MLLM-CITBench: Multimodal Task Benchmark with 7 Challenge Domains

MLLM-CITBench provides a benchmark for evaluating multimodal large language models across seven distinct task domains. The dataset includes independent training and test splits for tasks in OCR, art, financial policy, math, medicine, numerical reasoning, and science. It was created by yueluoshuangtian and last updated on Hugging Face in October 2025.

Use Cases

• Benchmarking OCR performance on images containing text, based on the OCR task description
• Evaluating model reasoning on financial and monetary policy documents, based on the FOMC task
• Testing numerical and mathematical problem-solving capabilities, based on the math and numglue tasks
• Assessing performance on medical and scientific domain knowledge, based on the medical and science tasks
• Training or fine-tuning models for art-related visual understanding, based on the art task

Strengths

• Covers seven distinct and challenging multimodal task domains for comprehensive evaluation
• Provides independent training and test splits for each task to support standard ML workflows
• Last updated on 2025-10-24, suggesting recent maintenance

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Row count and dataset size are unknown, which may limit suitability assessment

Provenance

Source

huggingface

Freshness

Last updated 2025-10-24 04:41:01