{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NONTTJ6G2LSMYDFCHRQ2IXZBVJ","short_pith_number":"pith:NONTTJ6G","schema_version":"1.0","canonical_sha256":"6b9b39a7c6d2e4cc0ca23c61a45f21aa722e3f2a52e1c144c47258c973b4a597","source":{"kind":"arxiv","id":"2605.23215","version":1},"attestation_state":"computed","paper":{"title":"FastKernels: Benchmarking GPU Kernel Generation in Production","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Gabriele Oliaro, Hao Zhang, Junli Wang, May Jiang, Owen Lu, Samyam Rajbhandari, Yichao Fu, Zhihao Jia","submitted_at":"2026-05-22T04:19:04Z","abstract_excerpt":"LLM-based agents for GPU kernel generation are advancing rapidly, yet their progress is fundamentally constrained by the benchmarks they optimize against. Existing benchmarks are poorly aligned with production inference frameworks: they evaluate kernels on a single GPU with synthetic inputs, ignore the surrounding compilation stack, and reward replicating known optimizations rather than discovering new ones. The resulting reward signals are misleading: agents learn to generate kernels that score well in sandboxes but introduce interface incompatibilities, compilation-stack conflicts, and silen"},"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":"2605.23215","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-22T04:19:04Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"a198a44aa7e9a120262609be63f456d0fcaf0c782b64871ba1478db58354cfe0","abstract_canon_sha256":"4b7d518a323054661c8a9ffc508f2fd81c862b72e59fafefdaa5f38a95820caf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:01:44.003543Z","signature_b64":"1b9ikkQQP1WMwKNXuwJXcyisKrQMdSNjlx1oZE1ErnC5cWoPjMavc13vgdpRM7Gnl+F3aofADSQQo7pt5LR6Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6b9b39a7c6d2e4cc0ca23c61a45f21aa722e3f2a52e1c144c47258c973b4a597","last_reissued_at":"2026-05-25T02:01:44.002827Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:01:44.002827Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FastKernels: Benchmarking GPU Kernel Generation in Production","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Gabriele Oliaro, Hao Zhang, Junli Wang, May Jiang, Owen Lu, Samyam Rajbhandari, Yichao Fu, Zhihao Jia","submitted_at":"2026-05-22T04:19:04Z","abstract_excerpt":"LLM-based agents for GPU kernel generation are advancing rapidly, yet their progress is fundamentally constrained by the benchmarks they optimize against. Existing benchmarks are poorly aligned with production inference frameworks: they evaluate kernels on a single GPU with synthetic inputs, ignore the surrounding compilation stack, and reward replicating known optimizations rather than discovering new ones. The resulting reward signals are misleading: agents learn to generate kernels that score well in sandboxes but introduce interface incompatibilities, compilation-stack conflicts, and silen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.23215","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/2605.23215/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":"2605.23215","created_at":"2026-05-25T02:01:44.002943+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.23215v1","created_at":"2026-05-25T02:01:44.002943+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.23215","created_at":"2026-05-25T02:01:44.002943+00:00"},{"alias_kind":"pith_short_12","alias_value":"NONTTJ6G2LSM","created_at":"2026-05-25T02:01:44.002943+00:00"},{"alias_kind":"pith_short_16","alias_value":"NONTTJ6G2LSMYDFC","created_at":"2026-05-25T02:01:44.002943+00:00"},{"alias_kind":"pith_short_8","alias_value":"NONTTJ6G","created_at":"2026-05-25T02:01:44.002943+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NONTTJ6G2LSMYDFCHRQ2IXZBVJ","json":"https://pith.science/pith/NONTTJ6G2LSMYDFCHRQ2IXZBVJ.json","graph_json":"https://pith.science/api/pith-number/NONTTJ6G2LSMYDFCHRQ2IXZBVJ/graph.json","events_json":"https://pith.science/api/pith-number/NONTTJ6G2LSMYDFCHRQ2IXZBVJ/events.json","paper":"https://pith.science/paper/NONTTJ6G"},"agent_actions":{"view_html":"https://pith.science/pith/NONTTJ6G2LSMYDFCHRQ2IXZBVJ","download_json":"https://pith.science/pith/NONTTJ6G2LSMYDFCHRQ2IXZBVJ.json","view_paper":"https://pith.science/paper/NONTTJ6G","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.23215&json=true","fetch_graph":"https://pith.science/api/pith-number/NONTTJ6G2LSMYDFCHRQ2IXZBVJ/graph.json","fetch_events":"https://pith.science/api/pith-number/NONTTJ6G2LSMYDFCHRQ2IXZBVJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NONTTJ6G2LSMYDFCHRQ2IXZBVJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NONTTJ6G2LSMYDFCHRQ2IXZBVJ/action/storage_attestation","attest_author":"https://pith.science/pith/NONTTJ6G2LSMYDFCHRQ2IXZBVJ/action/author_attestation","sign_citation":"https://pith.science/pith/NONTTJ6G2LSMYDFCHRQ2IXZBVJ/action/citation_signature","submit_replication":"https://pith.science/pith/NONTTJ6G2LSMYDFCHRQ2IXZBVJ/action/replication_record"}},"created_at":"2026-05-25T02:01:44.002943+00:00","updated_at":"2026-05-25T02:01:44.002943+00:00"}