This repository serves as a curated collection of high-quality Chinese pre-trained NLP models, including large language models (LLMs), multimodal models, and their associated resources. It aims to provide researchers and developers with a centralized hub for discovering and accessing state-of-the-art models for Chinese natural language processing tasks.

How It Works

The project meticulously gathers and organizes information on a vast array of Chinese NLP models, categorizing them by architecture (e.g., BERT, GPT, T5, RoFormer), domain (e.g., general, finance, medical, code), and modality (text-only, multimodal). It provides links to Hugging Face, model repositories, papers, and project pages, facilitating easy access and evaluation.

Quick Start & Requirements

• Models are primarily accessed via Hugging Face (🤗HF) or ModelScope.
• Installation typically involves using libraries like transformers or modelscope.
• Specific hardware requirements (e.g., GPU, VRAM) depend on the model size and complexity.
• Links to official documentation, demos, and quick-start guides are provided for individual models.

Highlighted Details

• Comprehensive coverage of foundational NLP models (BERT, RoBERTa, etc.) and modern LLMs.
• Extensive lists of Chinese-specific models, including domain-specific and multimodal variants.
• Curated resources for datasets, evaluation benchmarks (e.g., C-Eval, FlagEval), and related "awesome" lists.
• Regular updates to include the latest models and research advancements.

Maintenance & Community

• The project is community-driven, with contributions from various institutions and individuals.
• Links to Hugging Face, ModelScope, and other relevant platforms are provided for community engagement.

Licensing & Compatibility

• Model licenses vary, with many available under permissive licenses (e.g., Apache 2.0) allowing commercial use.
• Users should verify the specific license of each model before deployment.

Limitations & Caveats

• The sheer volume of models means direct evaluation of each is impractical; users must assess suitability for their specific needs.
• Access to some models may require registration or application.