Valley-Eagle-7B is a multimodal large model developed by ByteDance, designed to handle a variety of tasks involving text, image, and video data. The model has achieved top results in internal e-commerce and short video benchmark tests and has demonstrated outstanding performance in OpenCompass tests compared to models of similar scale. Valley-Eagle-7B incorporates a combination of LargeMLP and ConvAdapter to build its projector, and introduces a VisionEncoder to enhance performance in extreme scenarios.