fighting41love/funnlp

funNLP is a large index of Chinese natural-language-processing (NLP) resources collected in one place. It is not a single program but a curated list. Each entry points to a tool, a dataset, a model, a paper, or a piece of code that is useful when working with Chinese text. The README is itself the catalogue, organised by what each item does. The author describes it as the playground for NLP workers, and notes that it is updated irregularly. The top section, which has been growing fastest, is about ChatGPT-style large language models: evaluations and comparisons, background reading, open-source frameworks, training and low-resource fine-tuning, prompt engineering, document question answering, industry applications, course material, safety issues, multi-modal LLMs, and LLM datasets. The wider collection covers the bread and butter of Chinese NLP. There are dictionaries and word lists for sensitive words, stopwords, synonyms and antonyms, idioms, place names, historical figures, medical terms, legal terms, surname databases for Chinese and Japanese, and traditional-to-simplified conversion. There are extractors for common pieces of information such as phone numbers, ID numbers, email addresses, and inferring gender from a name. There are task-specific tools for Chinese word segmentation, named-entity recognition, sentiment analysis, summarisation, keyword extraction, OCR for handwritten Chinese, speech recognition, text-to-SQL, and question answering. It also indexes resources for the deep-learning side of the field: pretrained models such as BERT, ALBERT, ELECTRA and GPT-2 variants for Chinese, knowledge-graph projects in medicine, finance, and law, dialog-system frameworks like Rasa, and benchmark suites and corpora for training and evaluation. Many entries link to Python packages or training code, which is why the repository language tag is Python. Someone would use funNLP as a starting point for a Chinese-language project: to find the right library before writing one from scratch, to discover labelled datasets, or to keep up with the field.