نسخة أولية وصول مفتوح
MMVistaReason: Toward Open-Data and Post-Training Recipes for Multimodal Reasoning
Open multimodal reasoning models have benefited from large-scale reasoning supervision, yet reliable post-training remains challenging due to uneven data quality, inefficient supervision construction, imbalanced difficulty, and cross-domain interference. We introduce MMVistaReason (MVR), an open-data post-training reci …