Project Icon

TinyLLaVA_Factory

Modular Codebase for Developing Compact Large Multimodal Models

Product DescriptionTinyLLaVA Factory provides an open-source, modular framework for developing small-scale large multimodal models with ease. Built on PyTorch and HuggingFace, the platform emphasizes user-friendly implementation, feature expansion, and reliable training reproducibility. This codebase simplifies model customization and reduces coding errors, integrating advanced tools and models including CLIP, SigLIP, and Qformer. It supports diverse training methods such as LoRA tuning and offers updates like a visualization tool for prediction analysis, enhancing its functionality to effectively compete with larger models.
Project Details