Update 2022-2-8-quantization-in-practice.md

Author: subramen

_posts/2022-2-8-quantization-in-practice.md

@@ -460,11 +460,9 @@ for key in act_compare_dict:
   - If the above workflow didn't work for you, we want to know more. Post a thread with details of your code (model architecture, accuracy metric, techniques tried). Feel free to cc me [@suraj.pt](https://discuss.pytorch.org/u/suraj.pt/).
 
 
-If you stuck with it this far, congrats! You can confidently claim to be abreast with practical quantization with PyTorch. 
-In the next post, we'll take a look at quantizing a "real-world" model that is chock full of loops and if-else blocks, and uses third-party libraries.
-Dynamic control flow structures make model building with PyTorch really convenient, but they disallow symbolic tracing a model. This makes it hard to directly quantize a model out of the box, and requires some refactoring.
-In the next post, we'll also cover a cool new feature in PyTorch Quantization called Define-by-Run, that tries to ease this constraint by needing only subsets of the model's computational graph to be free of dynamic flow. Check out the 
-[Define-by-Run poster at PTDD'21](https://s3.amazonaws.com/assets.pytorch.org/ptdd2021/posters/C8.png) for a preview.
+That was a lot to digest, congratulations for sticking with it! Next, we'll take a look at quantizing a "real-world" model that uses dynamic control structures (if-else, loops). These elements disallow symbolic tracing a model, which makes it a bit tricky to directly quantize the model out of the box. In the next post of this series, we'll get our hands dirty on a model that is chock full of loops and if-else blocks, and even uses third-party libraries in the `forward` call. 
+
+We'll also cover a cool new feature in PyTorch Quantization called Define-by-Run, that tries to ease this constraint by needing only subsets of the model's computational graph to be free of dynamic flow. Check out the [Define-by-Run poster at PTDD'21](https://s3.amazonaws.com/assets.pytorch.org/ptdd2021/posters/C8.png) for a preview.
 
 ***Thanks to Mark Saroufim for useful comments and feedback!***