{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.28819", "kind": "paper", "url": "https://arxiv.org/abs/2605.28819"}, {"id": "gh:sphere-ai-lab/peft-arena", "kind": "repo", "url": "https://github.com/Sphere-AI-Lab/PEFT-Arena"}, {"kind": "website", "url": "https://spherelab.ai/PEFT-Arena/"}], "authors": ["Yangyi Huang", "Ruotian Peng", "Zeju Qiu", "Jiale Kang", "Yandong Wen", "Bernhard Schölkopf", "Weiyang Liu"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.28819", "source_url": "https://arxiv.org/abs/2605.28819"}, "categories": ["General AI", "Language & Knowledge", "cs.LG"], "description": {"en": "PEFT-Arena benchmark jointly evaluates target-domain performance and retention of pretrained capabilities for parameter-efficient finetuning methods. It covers mathematical and medical reasoning as target domains and measures general capability retention on benchmarks including BBH, IFEval, and NQ with SFT and RLVR training settings."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.28819", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.28819", "title": "PEFT-Arena"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2605.28819", "modality": null, "name": "PEFT-Arena", "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_peft-arena_b7338aaa", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "cf55a767eb85f7cf63f3451009efdc2e655cc3f4ac5df35587bd7a1690dc01ac", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.28819"}, "publisher": null, "released": "2026-05-27", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.28819", "source_url": "https://arxiv.org/abs/2605.28819"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-28819", "source": "claire_radar", "source_benchmark_id": "2605.28819", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/Sphere-AI-Lab/PEFT-Arena", "githubScope": "benchmark_repo", "githubStars": 29, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 7, "hfPaperUrl": "https://huggingface.co/papers/2605.28819", "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": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "deabbebd645b7ebe77205d30ddd4a8cce565c608114c85c53cd71e0d4157c0bd", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "PEFT-Arena benchmark jointly evaluates target-domain performance and retention of pretrained capabilities for parameter-efficient finetuning methods. It covers mathematical and medical reasoning as target domains and measures general capability retention on benchmarks including BBH, IFEval, and NQ with SFT and RLVR training settings.", "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 PEFT-Arena, a benchmark that jointly measures downstream performance and general capability retention."}, "familyId": "bmf_f5e847eb890b", "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/Sphere-AI-Lab/PEFT-Arena", "url": "https://github.com/Sphere-AI-Lab/PEFT-Arena"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/Sphere-AI-Lab/PEFT-Arena", "url": "https://huggingface.co/papers/2605.28819"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/Sphere-AI-Lab/PEFT-Arena", "url": "https://spherelab.ai/PEFT-Arena/"}], "scope": "benchmark_repo", "url": "https://github.com/Sphere-AI-Lab/PEFT-Arena", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_peft-arena_b7338aaa", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/Sphere-AI-Lab/PEFT-Arena", "data": null, "hfPaper": "https://huggingface.co/papers/2605.28819", "paper": "https://arxiv.org/abs/2605.28819", "pdf": "https://arxiv.org/pdf/2605.28819", "project": "https://spherelab.ai/PEFT-Arena/", "report": "https://arxiv.org/abs/2605.28819"}, "metrics": [], "motivation": "Parameter-efficient finetuning (PEFT) has become the standard approach for adapting large language models, yet evaluations largely emphasize downstream accuracy while overlooking the retention of pretrained capabilities.", "name": "PEFT-Arena", "oneLine": "PEFT-Arena benchmark jointly evaluates target-domain performance and retention of pretrained capabilities for parameter-efficient finetuning methods. It covers mathematical and medical reasoning as target domains and measures general capability retention on benchmarks including BBH, IFEval, and NQ with SFT and RLVR training settings.", "paperTitle": "PEFT-Arena: Understanding Parameter-Efficient Finetuning from a Stability-Plasticity Perspective", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.28819", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Sphere AI Lab", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://spherelab.ai/PEFT-Arena/"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-27", "source": {"authors": ["Yangyi Huang and Ruotian Peng and Zeju Qiu and Jiale Kang and Yandong Wen and Bernhard Sch\\\"olkopf and Weiyang Liu"], "categories": ["cs.LG", "cs.CL"], "id": "2605.28819", "title": "PEFT-Arena: Understanding Parameter-Efficient Finetuning from a Stability-Plasticity Perspective", "type": "arxiv", "url": "https://arxiv.org/abs/2605.28819"}, "sourceUpdatedAt": "2026-05-27T17:59:51Z", "topics": ["cs.LG"], "venueAttempts": [], "whyItMatters": "Existing PEFT evaluations focus mainly on downstream accuracy, overlooking retention of pretrained abilities. This benchmark provides a stability-plasticity perspective, enabling selection of fine-tuning methods that balance task adaptation and forgetting resistance, offering a more complete assessment for practical deployment decisions."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
