Macbert: Load a pretrained model for Chinese NLP

Understand the project

Practical overview

Use MacBERT when a Chinese NLP project needs a BERT-style pretrained checkpoint and the team plans to build or fine-tune the downstream task system.

Load a pretrained Chinese language model as a starting point for NLP development
Fine-tune a checkpoint for text classification or named-entity recognition
Adapt a pretrained backbone for sentence-pair tasks

Before you choose

The repository identifies the project as open source under the Apache-2.0 license. It lists base and large checkpoints for download.

The base checkpoint has 102 million parameters and a listed size of 383 MB; the large checkpoint has 324 million parameters and a listed size of 1.2 GB.
The documented loading path uses Hugging Face Transformers with BertTokenizer and BertModel.
MacBERT's pretraining substitutes similar words for [MASK] tokens and also applies whole-word and N-gram masking.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Load the documented base or large MacBERT model using the BERT tokenizer and model classes.

Choose the base checkpoint, hfl/chinese-macbert-base, or the large checkpoint, hfl/chinese-macbert-large.
Load the tokenizer with BertTokenizer.from_pretrained using the chosen model name.
Load the model with BertModel.from_pretrained using the same model name.

The selected MacBERT tokenizer and model are loaded through the documented Transformers interface.

The README's loading example covers model and tokenizer loading; it does not provide an end-to-end downstream fine-tuning recipe.

Evidence and transparency

Source record details

Source availability: Active
Sources checked: 14 October 2026
Original website title: ymcui/MacBERT · Official homepage title · retained in its source language