Matthieu Le

Eagle 2.5: boosting long-context post-training for frontier vision-language models