{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:PTPA5FNC4VIHJWV352KCMGIP5C","short_pith_number":"pith:PTPA5FNC","schema_version":"1.0","canonical_sha256":"7cde0e95a2e55074dabbee9426190fe885c6406e7f6fa9a404a68273d9e1a8f2","source":{"kind":"arxiv","id":"2310.04363","version":2},"attestation_state":"computed","paper":{"title":"Amortizing intractable inference in large language models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Edward J. Hu, Eric Elmoznino, Guillaume Lajoie, Moksh Jain, Nikolay Malkin, Yoshua Bengio, Younesse Kaddar","submitted_at":"2023-10-06T16:36:08Z","abstract_excerpt":"Autoregressive large language models (LLMs) compress knowledge from their training data through next-token conditional distributions. This limits tractable querying of this knowledge to start-to-end autoregressive sampling. However, many tasks of interest -- including sequence continuation, infilling, and other forms of constrained generation -- involve sampling from intractable posterior distributions. We address this limitation by using amortized Bayesian inference to sample from these intractable posteriors. Such amortization is algorithmically achieved by fine-tuning LLMs via diversity-see"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2310.04363","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-10-06T16:36:08Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"b441b444820e4fede10ce628345c21c19f1b0d310558bda99fa4cd61a98a3dda","abstract_canon_sha256":"1da1e6f1dd1ee0077fc7b672939c22f9f02c6377ce5b2f0c52fab53471872b42"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:55:45.298878Z","signature_b64":"DjuMu4IxJBu69jxaHJ831OcIoKaCrMI8ge75wRPmSfnKybqIa7OoFmlQd3yikJr1R1wlGholDF4sC3LhSc41BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7cde0e95a2e55074dabbee9426190fe885c6406e7f6fa9a404a68273d9e1a8f2","last_reissued_at":"2026-07-05T07:55:45.298369Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:55:45.298369Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Amortizing intractable inference in large language models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Edward J. Hu, Eric Elmoznino, Guillaume Lajoie, Moksh Jain, Nikolay Malkin, Yoshua Bengio, Younesse Kaddar","submitted_at":"2023-10-06T16:36:08Z","abstract_excerpt":"Autoregressive large language models (LLMs) compress knowledge from their training data through next-token conditional distributions. This limits tractable querying of this knowledge to start-to-end autoregressive sampling. However, many tasks of interest -- including sequence continuation, infilling, and other forms of constrained generation -- involve sampling from intractable posterior distributions. We address this limitation by using amortized Bayesian inference to sample from these intractable posteriors. Such amortization is algorithmically achieved by fine-tuning LLMs via diversity-see"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.04363","kind":"arxiv","version":2},"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/2310.04363/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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2310.04363","created_at":"2026-07-05T07:55:45.298433+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.04363v2","created_at":"2026-07-05T07:55:45.298433+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.04363","created_at":"2026-07-05T07:55:45.298433+00:00"},{"alias_kind":"pith_short_12","alias_value":"PTPA5FNC4VIH","created_at":"2026-07-05T07:55:45.298433+00:00"},{"alias_kind":"pith_short_16","alias_value":"PTPA5FNC4VIHJWV3","created_at":"2026-07-05T07:55:45.298433+00:00"},{"alias_kind":"pith_short_8","alias_value":"PTPA5FNC","created_at":"2026-07-05T07:55:45.298433+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":8,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06145","citing_title":"Prompting Complexity: Shortest Prompts for Texts and Behaviors in LLMs","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2607.00461","citing_title":"Multimodal Continuous Reasoning via Asymmetric Mutual Variational Learning","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30844","citing_title":"Fine-Tuning Improves Information Conveyance in Language Models","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19594","citing_title":"Unsupervised Causal Abstractions Discovery","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17295","citing_title":"DISA: Offline Importance Sampling for Distribution-Matching LLM-RL","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10810","citing_title":"Likelihood scoring for continuations of mathematical text: a self-supervised benchmark with tests for shortcut vulnerabilities","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20707","citing_title":"Generative Flow Networks for Model Adaptation in Digital Twins of Natural Systems","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16259","citing_title":"Beyond Distribution Sharpening: The Importance of Task Rewards","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PTPA5FNC4VIHJWV352KCMGIP5C","json":"https://pith.science/pith/PTPA5FNC4VIHJWV352KCMGIP5C.json","graph_json":"https://pith.science/api/pith-number/PTPA5FNC4VIHJWV352KCMGIP5C/graph.json","events_json":"https://pith.science/api/pith-number/PTPA5FNC4VIHJWV352KCMGIP5C/events.json","paper":"https://pith.science/paper/PTPA5FNC"},"agent_actions":{"view_html":"https://pith.science/pith/PTPA5FNC4VIHJWV352KCMGIP5C","download_json":"https://pith.science/pith/PTPA5FNC4VIHJWV352KCMGIP5C.json","view_paper":"https://pith.science/paper/PTPA5FNC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.04363&json=true","fetch_graph":"https://pith.science/api/pith-number/PTPA5FNC4VIHJWV352KCMGIP5C/graph.json","fetch_events":"https://pith.science/api/pith-number/PTPA5FNC4VIHJWV352KCMGIP5C/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PTPA5FNC4VIHJWV352KCMGIP5C/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PTPA5FNC4VIHJWV352KCMGIP5C/action/storage_attestation","attest_author":"https://pith.science/pith/PTPA5FNC4VIHJWV352KCMGIP5C/action/author_attestation","sign_citation":"https://pith.science/pith/PTPA5FNC4VIHJWV352KCMGIP5C/action/citation_signature","submit_replication":"https://pith.science/pith/PTPA5FNC4VIHJWV352KCMGIP5C/action/replication_record"}},"created_at":"2026-07-05T07:55:45.298433+00:00","updated_at":"2026-07-05T07:55:45.298433+00:00"}