{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SLACG3NTHTXJGZVLW7SFPNJFIL","short_pith_number":"pith:SLACG3NT","schema_version":"1.0","canonical_sha256":"92c0236db33cee9366abb7e457b52542e169c2379007fd33661a3f3a6f6aef40","source":{"kind":"arxiv","id":"2507.11948","version":1},"attestation_state":"computed","paper":{"title":"Kevin: Multi-Turn RL for Generating CUDA Kernels","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.PF","cs.SE"],"primary_cat":"cs.LG","authors_text":"Ben Pan, Carlo Baronio, Pietro Marsella, Silas Alberti, Simon Guo","submitted_at":"2025-07-16T06:33:07Z","abstract_excerpt":"Writing GPU kernels is a challenging task and critical for AI systems' efficiency. It is also highly iterative: domain experts write code and improve performance through execution feedback. Moreover, it presents verifiable rewards like correctness and speedup, making it a natural environment to apply Reinforcement Learning (RL). To explicitly incorporate the iterative nature of this process into training, we develop a flexible multi-turn RL recipe that addresses unique challenges encountered in real-world settings, such as learning from long trajectories and effective reward attribution across"},"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":"2507.11948","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-16T06:33:07Z","cross_cats_sorted":["cs.AI","cs.PF","cs.SE"],"title_canon_sha256":"c11a94431d45ccd9b30d036c3941064e2b95c95888fe57e55835d1b2afc09ec4","abstract_canon_sha256":"a0aa03e2ed9cfe2e695fc274f63d8cc81f5204f7f524eeaf8e675c32e078ef5a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:38:11.333663Z","signature_b64":"6p8bha7pc8xlpVIc5oSzzSSLGgu2G4LN/jr91gHoFo55hVmUuRzeKJ5SJiLu9JbbdEJYeXzzPu8GX8b9JVq7Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"92c0236db33cee9366abb7e457b52542e169c2379007fd33661a3f3a6f6aef40","last_reissued_at":"2026-07-05T11:38:11.333145Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:38:11.333145Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Kevin: Multi-Turn RL for Generating CUDA Kernels","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.PF","cs.SE"],"primary_cat":"cs.LG","authors_text":"Ben Pan, Carlo Baronio, Pietro Marsella, Silas Alberti, Simon Guo","submitted_at":"2025-07-16T06:33:07Z","abstract_excerpt":"Writing GPU kernels is a challenging task and critical for AI systems' efficiency. It is also highly iterative: domain experts write code and improve performance through execution feedback. Moreover, it presents verifiable rewards like correctness and speedup, making it a natural environment to apply Reinforcement Learning (RL). To explicitly incorporate the iterative nature of this process into training, we develop a flexible multi-turn RL recipe that addresses unique challenges encountered in real-world settings, such as learning from long trajectories and effective reward attribution across"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.11948","kind":"arxiv","version":1},"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/2507.11948/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":"2507.11948","created_at":"2026-07-05T11:38:11.333203+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.11948v1","created_at":"2026-07-05T11:38:11.333203+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.11948","created_at":"2026-07-05T11:38:11.333203+00:00"},{"alias_kind":"pith_short_12","alias_value":"SLACG3NTHTXJ","created_at":"2026-07-05T11:38:11.333203+00:00"},{"alias_kind":"pith_short_16","alias_value":"SLACG3NTHTXJGZVL","created_at":"2026-07-05T11:38:11.333203+00:00"},{"alias_kind":"pith_short_8","alias_value":"SLACG3NT","created_at":"2026-07-05T11:38:11.333203+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17518","citing_title":"SpecGen: Accelerating Agentic Kernel Optimization with Speculative Generation","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01590","citing_title":"Hawk: Harnessing Hardware-Aware Knowledge for High-Performance NPU Kernel Generation","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08455","citing_title":"CUDABeaver: Benchmarking LLM-Based Automated CUDA Debugging","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30359","citing_title":"Kernel Foundry: A Diagnosis-driven Evolutionary Kernel Optimizer with Multi-Experts","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30442","citing_title":"The FIL Hypothesis: Inductive Biases Help with Kernel Engineering","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25954","citing_title":"Step-TP: A Grounded, Step-Level Dataset with Chain-of-Thought Reasoning for LLM-Guided Tensor Program Optimization","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28213","citing_title":"Learning When to Optimize: Verified Optimization Skills from Expert GPU-Kernel Lineages","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29357","citing_title":"PassNet: Scaling Large Language Models for Graph Compiler Pass Generation","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23215","citing_title":"FastKernels: Benchmarking GPU Kernel Generation in Production","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2603.23566","citing_title":"AscendOptimizer: Episodic Agent for Ascend NPU Operator Optimization","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2603.28342","citing_title":"Kernel-Smith: A Unified Recipe for Evolutionary Kernel Optimization","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03144","citing_title":"InCoder-32B-Thinking: Industrial Code World Model for Thinking","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08467","citing_title":"CUDAHercules: Benchmarking Hardware-Aware Expert-level CUDA Optimization for LLMs","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08455","citing_title":"CUDABeaver: Benchmarking LLM-Based Automated CUDA Debugging","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16625","citing_title":"AdaExplore: Failure-Driven Adaptation and Diversity-Preserving Search for Efficient Kernel Generation","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SLACG3NTHTXJGZVLW7SFPNJFIL","json":"https://pith.science/pith/SLACG3NTHTXJGZVLW7SFPNJFIL.json","graph_json":"https://pith.science/api/pith-number/SLACG3NTHTXJGZVLW7SFPNJFIL/graph.json","events_json":"https://pith.science/api/pith-number/SLACG3NTHTXJGZVLW7SFPNJFIL/events.json","paper":"https://pith.science/paper/SLACG3NT"},"agent_actions":{"view_html":"https://pith.science/pith/SLACG3NTHTXJGZVLW7SFPNJFIL","download_json":"https://pith.science/pith/SLACG3NTHTXJGZVLW7SFPNJFIL.json","view_paper":"https://pith.science/paper/SLACG3NT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.11948&json=true","fetch_graph":"https://pith.science/api/pith-number/SLACG3NTHTXJGZVLW7SFPNJFIL/graph.json","fetch_events":"https://pith.science/api/pith-number/SLACG3NTHTXJGZVLW7SFPNJFIL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SLACG3NTHTXJGZVLW7SFPNJFIL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SLACG3NTHTXJGZVLW7SFPNJFIL/action/storage_attestation","attest_author":"https://pith.science/pith/SLACG3NTHTXJGZVLW7SFPNJFIL/action/author_attestation","sign_citation":"https://pith.science/pith/SLACG3NTHTXJGZVLW7SFPNJFIL/action/citation_signature","submit_replication":"https://pith.science/pith/SLACG3NTHTXJGZVLW7SFPNJFIL/action/replication_record"}},"created_at":"2026-07-05T11:38:11.333203+00:00","updated_at":"2026-07-05T11:38:11.333203+00:00"}