update textbook link (#427)

chapters/en/chapter7/6.mdx

@@ -36,7 +36,7 @@ This is actually showcasing the model that was trained and uploaded to the Hub u
 
 ## Gathering the data[[gathering-the-data]]
 
-Python code is abundantly available from code repositories such as GitHub, which we can use to create a dataset by scraping for every Python repository. This was the approach taken in the [Transformers textbook](https://learning.oreilly.com/library/view/natural-language-processing/9781098103231/) to pretrain a large GPT-2 model. Using a GitHub dump of about 180 GB containing roughly 20 million Python files called `codeparrot`, the authors built a dataset that they then shared on the [Hugging Face Hub](https://huggingface.co/datasets/transformersbook/codeparrot).
+Python code is abundantly available from code repositories such as GitHub, which we can use to create a dataset by scraping for every Python repository. This was the approach taken in the [Transformers textbook](https://learning.oreilly.com/library/view/natural-language-processing/9781098136789/) to pretrain a large GPT-2 model. Using a GitHub dump of about 180 GB containing roughly 20 million Python files called `codeparrot`, the authors built a dataset that they then shared on the [Hugging Face Hub](https://huggingface.co/datasets/transformersbook/codeparrot).
 
 However, training on the full corpus is time- and compute-consuming, and we only need the subset of the dataset concerned with the Python data science stack. So, let's start by filtering the `codeparrot` dataset for all files that include any of the libraries in this stack. Because of the dataset's size, we want to avoid downloading it; instead, we'll use the streaming feature to filter it on the fly. To help us filter the code samples using the libraries we mentioned earlier, we'll use the following function: