{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:4OFVLZZLOVAQPHVNJWKHMIRTAT","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":"90d565358fe1850a6c46e9f614103872ce80bb47f4c6d8f3850473d21558d259","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.IR","submitted_at":"2026-05-12T18:05:55Z","title_canon_sha256":"805391eadae9bcf536e9256d7da1afd222e74a5616b7d54437935f8d2427f49c"},"schema_version":"1.0","source":{"id":"2605.12617","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12617","created_at":"2026-05-18T03:10:00Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12617v1","created_at":"2026-05-18T03:10:00Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12617","created_at":"2026-05-18T03:10:00Z"},{"alias_kind":"pith_short_12","alias_value":"4OFVLZZLOVAQ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"4OFVLZZLOVAQPHVN","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"4OFVLZZL","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:65ec57d4ef09624c1661b00e3a4b477fa44c35fa841765becddff2ba120870a5","target":"graph","created_at":"2026-05-18T03:10:00Z","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":"Extensive experiments demonstrate that SID-MLP matches the accuracy of teacher models while accelerating inference by 8.74x. This distillation strategy can serve as a plug-and-play accelerator for different backbones and tokenizer settings."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The hierarchical nature of SIDs makes prediction difficulty drop sharply after the first token, rendering repeated attention computations highly redundant."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"SID-MLP distills autoregressive generative recommenders into efficient position-specific MLP heads for Semantic ID tasks, achieving 8.74x faster inference with matching accuracy."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Distilling generative recommenders into MLPs preserves accuracy while speeding up inference by 8.74x"}],"snapshot_sha256":"07bdc74cc6460efc6a67511efaa3b45150cdb3a996393ffa80da030f0551d571"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"45ceee976e77c80dc467ccff50a920c0c08e893e6405a463602a79e5c62d41ea"},"paper":{"abstract_excerpt":"Generative recommendation models employing Semantic IDs (SIDs) exhibit strong potential, yet their practical deployment is bottlenecked by the high inference latency of beam-expanded autoregressive decoding. In this work, we identify that standard attention-heavy Transformer decoders represent a structural overkill for this task: the hierarchical nature of SIDs makes prediction difficulty drops sharply after the first token, rendering repeated attention computations highly redundant. Driven by this insight, we propose SID-MLP, a lightweight MLP-centric distillation framework that fundamentally","authors_text":"Clark Mingxuan Ju, Julian McAuley, Neil Shah, Yupeng Hou, Zitian Guo","cross_cats":[],"headline":"Distilling generative recommenders into MLPs preserves accuracy while speeding up inference by 8.74x","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.IR","submitted_at":"2026-05-12T18:05:55Z","title":"MLPs are Efficient Distilled Generative Recommenders"},"references":{"count":77,"internal_anchors":5,"resolved_work":77,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Transformer memory as a differentiable search index","work_id":"86233c16-b959-4114-9574-a43c62cdee35","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Tran, Jonah Samost, Maciej Kula, Ed H","work_id":"ba5c0f0c-d332-440a-93c5-8606c39de2fc","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Adapting large language models by integrating collaborative semantics for recommenda- tion","work_id":"f51fe37c-a7aa-419f-a196-bf60fb106255","year":2024},{"cited_arxiv_id":"2502.18965","doi":"","is_internal_anchor":true,"ref_index":4,"title":"OneRec: Unifying Retrieve and Rank with Generative Recommender and Iterative Preference Alignment","work_id":"d1a07d92-e045-4af2-a79f-c7b0112cf824","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Session-based recommendations with recurrent neural networks","work_id":"52fdec4e-20e0-4895-a662-0d3deb4fbb2c","year":2016}],"snapshot_sha256":"95ab4ab1a560bd6719e2e877be70c7bae451ba16f2f3c5a52fad1cff53927393"},"source":{"id":"2605.12617","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:10:41.001869Z","id":"c375d6d3-cc80-41db-bcf6-3aa50bdcf746","model_set":{"reader":"grok-4.3"},"one_line_summary":"SID-MLP distills autoregressive generative recommenders into efficient position-specific MLP heads for Semantic ID tasks, achieving 8.74x faster inference with matching accuracy.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Distilling generative recommenders into MLPs preserves accuracy while speeding up inference by 8.74x","strongest_claim":"Extensive experiments demonstrate that SID-MLP matches the accuracy of teacher models while accelerating inference by 8.74x. This distillation strategy can serve as a plug-and-play accelerator for different backbones and tokenizer settings.","weakest_assumption":"The hierarchical nature of SIDs makes prediction difficulty drop sharply after the first token, rendering repeated attention computations highly redundant."}},"verdict_id":"c375d6d3-cc80-41db-bcf6-3aa50bdcf746"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:56924016b237d66991800c6115fd4ad839ad1a7b5c5a6d5a2695db91209760ae","target":"record","created_at":"2026-05-18T03:10:00Z","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":"90d565358fe1850a6c46e9f614103872ce80bb47f4c6d8f3850473d21558d259","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.IR","submitted_at":"2026-05-12T18:05:55Z","title_canon_sha256":"805391eadae9bcf536e9256d7da1afd222e74a5616b7d54437935f8d2427f49c"},"schema_version":"1.0","source":{"id":"2605.12617","kind":"arxiv","version":1}},"canonical_sha256":"e38b55e72b7541079ead4d9476223304c9ee1416cd7911867544650c80d1d435","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"e38b55e72b7541079ead4d9476223304c9ee1416cd7911867544650c80d1d435","first_computed_at":"2026-05-18T03:10:00.495604Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:10:00.495604Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"DI8NoU1FwK/mHPBUYoYDiHtZpsQYHU8GB3JZ7nET2voOZjnnisrRW2pUPZe9HOVQt9RcRjxgWI5A2jUtovSgCQ==","signature_status":"signed_v1","signed_at":"2026-05-18T03:10:00.496445Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.12617","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:56924016b237d66991800c6115fd4ad839ad1a7b5c5a6d5a2695db91209760ae","sha256:65ec57d4ef09624c1661b00e3a4b477fa44c35fa841765becddff2ba120870a5"],"state_sha256":"389a184d4c2cfa5e396520213c70c82cd0de1ba9329b2404a9736906b3521724"}