how rel_component works?

I am doing the relation extraction task, I saw the video you posted on youtube(SPACY v3: Custom trainable relation extraction component - YouTube) and completed the relation extraction task according to this tutorial(, but I don't quite understand the process. I see that your model consists of two parts, one create_instance_tensor, and one classification_layer. Can you give an example to illustrate the input and output of each part? like the picture below, Thank you.

hi @Zanejins!

Thanks for your comment and interest in the relations extractions tutorial.

The team is in the process of writing out a blog post that explains more of the tutorial with additional code snippets. We're hoping to publish that soon and I'll post back once it's up.

For example, here's the image you're referring to, which we'll describe more in the post:

hi @Zanejins!

Sorry for the delay.

Sofie and Victoria released their relations extractions blog late last week:

Hope this helps provide more context on how rel_component works!