In March 2026, Yufan Ren developed the Multi-Level Existence Benchmark (MLE-Bench) for the EvolvingLMMs-Lab/lmms-eval repository, targeting fine-grained visual perception evaluation in multimodal models. Leveraging Python and data processing techniques, Yufan designed a system that categorizes object existence by size, supporting comprehensive and subset-based evaluation tasks. The implementation included utilities for processing results and metrics, streamlining the benchmarking workflow. This work aligned with the ICLR 2026 Oral paper and associated datasets, providing a robust framework for model assessment. Yufan’s contribution demonstrated depth in machine learning and Python development, enabling more rigorous model selection and reporting practices.