File size: 983 Bytes
9b086ad c1ce8d9 996025f c1ce8d9 996025f c1ce8d9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 |
---
license: mit
---
现有的少数民族语言预训练模型仍然较为稀缺，尽管国内少数民族语言模型CINO具有较强的理解能力，但仍然缺乏面向生成与翻译领域的研究。
CMPT (Chinese Minority Pre-Trained Language Model) 是在BART的基础上，加入DeepNorm预训练的超深层生成模型。其最大具有128+128层。其在超过10G的汉英维藏蒙语料中进行受限预训练。其具有较强的理解与生成性能。
**Github Link:** https://github.com/WENGSYX/CMPT
## Usage
```python
>>> from modeling_cmpt import BartForConditionalGeneration
>>> from transformers import AutoTokenizer
>>> tokenizer = AutoTokenizer.from_pretrained('./CMTP')
>>> model = BartForConditionalGeneration.from_pretrained('./CMTP')
>>> inputs = tokenizer.encode("Hello world, 你好 世界", return_tensors='pt')
>>> pred_ids = model.generate(input_ids, num_beams=4, max_length=20)
>>> print(tokenizer.convert_ids_to_tokens(pred_ids[i]))
```
|