{"paper":{"title":"LMMs-Eval: Reality Check on the Evaluation of Large Multimodal Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Evaluating large multimodal models requires balancing wide task coverage, low computational cost, and zero data contamination in benchmarks.","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Bo Li, Chunyuan Li, Fanyi Pu, Jingkang Yang, Joshua Adrian Cahyono, Kaichen Zhang, Kairui Hu, Peiyuan Zhang, Shuai Liu, Yuanhan Zhang, Ziwei Liu","submitted_at":"2024-07-17T17:51:53Z","abstract_excerpt":"The advances of large foundation models necessitate wide-coverage, low-cost, and zero-contamination benchmarks. Despite continuous exploration of language model evaluations, comprehensive studies on the evaluation of Large Multi-modal Models (LMMs) remain limited. In this work, we introduce LMMS-EVAL, a unified and standardized multimodal benchmark framework with over 50 tasks and more than 10 models to promote transparent and reproducible evaluations. Although LMMS-EVAL offers comprehensive coverage, we find it still falls short in achieving low cost and zero contamination. To approach this e"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our work highlights the importance of considering the evaluation trilemma and provides practical solutions to navigate the trade-offs in evaluating large multi-modal models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the live data sources and pruning rules in LMMS-EVAL LITE and LIVEBENCH truly deliver zero contamination and maintained coverage without introducing new selection biases or missing important capabilities.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LMMS-EVAL delivers a standardized multimodal evaluation framework with lite and live variants that target the trade-offs among coverage, cost, and zero contamination.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Evaluating large multimodal models requires balancing wide task coverage, low computational cost, and zero data contamination in benchmarks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e3d49ea222040ef472c564e264ba79f6f27b0de3ed651315426759c8b1ffb32d"},"source":{"id":"2407.12772","kind":"arxiv","version":2},"verdict":{"id":"aea1499b-8ccf-4737-bb8a-8fa3e282bc8c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T05:15:22.504049Z","strongest_claim":"Our work highlights the importance of considering the evaluation trilemma and provides practical solutions to navigate the trade-offs in evaluating large multi-modal models.","one_line_summary":"LMMS-EVAL delivers a standardized multimodal evaluation framework with lite and live variants that target the trade-offs among coverage, cost, and zero contamination.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the live data sources and pruning rules in LMMS-EVAL LITE and LIVEBENCH truly deliver zero contamination and maintained coverage without introducing new selection biases or missing important capabilities.","pith_extraction_headline":"Evaluating large multimodal models requires balancing wide task coverage, low computational cost, and zero data contamination in benchmarks."},"references":{"count":24,"sample":[{"doi":"","year":null,"title":"InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning","work_id":"f3aac728-ded0-4e55-aa9e-4a1635d4313d","ref_index":1,"cited_arxiv_id":"2305.06500","is_internal_anchor":true},{"doi":"","year":null,"title":"Internlm-xcomposer2- 4khd: A pioneering large vision-language model handling resolutions from 336 pixels to 4k hd","work_id":"799b1f12-2edc-4b09-a83b-dbd87e1404e7","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models","work_id":"806d2e73-71b3-4d56-87e0-39d571cc15d6","ref_index":3,"cited_arxiv_id":"2306.13394","is_internal_anchor":true},{"doi":"","year":2023,"title":"Making llama see and draw with seed tokenizer","work_id":"9453feaf-a19d-4603-b9de-da50593c5d43","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"A diagram is worth a dozen images.ArXiv, abs/1603.07396","work_id":"3cc7177d-7f05-4d4d-8066-04b7d589017f","ref_index":5,"cited_arxiv_id":"1603.07396","is_internal_anchor":true}],"resolved_work":24,"snapshot_sha256":"8259cdf1001bbc4b76ac39e2a416d1dc500d9bc07d1609c7bd56eafeaa80ae19","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"0fe42ac1f960edabb5776079e4c5fb9f1ed2d110b322762081d32bcc4709db70"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}