This post explores the use and implications of temperature scaling in the language model output generation, specifically discussing its relationship to the beam search heuristic.
Deep learning has been deployed in many tasks in NLP, such as translation, image captioning, and dialogue systems. In machine translation, it is used to read source language (input) and generate the desired language (output). Similarly in a dialogue system, it is used to generate a response given a context. This is also known as Natural Language Generation (NLG).
