---
language:
  - zh
thumbnail: https://ckip.iis.sinica.edu.tw/files/ckip_logo.png
tags:
  - pytorch
  - lm-head
  - albert
  - zh
license: gpl-3.0
---

# CKIP ALBERT Tiny Chinese

## Table of Contents
- [Model Details](#model-details)
- [Uses](#uses)
- [Risks, Limitations and Biases](#risks-limitations-and-biases)
- [Training](#training)
- [Evaluation](#evaluation)
- [How to Get Started With the Model](#how-to-get-started-with-the-model)

## Model Details
- **Model Description:**

This project provides traditional Chinese transformers models (including ALBERT, BERT, GPT2) and NLP tools (including word segmentation, part-of-speech tagging, named entity recognition).

這個專案提供了繁體中文的 transformers 模型（包含 ALBERT、BERT、GPT2）及自然語言處理工具（包含斷詞、詞性標記、實體辨識）。

- **Developed by:** [Mu Yang](https://muyang.pro) at [CKIP](https://ckip.iis.sinica.edu.tw)
- **Model Type:** Fill-Mask
- **Language(s):** Chinese
- **License:** gpl-3.0
- **Parent Model:** See the [ALBERT base model](https://huggingface.co/albert-base-v2) for more information about the ALBERT base model.
- **Resources for more information:**
  - [GitHub Repo](https://github.com/ckiplab/ckip-transformers)
  - [CKIP Documentation](https://ckip-transformers.readthedocs.io/en/stable/)



## Uses

#### Direct Use

The model author suggests using BertTokenizerFast as tokenizer instead of AutoTokenizer.

請使用 BertTokenizerFast 而非 AutoTokenizer。

For full usage and more information, please refer to [github repository]  (https://github.com/ckiplab/ckip-transformers.)

有關完整使用方法及其他資訊，請參見  [github repository]  (https://github.com/ckiplab/ckip-transformers.)



## Risks, Limitations and Biases
**CONTENT WARNING: Readers should be aware this section contains content that is disturbing, offensive, and can propagate historical and current stereotypes.**

Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al. (2021)](https://aclanthology.org/2021.acl-long.330.pdf) and [Bender et al. (2021)](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)).


## Training


#### Training Data

The language models are trained on the ZhWiki and CNA datasets; the WS and POS tasks are trained on the ASBC dataset; the NER tasks are trained on the OntoNotes dataset.
以上的語言模型訓練於 ZhWiki 與 CNA 資料集上；斷詞（WS）與詞性標記（POS）任務模型訓練於 ASBC 資料集上；實體辨識（NER）任務模型訓練於 OntoNotes 資料集上。

#### Training Procedure
* **Parameters:** 4M 



## Evaluation




#### Results



* **Perplexity:** 4.40
* **WOS (Word Segmentation) [F1]:** 96.66%
* **POS (Part-of-speech) [ACC]:** 94.48%
* **NER (Named-entity recognition) [F1]:** 71.17% 

## How to Get Started With the Model

```
from transformers import (
  BertTokenizerFast,
  AutoModel,
)

tokenizer = BertTokenizerFast.from_pretrained('bert-base-chinese')
model = AutoModel.from_pretrained('ckiplab/albert-tiny-chinese')
```