Evaluating Model Adaptation Strategies for a Specialized Task
A small startup with limited computational resources and a small, specialized dataset wants to adapt a large, general-purpose pre-trained language model for a new task. Two main strategies are proposed:
- Strategy A: Update the model's internal parameters by training it extensively on the small, specialized dataset.
- Strategy B: Keep the model's parameters frozen and instead prepend carefully crafted instructions to the input to guide the model's behavior for the new task.
Evaluate the suitability of each strategy for the startup's situation. Justify which approach is more likely to be successful and explain the potential risks associated with the less suitable approach.
0
1
Tags
Ch.4 Alignment - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Evaluation in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
Prompting as a Motivator for Universal Foundation Models
A research team has a large, pre-trained language model designed for general text understanding. Initially, to make this model perform a specific task like classifying emails as 'spam' or 'not spam', they had to collect thousands of labeled emails and use them to update the model's internal parameters. This process was resource-intensive. Subsequently, a new approach was discovered that achieved similar results with far less effort. Which statement best analyzes the core principle of this more efficient new approach?
Evaluating Model Adaptation Strategies for a Specialized Task
Recommending an Adaptation Strategy for a Legal Tech Startup