diff --git a/readme.md b/readme.md
index 2fd61e042f953e89f1073f7ed898da84bbfb619b..ea8225fb61393c1b7b950f566e491b0a477aa628 100755
--- a/readme.md
+++ b/readme.md
@@ -98,7 +98,7 @@ The output is `{output_path}/scores.txt`, a line of a list of scores, following
 Note: *In the released codes, for simplicity, we did not expose the raw API , which supports some advanced generation modes, e.g. text and part of image.*
 
 ## Training
-Here we use a subset of our dataset from bird-and-animal for tutorial.
+Here we use a subset of our dataset from bird-and-animal for tutorial. The binary dataset is generated by our [cogdata toolkit](https://github.com/Sleepychord/cogdata). Please wait for a formal release with tutorials of cogdata (although it is available now).
 ### Single Node 
 After downloading the dataset, directly run
 ```