Learn Before
Designing a New NLP Task Format
Imagine you are designing a new natural language processing task: 'Question Generation'. The goal is to train a model to generate a relevant question based on a given statement. This model uses a unified 'Source Text -> Target Text' framework, where the task is specified by a prefix at the beginning of the source text.
Using the statement 'The Eiffel Tower was completed in 1889', create one complete, formatted training sample for this new task. Your answer should clearly show both the source text and the target text.
0
1
Tags
Ch.1 Pre-training - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Creation in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
Example of a T5 Translation Scoring Sample
A developer wants to adapt a pre-trained model that uses a unified text-to-text framework for a new task: sentiment analysis. The goal is for the model to read a customer review and output one of three labels: 'positive', 'negative', or 'neutral'. Which of the following represents the best-formatted source text to send to the model to classify the review 'The battery life is incredible!'?
Designing a New NLP Task Format
A research team is exploring the capabilities of a model that uses a unified 'text-to-text' framework. The model processes an input string (which includes a prefix indicating the task) and generates an output string. Match each of the following natural language processing tasks with the most appropriate input/output format that represents it within this framework.