{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:CXM4YKODNS3MZAN7QKPIWWE4CX","short_pith_number":"pith:CXM4YKOD","schema_version":"1.0","canonical_sha256":"15d9cc29c36cb6cc81bf829e8b589c15f41a7d0a2fba28832aa52dd38c836441","source":{"kind":"arxiv","id":"2403.03870","version":2},"attestation_state":"computed","paper":{"title":"Learning to Decode Collaboratively with Multiple Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bailin Wang, David Sontag, Hunter Lang, Shannon Zejiang Shen, Yoon Kim","submitted_at":"2024-03-06T17:23:28Z","abstract_excerpt":"We propose a method to teach multiple large language models (LLM) to collaborate by interleaving their generations at the token level. We model the decision of which LLM generates the next token as a latent variable. By optimizing the marginal likelihood of a training set under our latent variable model, the base LLM automatically learns when to generate itself and when to call on one of the ``assistant'' language models to generate, all without direct supervision. Token-level collaboration during decoding allows for a fusion of each model's expertise in a manner tailored to the specific task "},"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":"2403.03870","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-03-06T17:23:28Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"8426dc542cb59d7db65863a67ae114ab84972ff7c3adcbaf38f50744e19a3d29","abstract_canon_sha256":"2822f14201ce6a383e376819e362b3b246e7c39b82b8a897af099c5a3e6364e5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:59:26.490116Z","signature_b64":"SAr4oCHrAgagzeUgYhqqPHeAHdkGyxSTCd455kRU4sZFyHfSdgn/k3Zi7x+KkIucD35WgsgY75JnS/J+oThUDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"15d9cc29c36cb6cc81bf829e8b589c15f41a7d0a2fba28832aa52dd38c836441","last_reissued_at":"2026-07-05T08:59:26.489589Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:59:26.489589Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning to Decode Collaboratively with Multiple Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bailin Wang, David Sontag, Hunter Lang, Shannon Zejiang Shen, Yoon Kim","submitted_at":"2024-03-06T17:23:28Z","abstract_excerpt":"We propose a method to teach multiple large language models (LLM) to collaborate by interleaving their generations at the token level. We model the decision of which LLM generates the next token as a latent variable. By optimizing the marginal likelihood of a training set under our latent variable model, the base LLM automatically learns when to generate itself and when to call on one of the ``assistant'' language models to generate, all without direct supervision. Token-level collaboration during decoding allows for a fusion of each model's expertise in a manner tailored to the specific task "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.03870","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/2403.03870/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":"2403.03870","created_at":"2026-07-05T08:59:26.489653+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.03870v2","created_at":"2026-07-05T08:59:26.489653+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.03870","created_at":"2026-07-05T08:59:26.489653+00:00"},{"alias_kind":"pith_short_12","alias_value":"CXM4YKODNS3M","created_at":"2026-07-05T08:59:26.489653+00:00"},{"alias_kind":"pith_short_16","alias_value":"CXM4YKODNS3MZAN7","created_at":"2026-07-05T08:59:26.489653+00:00"},{"alias_kind":"pith_short_8","alias_value":"CXM4YKOD","created_at":"2026-07-05T08:59:26.489653+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2410.13181","citing_title":"AdaSwitch: Adaptive Switching between Small and Large Agents for Effective Cloud-Local Collaborative Learning","ref_index":20,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CXM4YKODNS3MZAN7QKPIWWE4CX","json":"https://pith.science/pith/CXM4YKODNS3MZAN7QKPIWWE4CX.json","graph_json":"https://pith.science/api/pith-number/CXM4YKODNS3MZAN7QKPIWWE4CX/graph.json","events_json":"https://pith.science/api/pith-number/CXM4YKODNS3MZAN7QKPIWWE4CX/events.json","paper":"https://pith.science/paper/CXM4YKOD"},"agent_actions":{"view_html":"https://pith.science/pith/CXM4YKODNS3MZAN7QKPIWWE4CX","download_json":"https://pith.science/pith/CXM4YKODNS3MZAN7QKPIWWE4CX.json","view_paper":"https://pith.science/paper/CXM4YKOD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.03870&json=true","fetch_graph":"https://pith.science/api/pith-number/CXM4YKODNS3MZAN7QKPIWWE4CX/graph.json","fetch_events":"https://pith.science/api/pith-number/CXM4YKODNS3MZAN7QKPIWWE4CX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CXM4YKODNS3MZAN7QKPIWWE4CX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CXM4YKODNS3MZAN7QKPIWWE4CX/action/storage_attestation","attest_author":"https://pith.science/pith/CXM4YKODNS3MZAN7QKPIWWE4CX/action/author_attestation","sign_citation":"https://pith.science/pith/CXM4YKODNS3MZAN7QKPIWWE4CX/action/citation_signature","submit_replication":"https://pith.science/pith/CXM4YKODNS3MZAN7QKPIWWE4CX/action/replication_record"}},"created_at":"2026-07-05T08:59:26.489653+00:00","updated_at":"2026-07-05T08:59:26.489653+00:00"}