dongx1997/TokenSwap-Bench

TokenSwap: Benchmarking and Reducing the Modality Gap in Multimodal LLMs Andong Hua1, Colton Bishop2, Igor Mordatch2, Arian Hosseini2, Jindong Gu2, Aleksandra Faust2, Rebecca Roelofs2, Yao Qin1,2 1University of California, Santa Barbara    2Google DeepMind NeurIPS 2026 Across 42 MLLMs, we observe a pervasive modality gap, with performance decreasing by 4.2% to 47.4% when moving from text-only to image-interleaved inputs. TokenSwap-Bench… See the full description on the dataset page: https://huggingface.co/datasets/dongx1997/TokenSwap-Bench.

<iframe src="https://silytics.si/embed/item/56b664cc-e1f7-45f3-aa02-18def57ed5db" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.