Does UI-TARS-7B-DPO have official multilingual benchmarks (en/ru/zh/etc)?
#8
by
unstopppable
- opened
The model bytedance-research/UI-TARS-7B-DPO seems to support English, Russian, and Chinese (with varying quality based on informal testing). Is there any official data or benchmarks on its multilingual performance? Specifically:
Supported languages: Is there a full list?
Per-language metrics: Accuracy, fluency, or task-specific scores (e.g., MMLU, FLORES)?
DPO impact: Does preference optimization favor certain languages?
Any links to papers, GitHub docs, or internal stats would be appreciated!
unstopppable
changed discussion title from
Does UI-TARS-7B-DPO have official multilingual benchmarks (en/ru/zh)?
to Does UI-TARS-7B-DPO have official multilingual benchmarks (en/ru/zh/etc)?