ModelScope
AI model hub with an Apache 2.0 Python library for local inference, training and evaluation, plus hosted demos and cloud notebooks.
ModelScope combines a hosted model and dataset hub with a Python library you can run locally. It's for developers and researchers who want to use AI models in their own applications, fine-tune them on their own data, or compare their performance. The library is open source under Apache 2.0.
Its scope extends beyond local LLM work. Models cover computer vision, speech, multimodal tasks and scientific computing, including text generation, image editing, speech recognition and protein structure prediction. The hub includes Qwen and DeepSeek models, along with projects such as Meta-Llama-3-8B-Instruct and LaMa image inpainting.
The library provides a shared interface for inference, training and evaluation across these different tasks. It supports PyTorch, TensorFlow and ONNX, with modular components that developers can customize for their applications. Support also covers model export, deployment and distributed training, including data and model parallel approaches.
The local library and hosted services have distinct roles. You can download public models for use in a local Python environment; the website provides online demos, Studios for displaying AI applications, and cloud notebooks backed by CPU or GPU environments. Those browser experiences run on hosted infrastructure.
The library connects to the model and dataset hubs for discovery, version management and caching. Most models on the hub are public and available for direct download.
