bert-base-chinese

bert-base-chinese

Predicts masked words in Chinese text, as Google's BERT base model pre-trained for Chinese. It suits developers and researchers who build Chinese language tools or fine-tune a model for a Chinese task.

Made by google-bert on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: bert-base-chinese is Google's BERT base model, pre-trained for Chinese. In training, some words were hidden at random, and the model learned to guess them from the words around them.

The card lists masked language modeling as its direct use. It builds on the BERT base model, which the card describes in a separate page.

Who it suits: Developers, researchers and students who work with Chinese text suit it best. Many people start from this model and fine-tune it for a task of their own.

Industry | Who uses it | How they use it | Example
Chinese NLP | Developers | Build on a Chinese language model | Predict a missing word in a sentence
Research | NLP researchers | Use it as a starting point for fine-tuning | Train a version for one Chinese task
Education | Students | Learn how masked language models work | Load the model and try a few sentences

Anyone who works only in other languages can skip it, because it was pre-trained for Chinese.

Where to use it: You run it in Python on your own computer with the transformers library.

• Windows, macOS or Linux with Python installed
• Python scripts that use the transformers library
• Fine-tuning projects that start from a Chinese BERT

Use case | Who | What they get back
Masked word prediction | Developers | Candidate words for a gap in a Chinese sentence
Fine-tuning base | Researchers | A pre-trained Chinese encoder to adapt
Learning and demos | Students | A working example of fill-mask

How to set it up: You need Python and the transformers library. The model downloads when you load it by name.

• Python on Windows, macOS or Linux
• The transformers library
• Disk space for about 412 MB of weights in F32

1. Install Python.
2. Install the transformers library with pip.
3. Optionally download the model files ahead of time with the hub tools.
4. Load the tokenizer and the model by name.

pip install -U huggingface_hub
hf download google-bert/bert-base-chinese

How to use it: Load the tokenizer and the masked language model by name. The card shows these two steps and nothing more.

from transformers import AutoTokenizer, AutoModelForMaskedLM

tokenizer = AutoTokenizer.from_pretrained("bert-base-chinese")

model = AutoModelForMaskedLM.from_pretrained("bert-base-chinese")

Tool | What it does
AutoTokenizer | Splits Chinese text into word pieces for the model
AutoModelForMaskedLM | Loads the model that predicts masked words

Example: You give the model a Chinese sentence with one word hidden. It proposes words that could fill the gap.

For newcomers and experts: Think of the model as a reader that has practiced filling blanks in Chinese sentences. Your first small try is to run the two loading lines above and check that both load.

Under the hood, it is a 12-layer BERT encoder with a vocabulary of 21128 entries. Masking was applied to word pieces independently, as in the original BERT paper. The record lists 103M parameters.

A Chinese sentence with a hidden word
The tokenizer splits it into word pieces
The BERT model reads the whole sentence
Candidate words for the gap come out

Good to know: The card leaves training data and evaluation results marked as more information needed, and it carries a content warning about bias.

• Size: 103M parameters, about 412 MB in F32, stored as Safetensors
• Language: Chinese
• Other versions: Hugging Face lists 4 quantized and 271 fine-tuned versions made by other people
• License: apache-2.0
• Cost to run: none named
• Data it reaches: the text you give it, and the model files downloaded from Hugging Face
• Not for: anything beyond masked language modeling, which is the only direct use the card names

On Hugging Face, 13 November 2026

• Task: fill mask
• Library: transformers
• Parameters: 103M
• Weights: about 412 MB in F32; about 51 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 353,460
• Likes: 1,502
• License: apache-2.0
• Languages: zh
• Quantized versions: 4
• Fine-tunes: 275
• Tags: transformers, pytorch, safetensors, bert, fill-mask

Links

• Model: https://huggingface.co/google-bert/bert-base-chinese
• Files: https://huggingface.co/google-bert/bert-base-chinese/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download google-bert/bert-base-chinese

Then follow the usage in its model card. Nothing in this kit runs by itself.