How Transformers Generate Text

Medium3 min read

How decoding strategies (greedy search, top-k, nucleus sampling, and temperature) shape the text a model actually outputs.

Originally published on

Medium

Read the full article