{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:NWZ2LUZQPZPVMV62NUPT4I7GKI","short_pith_number":"pith:NWZ2LUZQ","schema_version":"1.0","canonical_sha256":"6db3a5d3307e5f5657da6d1f3e23e65215f3144e17649beb2da19056b17d8538","source":{"kind":"arxiv","id":"2203.13474","version":5},"attestation_state":"computed","paper":{"title":"CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"The same problem intent given to CodeGen in multiple turns produces substantially better programs than a single turn on the MTPB benchmark.","cross_cats":["cs.CL","cs.PL"],"primary_cat":"cs.LG","authors_text":"Bo Pang, Caiming Xiong, Erik Nijkamp, Hiroaki Hayashi, Huan Wang, Lifu Tu, Silvio Savarese, Yingbo Zhou","submitted_at":"2022-03-25T06:55:15Z","abstract_excerpt":"Program synthesis strives to generate a computer program as a solution to a given problem specification, expressed with input-output examples or natural language descriptions. The prevalence of large language models advances the state-of-the-art for program synthesis, though limited training resources and data impede open access to such models. To democratize this, we train and release a family of large language models up to 16.1B parameters, called CODEGEN, on natural language and programming language data, and open source the training library JAXFORMER. We show the utility of the trained mod"},"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":"2203.13474","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2022-03-25T06:55:15Z","cross_cats_sorted":["cs.CL","cs.PL"],"title_canon_sha256":"9ba27ade9785782edaa81753e0dc8d1830e1e65541a9e8a1d83f1c4511ff9249","abstract_canon_sha256":"d936db27aa7f28c7873b54f880d9ed3a292e1595afe3a75199ea36c7bd3f6dab"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:46:14.025883Z","signature_b64":"z7nuamhDPUIj6TLv/ClzJdmP+5Kd35AZ5/o0XdVbzyjnzAWL+dqmSTZguxnb+O3EkF3+PjFAHaTcpXwdaHNjDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6db3a5d3307e5f5657da6d1f3e23e65215f3144e17649beb2da19056b17d8538","last_reissued_at":"2026-07-05T05:46:14.025352Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:46:14.025352Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"The same problem intent given to CodeGen in multiple turns produces substantially better programs than a single turn on the MTPB benchmark.","cross_cats":["cs.CL","cs.PL"],"primary_cat":"cs.LG","authors_text":"Bo Pang, Caiming Xiong, Erik Nijkamp, Hiroaki Hayashi, Huan Wang, Lifu Tu, Silvio Savarese, Yingbo Zhou","submitted_at":"2022-03-25T06:55:15Z","abstract_excerpt":"Program synthesis strives to generate a computer program as a solution to a given problem specification, expressed with input-output examples or natural language descriptions. The prevalence of large language models advances the state-of-the-art for program synthesis, though limited training resources and data impede open access to such models. To democratize this, we train and release a family of large language models up to 16.1B parameters, called CODEGEN, on natural language and programming language data, and open source the training library JAXFORMER. We show the utility of the trained mod"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our analysis on MTPB shows that the same intent provided to CODEGEN in multi-turn fashion significantly improves program synthesis over that provided as a single turn.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The multi-turn prompts in MTPB accurately capture natural human problem decomposition without introducing selection bias or making the benchmark easier than real-world use cases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CodeGen models match prior SOTA on HumanEval zero-shot code generation, and multi-turn prompting on the new MTPB benchmark significantly improves synthesis over single-turn inputs.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"The same problem intent given to CodeGen in multiple turns produces substantially better programs than a single turn on the MTPB benchmark.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"a62e9dd77e44d9ae3f1646fec1845d4cb3a2b0969e52520f6ca5741ac7e1341a"},"source":{"id":"2203.13474","kind":"arxiv","version":5},"verdict":{"id":"0616d8ac-1155-4b57-ba6b-dd66dd6d1a1c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T16:59:25.733083Z","strongest_claim":"Our analysis on MTPB shows that the same intent provided to CODEGEN in multi-turn fashion significantly improves program synthesis over that provided as a single turn.","one_line_summary":"CodeGen models match prior SOTA on HumanEval zero-shot code generation, and multi-turn prompting on the new MTPB benchmark significantly improves synthesis over single-turn inputs.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The multi-turn prompts in MTPB accurately capture natural human problem decomposition without introducing selection bias or making the benchmark easier than real-world use cases.","pith_extraction_headline":"The same problem intent given to CodeGen in multiple turns produces substantially better programs than a single turn on the MTPB benchmark."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2203.13474/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":"2203.13474","created_at":"2026-07-05T05:46:14.025406+00:00"},{"alias_kind":"arxiv_version","alias_value":"2203.13474v5","created_at":"2026-07-05T05:46:14.025406+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2203.13474","created_at":"2026-07-05T05:46:14.025406+00:00"},{"alias_kind":"pith_short_12","alias_value":"NWZ2LUZQPZPV","created_at":"2026-07-05T05:46:14.025406+00:00"},{"alias_kind":"pith_short_16","alias_value":"NWZ2LUZQPZPVMV62","created_at":"2026-07-05T05:46:14.025406+00:00"},{"alias_kind":"pith_short_8","alias_value":"NWZ2LUZQ","created_at":"2026-07-05T05:46:14.025406+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":60,"internal_anchor_count":60,"sample":[{"citing_arxiv_id":"2607.07619","citing_title":"Rethinking Code Performance Benchmarks for LLMs","ref_index":74,"is_internal_anchor":true},{"citing_arxiv_id":"2606.24834","citing_title":"Accuracy and Satisfaction in Multi-Turn LLM Dialogues for NFR Assessment","ref_index":65,"is_internal_anchor":true},{"citing_arxiv_id":"2606.26490","citing_title":"An Empirical Study of LLM-Generated Specifications for VeriFast","ref_index":47,"is_internal_anchor":true},{"citing_arxiv_id":"2606.22082","citing_title":"CodeTeam: An LLM-Powered Multi-Agent Framework for Repository-Level Code Generation","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2606.20517","citing_title":"Multi-LCB: Extending LiveCodeBench to Multiple Programming Languages","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2606.20683","citing_title":"From Question Answering to Task Completion: A Survey on Agent System and Harness Design","ref_index":70,"is_internal_anchor":true},{"citing_arxiv_id":"2607.02186","citing_title":"UA-ChatDev: Uncertainty-Aware Multi-Agent Collaboration for Reliable Software Development","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2606.08476","citing_title":"FlashCP: Load-Balanced Communication-Efficient Context Parallelism for LLM Training","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2606.07999","citing_title":"Efficient Skill Grounding via Code Refactoring with Small Language Models","ref_index":71,"is_internal_anchor":true},{"citing_arxiv_id":"2607.00035","citing_title":"Making Failure Safe: A Constrained, Verifiable Agent Framework for Open-Web Data Collection","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2606.00186","citing_title":"How to Compare the Security of Code Written by Humans to LLM-generated Code","ref_index":38,"is_internal_anchor":true},{"citing_arxiv_id":"2606.07568","citing_title":"A Systematic Study of Behavioral Cloning for Scientific Data Annotation","ref_index":289,"is_internal_anchor":true},{"citing_arxiv_id":"2606.00186","citing_title":"How to Compare the Security of Code Written by Humans to LLM-generated Code","ref_index":38,"is_internal_anchor":true},{"citing_arxiv_id":"2606.13097","citing_title":"Functional Cache Grafting: Robust and Rapid Code-Policy Synthesis for Embodied Agents","ref_index":19,"is_internal_anchor":true},{"citing_arxiv_id":"2402.01411","citing_title":"CodePori: Large-Scale System for Autonomous Software Development Using Multi-Agent Technology","ref_index":64,"is_internal_anchor":true},{"citing_arxiv_id":"2404.01535","citing_title":"Assessing, Exploiting, and Mitigating Syntactic Robustness Failures in LLM-Based Code Generation","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2408.01055","citing_title":"Towards Agentic Runtime Healing","ref_index":38,"is_internal_anchor":true},{"citing_arxiv_id":"2412.04590","citing_title":"Specification-Driven Code Translation Powered by Large Language Models: How Far Are We?","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2503.02497","citing_title":"A PennyLane-Centric Dataset to Enhance LLM-based Quantum Code Generation using RAG","ref_index":30,"is_internal_anchor":true},{"citing_arxiv_id":"2504.16584","citing_title":"Case Study: Fine-tuning Small Language Models for Accurate and Private CWE Detection in Python Code","ref_index":24,"is_internal_anchor":true},{"citing_arxiv_id":"2504.16584","citing_title":"Case Study: Fine-tuning Small Language Models for Accurate and Private CWE Detection in Python Code","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15585","citing_title":"See Before You Code: Learning Visual Priors for Spatially Aware Educational Animation Generation","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17458","citing_title":"ClaHF: A Human Feedback-inspired Reinforcement Learning Framework for Improving Classification Tasks","ref_index":40,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18747","citing_title":"Code as Agent Harness","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17957","citing_title":"Contextualized Code Pretraining for Code Generation","ref_index":34,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NWZ2LUZQPZPVMV62NUPT4I7GKI","json":"https://pith.science/pith/NWZ2LUZQPZPVMV62NUPT4I7GKI.json","graph_json":"https://pith.science/api/pith-number/NWZ2LUZQPZPVMV62NUPT4I7GKI/graph.json","events_json":"https://pith.science/api/pith-number/NWZ2LUZQPZPVMV62NUPT4I7GKI/events.json","paper":"https://pith.science/paper/NWZ2LUZQ"},"agent_actions":{"view_html":"https://pith.science/pith/NWZ2LUZQPZPVMV62NUPT4I7GKI","download_json":"https://pith.science/pith/NWZ2LUZQPZPVMV62NUPT4I7GKI.json","view_paper":"https://pith.science/paper/NWZ2LUZQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2203.13474&json=true","fetch_graph":"https://pith.science/api/pith-number/NWZ2LUZQPZPVMV62NUPT4I7GKI/graph.json","fetch_events":"https://pith.science/api/pith-number/NWZ2LUZQPZPVMV62NUPT4I7GKI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NWZ2LUZQPZPVMV62NUPT4I7GKI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NWZ2LUZQPZPVMV62NUPT4I7GKI/action/storage_attestation","attest_author":"https://pith.science/pith/NWZ2LUZQPZPVMV62NUPT4I7GKI/action/author_attestation","sign_citation":"https://pith.science/pith/NWZ2LUZQPZPVMV62NUPT4I7GKI/action/citation_signature","submit_replication":"https://pith.science/pith/NWZ2LUZQPZPVMV62NUPT4I7GKI/action/replication_record"}},"created_at":"2026-07-05T05:46:14.025406+00:00","updated_at":"2026-07-05T05:46:14.025406+00:00"}