{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.23215", "kind": "paper", "url": "https://arxiv.org/abs/2605.23215"}, {"id": "gh:snowflake-ai-research/fastkernels", "kind": "repo", "url": "https://github.com/Snowflake-AI-Research/fastkernels"}], "authors": ["Gabriele Oliaro", "Jaeseong Lee", "Yichao Fu", "Zhaoyuan Su", "Owen Lu", "Sreeram Vennam", "May Jiang", "Junli Wang", "Hao Zhang", "Zhihao Jia", "Samyam Rajbhandari"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.23215", "source_url": "https://arxiv.org/abs/2605.23215"}, "categories": ["Code", "Code & Software", "Code generation", "Software & AI Compute"], "description": {"en": "FastKernels evaluates GPU kernel generation for production inference across 46 representative architectures spanning 8 categories. It measures correctness and speedup of candidate kernels against reference implementations at multiple abstraction levels, from single-kernel to end-to-end serving."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.23215", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.23215", "title": "FastKernels"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Carnegie Mellon University", "Oldham Council"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2605.23215", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2605.23215", "source_url": "https://openalex.org/W7162325420"}, "key": "claire-radar:2605.23215", "modality": null, "name": "FastKernels", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_fastkernels_60e561f1", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "72dece719c4b7ea1bc0171227320ab311dd77bb9a16b8648d87c68cbc71e2c86", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.23215"}, "publisher": null, "released": "2026-05-22", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.23215", "source_url": "https://arxiv.org/abs/2605.23215"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-23215", "source": "claire_radar", "source_benchmark_id": "2605.23215", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Machine Generated", "applicationDomains": ["Software & AI Compute"], "area": "Code & Software", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/Snowflake-AI-Research/fastkernels", "githubScope": "benchmark_repo", "githubStars": 6, "hfDailySubmittedAt": "2026-05-27T00:00:00.000Z", "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 4, "hfPaperUrl": "https://huggingface.co/papers/2605.23215", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}}, "status": "fresh"}, "capabilities": ["Code generation"], "confidence": "High", "construction": "Original Synthetic", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "c697a4fc536c397d33cd9246d6cabdeef8722d60539728d93d563042db782833", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "FastKernels evaluates GPU kernel generation for production inference across 46 representative architectures spanning 8 categories. It measures correctness and speedup of candidate kernels against reference implementations at multiple abstraction levels, from single-kernel to end-to-end serving.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce FastKernels, a kernel benchmark built around a minimal set of 46 representative architectures spanning 8 categories, whose kernels collectively subsume those of 96.2% (409/425) of HuggingFace Transformers architectures."}, "familyId": "bmf_c3b50d55e8f5", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/Snowflake-AI-Research/fastkernels", "url": "https://github.com/Snowflake-AI-Research/fastkernels"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/Snowflake-AI-Research/fastkernels", "url": "https://huggingface.co/papers/2605.23215"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/Snowflake-AI-Research/fastkernels", "url": "https://arxiv.org/abs/2605.23215"}], "scope": "benchmark_repo", "url": "https://github.com/Snowflake-AI-Research/fastkernels", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_fastkernels_60e561f1", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Software & Cloud"], "links": {"code": "https://github.com/Snowflake-AI-Research/fastkernels", "data": null, "hfPaper": "https://huggingface.co/papers/2605.23215", "paper": "https://arxiv.org/abs/2605.23215", "pdf": "https://arxiv.org/pdf/2605.23215", "project": null, "report": "https://arxiv.org/abs/2605.23215"}, "metrics": [], "motivation": "LLM-based agents for GPU kernel generation are advancing rapidly, yet their progress is fundamentally constrained by the benchmarks they optimize against.", "name": "FastKernels", "oneLine": "FastKernels evaluates GPU kernel generation for production inference across 46 representative architectures spanning 8 categories. It measures correctness and speedup of candidate kernels against reference implementations at multiple abstraction levels, from single-kernel to end-to-end serving.", "paperTitle": "FastKernels: Benchmarking GPU Kernel Generation in Production", "primaryDomain": "Software & AI Compute", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.23215", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Snowflake AI Research", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/Snowflake-AI-Research/fastkernels"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-22", "source": {"authors": ["Gabriele Oliaro", "Yichao Fu", "May Jiang", "Owen Lu", "Junli Wang", "Zhihao Jia", "Hao Zhang", "and Samyam Rajbhandari"], "categories": ["cs.LG", "cs.AI", "cs.CL"], "id": "2605.23215", "title": "FastKernels: Benchmarking GPU Kernel Generation in Production", "type": "arxiv", "url": "https://arxiv.org/abs/2605.23215"}, "sourceUpdatedAt": "2026-05-22T04:19:04Z", "topics": ["Code"], "venueAttempts": [], "whyItMatters": "Existing kernel benchmarks reward sandbox optimizations that fail in production. FastKernels aligns evaluation with real inference frameworks, providing a benchmark where scores translate to throughput improvements in production codebases."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
