c17c7fdf9f5e07145331c50889f3026c26f261e12db3e34d8d31aa895e88852e","source":{"kind":"arxiv","id":"2602.11618","version":4},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.11618","created_at":"2026-05-18T03:09:23Z"},{"alias_kind":"arxiv_version","alias_value":"2602.11618v4","created_at":"2026-05-18T03:09:23Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.11618","created_at":"2026-05-18T03:09:23Z"},{"alias_kind":"pith_short_12","alias_value":"YF6H7X47LYDR","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"YF6H7X47LYDRIUZR","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"YF6H7X47","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:YF6H7X47LYDRIUZRYUEIT4YCNQ","target":"record","payload":{"canonical_record":{"source":{"id":"2602.11618","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-12T06:14:34Z","cross_cats_sorted":["q-bio.QM"],"title_canon_sha256":"51df8ba47e3cab93408e9849060252bde98832c80bfd97c054fec22b4e4baefa","abstract_canon_sha256":"453216340a058ee8fad0d5a80c53395ccace2e0f5523edaea5845ee9bc11fd9b"},"schema_version":"1.0"},"canonical_sha256":"c17c7fdf9f5e07145331c50889f3026c26f261e12db3e34d8d31aa895e88852e","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:09:23.631129Z","signature_b64":"Zyi2XIrkHPkxdjqHPJ27nSEqbpQgGW3W93XJx/YGOUIfu7IH1tMr2r0yvyiZo0yyXmn15aQCH7avjTaBa8H7Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c17c7fdf9f5e07145331c50889f3026c26f261e12db3e34d8d31aa895e88852e","last_reissued_at":"2026-05-18T03:09:23.630097Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:09:23.630097Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2602.11618","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":"2026-05-18T03:09:23Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"fQjYPQ33ZLtGFAmPLcXGCpNSt/CZ5x+KyxdmX6YFnrAwXGLQp0waL7DI/Dvm9xc+6wpzE9f1xAxmqYxhjXxVBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-05T05:58:02.182368Z"},"content_sha256":"788424b1ec7fff9cd8d93c67b0f20b07a3a1b716caf704204f0ee1245737bb5d","schema_version":"1.0","event_id":"sha256:788424b1ec7fff9cd8d93c67b0f20b07a3a1b716caf704204f0ee1245737bb5d"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:YF6H7X47LYDRIUZRYUEIT4YCNQ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"How Well Do Large-Scale Chemical Language Models Transfer to Downstream Tasks?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Scaling chemical language models reduces pretraining loss but delivers limited gains on downstream molecular tasks.","cross_cats":["q-bio.QM"],"primary_cat":"cs.LG","authors_text":"Ryosuke Kojima, Tatsuya Sagawa","submitted_at":"2026-02-12T06:14:34Z","abstract_excerpt":"Chemical Language Models (CLMs) pre-trained on large scale molecular data are widely used for molecular property prediction. However, the common belief that increasing training resources such as model size, dataset size, and training compute improves both pretraining loss and downstream task performance has not been systematically validated in the chemical domain. In this work, we evaluate this assumption by pretraining CLMs while scaling training resources and measuring transfer performance across diverse molecular property prediction (MPP) tasks. We find that while pretraining loss consisten"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"while pretraining loss consistently decreases with increased training resources, downstream task performance shows limited improvement","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the chosen downstream molecular property prediction tasks and evaluation protocol are representative enough that limited observed gains reflect a general scaling failure rather than task-specific or experimental artifacts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Scaling chemical language models reduces pretraining loss but yields only limited or saturating gains on downstream molecular property prediction tasks, exposing a disconnect between pretraining metrics and actual transfer performance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Scaling chemical language models reduces pretraining loss but delivers limited gains on downstream molecular tasks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e7f48dd40117263f661d684c2f6ae12da7e9d1b0faa216a64a60053a25091598"},"source":{"id":"2602.11618","kind":"arxiv","version":4},"verdict":{"id":"8663b6f7-d280-4fa6-bd44-e1304f04d88a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T02:43:18.191007Z","strongest_claim":"while pretraining loss consistently decreases with increased training resources, downstream task performance shows limited improvement","one_line_summary":"Scaling chemical language models reduces pretraining loss but yields only limited or saturating gains on downstream molecular property prediction tasks, exposing a disconnect between pretraining metrics and actual transfer performance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the chosen downstream molecular property prediction tasks and evaluation protocol are representative enough that limited observed gains reflect a general scaling failure rather than task-specific or experimental artifacts.","pith_extraction_headline":"Scaling chemical language models reduces pretraining loss but delivers limited gains on downstream molecular tasks."},"references":{"count":32,"sample":[{"doi":"","year":2017,"title":"Pappu, and Vijay Pande","work_id":"a72fff8f-c90d-4557-95a1-022243633dac","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"ChemBERTa: Large - scale self -supervised pretraining for molecular property prediction","work_id":"f24a680c-0d43-4fe9-b0c3-e11d238a37a6","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding","work_id":"59c38c1b-7b9d-453e-8046-13bbd54c753f","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"UniCorn: A unified contrastive learning approach for multi-view molecular representation learning","work_id":"779b1959-5c04-476c-86de-a28e2b4890d5","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Frey, Ryan Soklaski, Simon Ax- elrod, Siddharth Samsi, Rafael Gómez -Bombarelli, Con- nor W","work_id":"74d6bbe4-3a65-4434-a9db-bb3da2453dec","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":32,"snapshot_sha256":"0c9cdf3eafbdf216ff84e689fd7d2db9254cfa1f61b07d82271b6801685dbd1e","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":"8663b6f7-d280-4fa6-bd44-e1304f04d88a"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:09:23Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"SKnC4nDFUnPnldNQvmwHMWbttXz+GuflUxoLid7BCG8dwAGavFZ/PJ6ZpIn4NNchlGQmaDDwbrfcertUPxOtCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-05T05:58:02.182910Z"},"content_sha256":"f8c0be501487bfd86df16452a9b734b33080f05a53d6b13876e2be975502475f","schema_version":"1.0","event_id":"sha256:f8c0be501487bfd86df16452a9b734b33080f05a53d6b13876e2be975502475f"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/YF6H7X47LYDRIUZRYUEIT4YCNQ/bundle.json","state_url":"https://pith.science/pith/YF6H7X47LYDRIUZRYUEIT4YCNQ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/YF6H7X47LYDRIUZRYUEIT4YCNQ/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":"2026-06-05T05:58:02Z","links":{"resolver":"https://pith.science/pith/YF6H7X47LYDRIUZRYUEIT4YCNQ","bundle":"https://pith.science/pith/YF6H7X47LYDRIUZRYUEIT4YCNQ/bundle.json","state":"https://pith.science/pith/YF6H7X47LYDRIUZRYUEIT4YCNQ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/YF6H7X47LYDRIUZRYUEIT4YCNQ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:YF6H7X47LYDRIUZRYUEIT4YCNQ","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":"453216340a058ee8fad0d5a80c53395ccace2e0f5523edaea5845ee9bc11fd9b","cross_cats_sorted":["q-bio.QM"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-12T06:14:34Z","title_canon_sha256":"51df8ba47e3cab93408e9849060252bde98832c80bfd97c054fec22b4e4baefa"},"schema_version":"1.0","source":{"id":"2602.11618","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.11618","created_at":"2026-05-18T03:09:23Z"},{"alias_kind":"arxiv_version","alias_value":"2602.11618v4","created_at":"2026-05-18T03:09:23Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.11618","created_at":"2026-05-18T03:09:23Z"},{"alias_kind":"pith_short_12","alias_value":"YF6H7X47LYDR","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"YF6H7X47LYDRIUZR","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"YF6H7X47","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:f8c0be501487bfd86df16452a9b734b33080f05a53d6b13876e2be975502475f","target":"graph","created_at":"2026-05-18T03:09:23Z","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":"while pretraining loss consistently decreases with increased training resources, downstream task performance shows limited improvement"},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the chosen downstream molecular property prediction tasks and evaluation protocol are representative enough that limited observed gains reflect a general scaling failure rather than task-specific or experimental artifacts."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Scaling chemical language models reduces pretraining loss but yields only limited or saturating gains on downstream molecular property prediction tasks, exposing a disconnect between pretraining metrics and actual transfer performance."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Scaling chemical language models reduces pretraining loss but delivers limited gains on downstream molecular tasks."}],"snapshot_sha256":"e7f48dd40117263f661d684c2f6ae12da7e9d1b0faa216a64a60053a25091598"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Chemical Language Models (CLMs) pre-trained on large scale molecular data are widely used for molecular property prediction. However, the common belief that increasing training resources such as model size, dataset size, and training compute improves both pretraining loss and downstream task performance has not been systematically validated in the chemical domain. In this work, we evaluate this assumption by pretraining CLMs while scaling training resources and measuring transfer performance across diverse molecular property prediction (MPP) tasks. We find that while pretraining loss consisten","authors_text":"Ryosuke Kojima, Tatsuya Sagawa","cross_cats":["q-bio.QM"],"headline":"Scaling chemical language models reduces pretraining loss but delivers limited gains on downstream molecular tasks.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-12T06:14:34Z","title":"How Well Do Large-Scale Chemical Language Models Transfer to Downstream Tasks?"},"references":{"count":32,"internal_anchors":0,"resolved_work":32,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Pappu, and Vijay Pande","work_id":"a72fff8f-c90d-4557-95a1-022243633dac","year":2017},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"ChemBERTa: Large - scale self -supervised pretraining for molecular property prediction","work_id":"f24a680c-0d43-4fe9-b0c3-e11d238a37a6","year":2020},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding","work_id":"59c38c1b-7b9d-453e-8046-13bbd54c753f","year":2019},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"UniCorn: A unified contrastive learning approach for multi-view molecular representation learning","work_id":"779b1959-5c04-476c-86de-a28e2b4890d5","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Frey, Ryan Soklaski, Simon Ax- elrod, Siddharth Samsi, Rafael Gómez -Bombarelli, Con- nor W","work_id":"74d6bbe4-3a65-4434-a9db-bb3da2453dec","year":2023}],"snapshot_sha256":"0c9cdf3eafbdf216ff84e689fd7d2db9254cfa1f61b07d82271b6801685dbd1e"},"source":{"id":"2602.11618","kind":"arxiv","version":4},"verdict":{"created_at":"2026-05-16T02:43:18.191007Z","id":"8663b6f7-d280-4fa6-bd44-e1304f04d88a","model_set":{"reader":"grok-4.3"},"one_line_summary":"Scaling chemical language models reduces pretraining loss but yields only limited or saturating gains on downstream molecular property prediction tasks, exposing a disconnect between pretraining metrics and actual transfer performance.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Scaling chemical language models reduces pretraining loss but delivers limited gains on downstream molecular tasks.","strongest_claim":"while pretraining loss consistently decreases with increased training resources, downstream task performance shows limited improvement","weakest_assumption":"That the chosen downstream molecular property prediction tasks and evaluation protocol are representative enough that limited observed gains reflect a general scaling failure rather than task-specific or experimental artifacts."}},"verdict_id":"8663b6f7-d280-4fa6-bd44-e1304f04d88a"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:788424b1ec7fff9cd8d93c67b0f20b07a3a1b716caf704204f0ee1245737bb5d","target":"record","created_at":"2026-05-18T03:09:23Z","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":"453216340a058ee8fad0d5a80c53395ccace2e0f5523edaea5845ee9bc11fd9b","cross_cats_sorted":["q-bio.QM"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-12T06:14:34Z","title_canon_sha256":"51df8ba47e3cab93408e9849060252bde98832c80bfd97c054fec22b4e4baefa"},"schema_version":"1.0","source":{"id":"2602.11618","kind":"arxiv","version":4}},"canonical_sha256":"c17c7fdf9f5e07145331c50889f3026c26f261e12db3e34d8d31aa895e88852e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"c17c7fdf9f5e07145331c50889f3026c26f261e12db3e34d8d31aa895e88852e","first_computed_at":"2026-05-18T03:09:23.630097Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:09:23.630097Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Zyi2XIrkHPkxdjqHPJ27nSEqbpQgGW3W93XJx/YGOUIfu7IH1tMr2r0yvyiZo0yyXmn15aQCH7avjTaBa8H7Bw==","signature_status":"signed_v1","signed_at":"2026-05-18T03:09:23.631129Z","signed_message":"canonical_sha256_bytes"},"source_id":"2602.11618","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:788424b1ec7fff9cd8d93c67b0f20b07a3a1b716caf704204f0ee1245737bb5d","sha256:f8c0be501487bfd86df16452a9b734b33080f05a53d6b13876e2be975502475f"],"state_sha256":"1454a09fb5841d043c1b10e9efeae316bd485538f6f2d43fedb3d5fa8b25a82b"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"numUbu72P3guO4Se7mqdsmvO5h9wwrA1gNxXc8lvA0bJSiajWwUaGo00DsbdSPGv/W6VbUAAuHUTuaI2N76fBw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-05T05:58:02.185635Z","bundle_sha256":"d1efdb3116d9241324af8cba2f9fbecbdecb93006f4f28505c7a86db06947f81"}}