Speed improvements using hash tables

But why? I don't see that the author explains why he doesn't use what's already in the language which he uses to implement all his "dictionaries":

https://docs.python.org/3/tutorial/datastructures.html#dicti...

Also, the "improvement" is when starting from "a linear search through all of keys"?

Edit: Thanks to anonymoushn, so the language in which the "dictionaries" are implemented is actually an independent language:

"Almost all of Flpc is written in Flpc itself and defined at runtime. This means almost everything is modifiable at runtime using rebind: including the loop for reading input, compiling functions, looking up names and rebind: itself!"

From: https://github.com/asrp/flpc

So the author is writing a compiler but only learned about hashtables?