Model Selection Recommendation
#1
by ffy2000 - opened
Hi, thank you very much for your interest in Lance and for sharing the quantization workflow.
One suggestion from our side: for image-related tasks such as image generation, image understanding, and image editing, we recommend using Lance_3B rather than Lance_3B_Video whenever possible, as it can better preserve performance on these image-centric capabilities.
Although Lance_3B_Video is a more general model, its training is more focused on video-related performance. In contrast, Lance_3B is specifically trained and optimized for image tasks, so it is usually the better choice for image generation, understanding, and editing scenarios.
Thanks again for your attention and support for Lance!