Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
使用预训练语言模型ALBERT做中文NER
| Date | Stars |
|---|---|
| 2026-07-24 | 476 |
| 2026-07-25 | 476 |
| 2026-07-28 | 476 |
| 2026-07-30 | 476 |
| 2026-08-06 | 476 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# albert-chinese-ner ## 前言 这次的albert某种程度上可能比bert本身更具有意义，恰逢中文预训练模型出来，还是按照之前的数据来做NER方面的fine-tune PS: 移步传统[**bert ner**](https://github.com/ProHiryu/bert-chinese-ner)模型 ## Resources - [Bert](https://github.com/google-research/bert) - [ALBert](https://github.com/google-research/albert) - [ALBert_zh](https://github.com/brightmart/albert_zh) ## Papers - [ALBERT](https://arxiv.org/pdf/1909.11942.pdf) ## 配置 1. 下载albert中文模型，这里使用的是base 2. 将模型文件夹重命名为albert_base_zh，放入项目中 3. 运行 ```bash python albert_ner.py --task_name ner --do_train true --do_eval true --data_dir data --vocab_file ./albert_config/vocab.txt --bert_config_file ./albert_base_zh/albert_config_base.json --max_seq_length 128 --train_batch_size 64 --learning_rate 2e-5 --num_train_epochs 3 --output_dir albert_base_ner_checkpoints ``` 4.最好使用tensorflow > 1.13, 这里运行的是1.15，不支持tf2.0 ## 结果 Base模型下训练3个epoch后： ```bash INFO:tensorflow: eval_f = 0.9280548 INFO:tensorflow: eval_precision = 0.923054 INFO:tensorflow: eval_recall = 0.9331808 INFO:tensorflow: global_step = 2374 INFO:tensorflow: loss = 13.210413 ``` 测试结果同样： ``` [CLS] B-LOC I-LOC O B-LOC I-LOC I-PER O O O O O O O O O [SEP] [CLS] ``` ## 总结 比起Bert本体，模型确实小了很多，效果却基本相当甚至领先bert，训练时间大幅缩小，NLP的“大舰巨炮”时代可能真的要过去了
Excerpt of 1,247 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:7aa5900d27cbb59a, topic:tensorflow