Update README.md

Updating the first version of readme.

README.md

@@ -1,3 +1,10 @@
 ---
 license: afl-3.0
 ---
+This project page is about the pytorch code implementation of GlyphBERT by the HITsz-TMG research group.
+
+GlyphBERT is a Chinese pre-training model that includes Chinese character glyph features.It renders the input characters into images and designs them in the form of multi-channel location feature maps, and designs a two-layer residual convolutional neural network module to extract the image features of the characters for training.
+
+The experimental results show that the performance of the pre-training model can be well improved by fusing the features of Chinese glyphs. GlyphBERT is much better than BERT in multiple downstream tasks, and has strong transferability.
+
+For more details about using it, you can check the [github repo](https://github.com/HITsz-TMG/GlyphBERT)