https://github.com/xusenlinzy/fastie
A training and inference framework for open ner and re models! 信息抽取（实体抽取、关系抽取、事件抽取）模型的统一训练和推理框架，包含丰富的开源SOTA模型
https://github.com/xusenlinzy/fastie
event-extraction information-extraction named-entity-recognition natural-language-processing pretrained-models relation-extraction text-classification uie
Last synced: 10 days ago
JSON representation
A training and inference framework for open ner and re models! 信息抽取（实体抽取、关系抽取、事件抽取）模型的统一训练和推理框架，包含丰富的开源SOTA模型
- Host: GitHub
- URL: https://github.com/xusenlinzy/fastie
- Owner: xusenlinzy
- License: apache-2.0
- Created: 2024-09-02T09:17:51.000Z (about 1 year ago)
- Default Branch: master
- Last Pushed: 2024-12-31T02:00:41.000Z (10 months ago)
- Last Synced: 2025-09-20T17:35:44.273Z (22 days ago)
- Topics: event-extraction, information-extraction, named-entity-recognition, natural-language-processing, pretrained-models, relation-extraction, text-classification, uie
- Language: Python
- Homepage:
- Size: 12.7 MB
- Stars: 13
- Watchers: 1
- Forks: 3
- Open Issues: 5
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# FastIE
此项目为开源**文本分类、实体抽取、关系抽取和事件抽取**[模型](MODELS.md)的训练和推理提供统一的框架，具有以下特性
+ ✨ 支持多种开源文本分类、实体抽取、关系抽取和事件抽取模型
+ 👑 支持百度 [UIE](https://github.com/PaddlePaddle/PaddleNLP) 模型的训练和推理
+ 🚀 统一的训练和推理框架
+ 🎯 集成对抗训练方法，简便易用
## 📢 更新日志
+ 【2024.8.30】 发布初始版本
---
## 📦 快速安装
源码安装
```shell
git clone https://github.com/xusenlinzy/FastIE.git
pip install -e .
```
docker启动
```shell
docker build -t fastie .
docker compose up -d
docker exec -it fastie bash
```
## 🚀 模型训练
### 实体抽取
```shell
cd examples/named_entity_recognition
fastie-cli train global_pointer.yaml
```
具体参数详见 [named_entity_recognition](./examples/named_entity_recognition)
### 关系抽取
```shell
cd examples/relation_extraction
fastie-cli train gplinker.yaml
```
具体参数详见 [relation_extraction](./examples/relation_extraction)
### 事件抽取
```shell
cd examples/event_extraction
fastie-cli train gplinker.yaml
```
具体参数详见 [event_extraction](./examples/event_extraction)
## 📊 模型推理
本项目实现了对各类模型推理代码的封装，只需要4行代码即可推理！
```python
from transformers import AutoModel, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("path_to_model", trust_remote_code=True)
model = AutoModel.from_pretrained("path_to_model", trust_remote_code=True)
print(model.predict(tokenizer, "因肺过度充气，常将肝脏推向下方。"))
```
一键启动模型接口或DEMO
```shell
# fastie-cli api --model_name_or_path path_to_model --port 9000
fastie-cli demo --model_name_or_path path_to_model --port 9000 --device cuda
```
## 致谢
本项目受益于 [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory)，感谢以上诸位作者的付出。