EvenVLM: 50K-entry multimodal dataset and fine-tuned VLM for endangered Even language

Daniil Makovetskii · hf · 2026-11-17

EvenVLM on Hugging Face provides a text and multimodal VQA dataset (50K entries, 107K examples) for the endangered Even language, with tokenizer expansion and LoRA fine-tuning cutting token fragmentation by 27% and improving translation and VQA accuracy; presented at an EMNLP 2026 workshop.

Original post →

More from Research

Research channel →