Update README.md
Browse files
README.md
CHANGED
@@ -22,6 +22,7 @@ The Variational Autoencoder (VAE) network comprises an encoder using Bert with 1
|
|
22 |
* Wudao Corpus (with 280G samples)
|
23 |
|
24 |
**防止后验崩塌 Avoiding posterior collapse**
|
|
|
25 |
为了防止在通用语料上后验崩塌,我们在训练中加入以下措施,
|
26 |
1. 使用KL annealing。对于正则项系数,采用梯形scheduler
|
27 |
2. 加入free bits。设置free bits,避免过分靠近先验
|
@@ -39,6 +40,7 @@ We used several methods to avoid posterior collapse, as what follows,
|
|
39 |
|
40 |
|
41 |
## 使用 Usage
|
|
|
42 |
```shell
|
43 |
git clone https://github.com/IDEA-CCNL/Fengshenbang-LM.git
|
44 |
cd Fengshenbang-LM
|
|
|
22 |
* Wudao Corpus (with 280G samples)
|
23 |
|
24 |
**防止后验崩塌 Avoiding posterior collapse**
|
25 |
+
|
26 |
为了防止在通用语料上后验崩塌,我们在训练中加入以下措施,
|
27 |
1. 使用KL annealing。对于正则项系数,采用梯形scheduler
|
28 |
2. 加入free bits。设置free bits,避免过分靠近先验
|
|
|
40 |
|
41 |
|
42 |
## 使用 Usage
|
43 |
+
|
44 |
```shell
|
45 |
git clone https://github.com/IDEA-CCNL/Fengshenbang-LM.git
|
46 |
cd Fengshenbang-LM
|