Awesome Pretrained Chinese NLP Models is a curated collection of pretrained language models, large language models, and multimodal models for Chinese natural language processing.
The repository addresses the challenge of discovering and accessing high-quality Chinese NLP models by aggregating publicly available pretrained models across multiple categories. It organizes models by type, including general-purpose large language models with parameters exceeding 7 billion, domain-specific models for finance, medicine, and law, conversational models, multimodal models combining text and images, and reasoning-focused models. The collection also covers traditional pretrained architectures organized by their primary function: NLU models like BERT and RoBERTa variants, NLG models including GPT and T5 variants, combined NLU-NLG architectures, and multimodal models. Beyond model listings, the repository provides links to evaluation benchmarks, open-source model repositories, Chinese instruction datasets, and embedding resources.
Developers building Chinese NLP applications should use this repository as a reference guide when selecting models for their specific use case. It suits projects requiring domain-specific models in vertical areas like finance or healthcare, as well as those needing general conversational or reasoning capabilities. The repository is particularly valuable for teams working within China, as it includes recommendations for accessing HuggingFace models through domestic mirror services to improve download speeds.
The project maintains active curation with continuous updates to reflect new model releases. Documentation is organized into separate detailed files for each model category, allowing developers to explore full model lists beyond the summary tables. The repository includes structured metadata about model architectures, distinguishing between non-causal decoders, causal decoders, and encoder-decoder configurations to help with technical selection decisions.
