How Transformers Generate Text
How decoding strategies — greedy search, top-k, nucleus sampling, and temperature — shape the text a model actually outputs.
Originally published on Medium (~680 words). Read the full piece here: How Transformers Generate Text.