{"paper":{"title":"SMolLM: Small Language Models Learn Small Molecular Grammar","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A 53K-parameter transformer generates valid SMILES by resolving constraints in fixed order: brackets first, rings second, valence last.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Akhil Jindal, Harang Ju","submitted_at":"2026-05-07T14:21:26Z","abstract_excerpt":"Language models for molecular design have scaled to hundreds of millions of parameters, yet how they learn chemical grammar is poorly understood. We train SMolLM, a 53K-parameter weight-shared transformer, to generate novel SMILES with 95% validity on the ZINC-250K drug-like-molecule benchmark, outperforming a standard GPT with 10 times more parameters. Mechanistically, the same block resolves SMILES constraints across passes in a fixed hierarchy: brackets first, rings second, and valence last, as shown by error classification and linear probing, with ablation isolating the bracket-matching he"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"the same block resolves SMILES constraints across passes in a fixed order: brackets first, rings second, and valence last, as shown by error classification, linear probing, and sparse autoencoders. A systematic ablation across attention heads and passes further localizes the first bracket-matching step to a single attention head.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That linear probing, sparse autoencoders, and error classification reveal the actual causal computation rather than surface correlations, and that high validity on the benchmark reflects genuine grammar learning instead of dataset-specific pattern matching.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A 53K-parameter model generates 95% valid SMILES on ZINC-250K, outperforming larger models, by resolving chemical constraints in fixed order: brackets first, rings second, valence last.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A 53K-parameter transformer generates valid SMILES by resolving constraints in fixed order: brackets first, rings second, valence last.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c0a7ee730d16ebfc58f4f18a5e849258bcb6dc337f3528e1d23a5e1735c7890d"},"source":{"id":"2605.06322","kind":"arxiv","version":2},"verdict":{"id":"220f2527-1e8b-4f36-ae96-7aea40996fb5","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T12:54:00.434191Z","strongest_claim":"the same block resolves SMILES constraints across passes in a fixed order: brackets first, rings second, and valence last, as shown by error classification, linear probing, and sparse autoencoders. A systematic ablation across attention heads and passes further localizes the first bracket-matching step to a single attention head.","one_line_summary":"A 53K-parameter model generates 95% valid SMILES on ZINC-250K, outperforming larger models, by resolving chemical constraints in fixed order: brackets first, rings second, valence last.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That linear probing, sparse autoencoders, and error classification reveal the actual causal computation rather than surface correlations, and that high validity on the benchmark reflects genuine grammar learning instead of dataset-specific pattern matching.","pith_extraction_headline":"A 53K-parameter transformer generates valid SMILES by resolving constraints in fixed order: brackets first, rings second, valence last."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.06322/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T12:42:04.063215Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-20T08:33:30.496352Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T18:31:19.444859Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T12:47:06.925547Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"12bf0dab27bfc5e2c9a9bb4e2bdbe8d642f55fd11e61ed01bfa42f476e93c986"},"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"}