update: README.md
Browse files
README.md
CHANGED
@@ -85,6 +85,41 @@ print(predict_label)
|
|
85 |
# model output: 科技
|
86 |
```
|
87 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
88 |
## 引用 Citation
|
89 |
|
90 |
如果您在您的工作中使用了我们的模型,可以引用我们的[论文](https://arxiv.org/abs/2209.02970):
|
|
|
85 |
# model output: 科技
|
86 |
```
|
87 |
|
88 |
+
除了分类任务,其他任务的数据构造例子如下:
|
89 |
+
|
90 |
+
In addition to classification tasks, data construction examples of other tasks are as follows:
|
91 |
+
|
92 |
+
```python
|
93 |
+
example_dict={
|
94 |
+
"文本分类":{"text_a":"钢琴块3别踩白块儿3钢琴块3是一款简洁的钢琴模拟软件,在Android平台上,类似的软件还是比较多的。","choices":["相机","影视娱乐","棋牌中心","新闻","财经","策略","休闲益智","教育"]},
|
95 |
+
'新闻分类':{"text_a":"微软披露拓扑量子计算机计划!","choices":["故事","文化","娱乐","体育","财经","房产","汽车","教育","科技"]},
|
96 |
+
'情感分析':{"text_a":"刚买iphone13 pro 还不到一个月,天天死机最差的一次购物体验","choices":["好评","差评"]},
|
97 |
+
'意图识别':{"text_a":"打电话给吴小军。","choices":["放音乐","播放下一首","打电话","退出导航","开始导航","其他","暂停音乐","导航","开导航"]},
|
98 |
+
|
99 |
+
'语义匹配':{"text_a":"今天心情不好","text_b":"我很不开心","choices":["相似","不相似"]},
|
100 |
+
'自然语言推理':{"text_a":"小明正在上高中","text_b":"小明是一个初中生","choices":["无关","矛盾","蕴含"]},
|
101 |
+
|
102 |
+
'多项选择':{"text_a":"这大家千万不能着急,我们现在只是暂时输了7分。距离比赛结束还有20多分钟呢,我们是完全有机会转败为赢的,大家加油!","question":"说话人希望大家:","choices":["别得意","冷静一些","加快速度","提前预习"]},
|
103 |
+
'指代消解':{"text_a":"李鸣觉得董客这人,踏实得叫人难受。可因为孟野和森森太疯,他只好去找董客聊天,但在董客眼里,李鸣也是不正常,他竟然放着现成的大学不愿上。","question":"【他】指的是【李鸣】吗?","choices":["是","不是"]},
|
104 |
+
|
105 |
+
'实体识别':{"text_a":"北京大学是我国的一座历史名校,坐落在海淀区,蔡元培曾经担任校长","question":"机构"},
|
106 |
+
'抽取式阅读理解':{"text_a":"《H》正式定档3月7日下午两点整在京东商城独家平台开启第一批5000份预售,定价230元人民币,回馈最忠实的火星歌迷,意在用精品回馈三年来跟随华晨宇音乐不离不弃的粉丝们的支持与厚爱","question":"华晨宇专辑h预售价格是多少?"},
|
107 |
+
'关键词抽取':{"text_a":"今儿在大众点评,找到了口碑不错的老茶故事私房菜。"},
|
108 |
+
|
109 |
+
"生成式摘要":{"text_a":"针对传统的流量分类管理系统存在不稳定、结果反馈不及时、分类结果显示不直观等问题,设计一个基于web的在线的流量分类管理系统.该系统采用流中前5个包(排除3次握手包)所含信息作为特征值计算资源,集成一种或多种分类算法用于在线网络流量分类,应用数据可视化技术处理分类结果.实验表明:在采用适应在线分类的特征集和c4.5决策树算法做分类时,系统能快速做出分类,且精度达到94%以上;数据可视化有助于人机交互,改善分类指导."}
|
110 |
+
}
|
111 |
+
```
|
112 |
+
|
113 |
+
## 预训练或微调 prtrain or finetune
|
114 |
+
如果您对于怎么预训练Randeng-T5模型或者想在自己的下游任务中微调Randeng模型,欢迎使用[Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM/)项目,这里提供了完整的示例:
|
115 |
+
- [预训练](https://github.com/IDEA-CCNL/Fengshenbang-LM/tree/main/fengshen/examples/pretrain_t5)
|
116 |
+
- [微调](https://github.com/IDEA-CCNL/Fengshenbang-LM/tree/main/fengshen/examples/mt5_summary)
|
117 |
+
|
118 |
+
If you want to pre train the Randeng T5 model or fine tune the Randeng model in your downstream tasks, welcome to use [Fengshenbang LM]( https://github.com/IDEA-CCNL/Fengshenbang-LM/ )A complete example of the project is provided here:
|
119 |
+
|
120 |
+
- [Pre training](https://github.com/IDEA-CCNL/Fengshenbang-LM/tree/main/fengshen/examples/pretrain_t5)
|
121 |
+
- [Fine tune](https://github.com/IDEA-CCNL/Fengshenbang-LM/tree/main/fengshen/examples/mt5_summary)
|
122 |
+
|
123 |
## 引用 Citation
|
124 |
|
125 |
如果您在您的工作中使用了我们的模型,可以引用我们的[论文](https://arxiv.org/abs/2209.02970):
|