Pairing VL-PRMs trained with abstract reasoning problems results in strong generalization and reasoning performance improvements when used with strong vision-language models in test-time scaling ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback