AI model inference training

VastGaussian is an open source project for 3D scene reconstruction, which simulates the geometry and appearance information of large scenes by using 3D Gaussian. This project was implemented by the author from scratch and may have some errors, but it provides a new attempt in the field of 3D scene reconstruction. The main advantages of the project include the ability to handle large data sets, as well as improvements to the original 3DGS project to make it easier to understand and use.

Vary-toy is a small Vary model based on Qwen-1.8B as the basic "big" language model. Vary-toy introduces an improved visual vocabulary, giving the model not only all the features of Vary, but also broader generalization capabilities. Specifically, in the process of generating visual vocabulary, we replace the negative samples of natural images with positive sample data driven by object detection, more fully utilizing the capacity of the vocabulary network, allowing it to efficiently encode visual information corresponding to natural objects. In experiments, Vary-toy achieved 65.6% ANLS on DocVQA, 59.1% accuracy on ChartQA, 88.1% accuracy on RefCOCO, and 29% accuracy on MMVet. Pricing: Free trial, paid version pricing to be determined. Positioning: Provide researchers with solutions to train and deploy LVLMs on ordinary GPUs with limited resources.