{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:VBSDGLFV7JA35DN5MR73OSL4FV","short_pith_number":"pith:VBSDGLFV","schema_version":"1.0","canonical_sha256":"a864332cb5fa41be8dbd647fb7497c2d7553a07e73fee1dc6410093e82c41290","source":{"kind":"arxiv","id":"2603.02637","version":2},"attestation_state":"computed","paper":{"title":"StitchCUDA: An Automated Multi-Agents End-to-End GPU Programing Framework with Rubric-based Agentic Reinforcement Learning","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CL","cs.PL"],"primary_cat":"cs.MA","authors_text":"Caiwen Ding, Mingyi Hong, Shiyang Li, Winson Chen, Yuebo Luo, Zijian Zhang","submitted_at":"2026-03-03T06:04:49Z","abstract_excerpt":"Modern machine learning (ML) workloads increasingly rely on GPUs, yet achieving high end-to-end performance remains challenging due to dependencies on both GPU kernel efficiency and host-side settings. Although LLM-based methods show promise on automated GPU kernel generation, prior works mainly focus on single-kernel optimization and do not extend to end-to-end programs, hindering practical deployment.\n  To address the challenge, in this work, we propose StitchCUDA, a multi-agent framework for end-to-end GPU program generation, with three specialized agents: a Planner to orchestrate whole sys"},"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":"2603.02637","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.MA","submitted_at":"2026-03-03T06:04:49Z","cross_cats_sorted":["cs.CL","cs.PL"],"title_canon_sha256":"2f04b24ccc3d3e9ae3e710f35c1feb62d4268c6751e6c308c6c38d69e2ba1fb3","abstract_canon_sha256":"36f768c5d9cfe25caa99bee96ec0cd2c6bad20e561282d59db0c1e99b662172e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-11T02:23:51.953114Z","signature_b64":"aT5qFSKhazCSiLz0QyPYcoUrGoq0w5u+7qTLZX1sNNw8gfgjUnTlYdrzet0GXlzXaKhpJJblU/7z9YVecB8KCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a864332cb5fa41be8dbd647fb7497c2d7553a07e73fee1dc6410093e82c41290","last_reissued_at":"2026-08-11T02:23:51.951067Z","signature_status":"signed_v1","first_computed_at":"2026-08-11T02:23:51.951067Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"StitchCUDA: An Automated Multi-Agents End-to-End GPU Programing Framework with Rubric-based Agentic Reinforcement Learning","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CL","cs.PL"],"primary_cat":"cs.MA","authors_text":"Caiwen Ding, Mingyi Hong, Shiyang Li, Winson Chen, Yuebo Luo, Zijian Zhang","submitted_at":"2026-03-03T06:04:49Z","abstract_excerpt":"Modern machine learning (ML) workloads increasingly rely on GPUs, yet achieving high end-to-end performance remains challenging due to dependencies on both GPU kernel efficiency and host-side settings. Although LLM-based methods show promise on automated GPU kernel generation, prior works mainly focus on single-kernel optimization and do not extend to end-to-end programs, hindering practical deployment.\n  To address the challenge, in this work, we propose StitchCUDA, a multi-agent framework for end-to-end GPU program generation, with three specialized agents: a Planner to orchestrate whole sys"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.02637","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/2603.02637/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":"2603.02637","created_at":"2026-08-11T02:23:51.952018+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.02637v2","created_at":"2026-08-11T02:23:51.952018+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.02637","created_at":"2026-08-11T02:23:51.952018+00:00"},{"alias_kind":"pith_short_12","alias_value":"VBSDGLFV7JA3","created_at":"2026-08-11T02:23:51.952018+00:00"},{"alias_kind":"pith_short_16","alias_value":"VBSDGLFV7JA35DN5","created_at":"2026-08-11T02:23:51.952018+00:00"},{"alias_kind":"pith_short_8","alias_value":"VBSDGLFV","created_at":"2026-08-11T02:23:51.952018+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2606.26453","citing_title":"Optimizing CUDA like a Human: Micro-Profiling Tools as Expert Surrogates for LLM-Based GPU Kernel Optimization","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2603.23566","citing_title":"AscendOptimizer: Episodic Agent for Ascend NPU Operator Optimization","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2604.13602","citing_title":"Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges","ref_index":180,"is_internal_anchor":true},{"citing_arxiv_id":"2604.26666","citing_title":"FACT: Compositional Kernel Synthesis with a Three-Stage Agentic Workflow","ref_index":8,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VBSDGLFV7JA35DN5MR73OSL4FV","json":"https://pith.science/pith/VBSDGLFV7JA35DN5MR73OSL4FV.json","graph_json":"https://pith.science/api/pith-number/VBSDGLFV7JA35DN5MR73OSL4FV/graph.json","events_json":"https://pith.science/api/pith-number/VBSDGLFV7JA35DN5MR73OSL4FV/events.json","paper":"https://pith.science/paper/VBSDGLFV"},"agent_actions":{"view_html":"https://pith.science/pith/VBSDGLFV7JA35DN5MR73OSL4FV","download_json":"https://pith.science/pith/VBSDGLFV7JA35DN5MR73OSL4FV.json","view_paper":"https://pith.science/paper/VBSDGLFV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.02637&json=true","fetch_graph":"https://pith.science/api/pith-number/VBSDGLFV7JA35DN5MR73OSL4FV/graph.json","fetch_events":"https://pith.science/api/pith-number/VBSDGLFV7JA35DN5MR73OSL4FV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VBSDGLFV7JA35DN5MR73OSL4FV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VBSDGLFV7JA35DN5MR73OSL4FV/action/storage_attestation","attest_author":"https://pith.science/pith/VBSDGLFV7JA35DN5MR73OSL4FV/action/author_attestation","sign_citation":"https://pith.science/pith/VBSDGLFV7JA35DN5MR73OSL4FV/action/citation_signature","submit_replication":"https://pith.science/pith/VBSDGLFV7JA35DN5MR73OSL4FV/action/replication_record"}},"created_at":"2026-08-11T02:23:51.952018+00:00","updated_at":"2026-08-11T02:23:51.952018+00:00"}