{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:5SQVB75ABEDGFKZMBGKB73CMPF","short_pith_number":"pith:5SQVB75A","schema_version":"1.0","canonical_sha256":"eca150ffa0090662ab2c09941fec4c7978a2b13b91edde4e5bbe6368404895d3","source":{"kind":"arxiv","id":"2504.07247","version":2},"attestation_state":"computed","paper":{"title":"Resource-efficient Inference with Foundation Model Programs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chris Jermaine, Kevin Yu, Lunyiu Nie, Marco Cheung, Swarat Chaudhuri, Zhimin Ding","submitted_at":"2025-04-09T19:36:47Z","abstract_excerpt":"The inference-time resource costs of large language and vision models present a growing challenge in production deployments. We propose the use of foundation model programs, i.e., programs that can invoke foundation models with varying resource costs and performance, as an approach to this problem. Specifically, we present a method that translates a task into a program, then learns a policy for resource allocation that, on each input, selects foundation model \"backends\" for each program module. The policy uses smaller, cheaper backends to handle simpler subtasks, while allowing more complex su"},"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":"2504.07247","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-04-09T19:36:47Z","cross_cats_sorted":[],"title_canon_sha256":"54d80a539ba435f9f136c54109506d70f0b765221c43713c9e2eae26d74fb1d7","abstract_canon_sha256":"e06d098e78521258c8a821d212de90c8fc3f6c3b3d80c8a46747b3e3af82388c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:51:25.182531Z","signature_b64":"51MCyjHQzzGLctYmT2dcGnbZ+I1v/LojQP1/j/eZmR6GyaSpOuT5LxJQPMEra20YGlTTrUk4O++rkG4jsXkKBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"eca150ffa0090662ab2c09941fec4c7978a2b13b91edde4e5bbe6368404895d3","last_reissued_at":"2026-07-05T11:51:25.182006Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:51:25.182006Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Resource-efficient Inference with Foundation Model Programs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chris Jermaine, Kevin Yu, Lunyiu Nie, Marco Cheung, Swarat Chaudhuri, Zhimin Ding","submitted_at":"2025-04-09T19:36:47Z","abstract_excerpt":"The inference-time resource costs of large language and vision models present a growing challenge in production deployments. We propose the use of foundation model programs, i.e., programs that can invoke foundation models with varying resource costs and performance, as an approach to this problem. Specifically, we present a method that translates a task into a program, then learns a policy for resource allocation that, on each input, selects foundation model \"backends\" for each program module. The policy uses smaller, cheaper backends to handle simpler subtasks, while allowing more complex su"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.07247","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/2504.07247/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":"2504.07247","created_at":"2026-07-05T11:51:25.182063+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.07247v2","created_at":"2026-07-05T11:51:25.182063+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.07247","created_at":"2026-07-05T11:51:25.182063+00:00"},{"alias_kind":"pith_short_12","alias_value":"5SQVB75ABEDG","created_at":"2026-07-05T11:51:25.182063+00:00"},{"alias_kind":"pith_short_16","alias_value":"5SQVB75ABEDGFKZM","created_at":"2026-07-05T11:51:25.182063+00:00"},{"alias_kind":"pith_short_8","alias_value":"5SQVB75A","created_at":"2026-07-05T11:51:25.182063+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.13647","citing_title":"FlowCompile: An Optimizing Compiler for Structured LLM Workflows","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5SQVB75ABEDGFKZMBGKB73CMPF","json":"https://pith.science/pith/5SQVB75ABEDGFKZMBGKB73CMPF.json","graph_json":"https://pith.science/api/pith-number/5SQVB75ABEDGFKZMBGKB73CMPF/graph.json","events_json":"https://pith.science/api/pith-number/5SQVB75ABEDGFKZMBGKB73CMPF/events.json","paper":"https://pith.science/paper/5SQVB75A"},"agent_actions":{"view_html":"https://pith.science/pith/5SQVB75ABEDGFKZMBGKB73CMPF","download_json":"https://pith.science/pith/5SQVB75ABEDGFKZMBGKB73CMPF.json","view_paper":"https://pith.science/paper/5SQVB75A","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.07247&json=true","fetch_graph":"https://pith.science/api/pith-number/5SQVB75ABEDGFKZMBGKB73CMPF/graph.json","fetch_events":"https://pith.science/api/pith-number/5SQVB75ABEDGFKZMBGKB73CMPF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5SQVB75ABEDGFKZMBGKB73CMPF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5SQVB75ABEDGFKZMBGKB73CMPF/action/storage_attestation","attest_author":"https://pith.science/pith/5SQVB75ABEDGFKZMBGKB73CMPF/action/author_attestation","sign_citation":"https://pith.science/pith/5SQVB75ABEDGFKZMBGKB73CMPF/action/citation_signature","submit_replication":"https://pith.science/pith/5SQVB75ABEDGFKZMBGKB73CMPF/action/replication_record"}},"created_at":"2026-07-05T11:51:25.182063+00:00","updated_at":"2026-07-05T11:51:25.182063+00:00"}