Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GPT-2 Fine-tuning in Vietnamese News
|
| 2 |
+
## Model description
|
| 3 |
+
A Fine-tuned Vietnamese GPT2 model which can generate Vietnamese news based on context (category + headline), based on the Vietnamese Wiki GPT2 pretrained model (https://huggingface.co/danghuy1999/gpt2-viwiki)
|
| 4 |
+
|
| 5 |
+
## Purpose
|
| 6 |
+
This model was made only for fun and experimental study. However, It gives impressive results
|
| 7 |
+
Most of the generative news are fake with unconfirmed information. Honestly, I feel fun about this project =))
|
| 8 |
+
|
| 9 |
+
## Dataset
|
| 10 |
+
The dataset is about 30k Vietnamese news dataset from website thanhnien.vn
|
| 11 |
+
|
| 12 |
+
## Result
|
| 13 |
+
- Train Loss: 2.3
|
| 14 |
+
- Val loss: 2.5
|
| 15 |
+
- Rouge F1: 0.556
|
| 16 |
+
- Word error rate: 1.08
|
| 17 |
+
|
| 18 |
+
## Deployment
|
| 19 |
+
- You can run the model deployment in this Colab's [link](https://colab.research.google.com/drive/1ITnYPnngd_aqkFB2A5IhzSsX4jQSPOR1?usp=sharing)
|
| 20 |
+
- Then go to this link: https://gptvn.loca.lt
|
| 21 |
+
- You can choose any categories and give it some text for the headline, then generate. There we go
|
| 22 |
+
- P/s: I've already tried to deploy my model on Streamlit's cloud, but It was always being broken due to out of memory
|
| 23 |
+
|
| 24 |
+
## Github
|
| 25 |
+
- https://github.com/Tuan-Lee-23/Vietnamese-News-Generative-Model/blob/main/
|