Learn Before
Troubleshooting a Tool-Use Fine-Tuning Process
A development team is fine-tuning a language model to answer questions about recent events by using a search_web(query) tool. Their training dataset consists of prompts paired with complete, fact-based answers. For example:
Prompt: 'Who won the Best Picture award at the most recent Oscars?' Target Output: 'Oppenheimer won the Best Picture award at the most recent Oscars.'
After fine-tuning on thousands of similar examples, the team observes that the model still frequently provides incorrect or made-up answers for new questions about recent events, instead of learning to use the search tool. Based on the principles of preparing data for tool use, diagnose the fundamental error in the team's data preparation strategy and explain what the correct target output should have looked like for the given prompt.
0
1
Tags
Ch.3 Prompting - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Analysis in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
A developer is creating a training dataset to teach a language model how to use an external tool called
get_current_weather(location). The model should learn to insert a special command to call this tool when asked for weather information. Given the desired final output: 'The weather in Paris is currently 18°C and cloudy.', which of the following examples correctly annotates this instance for the training data?Troubleshooting a Tool-Use Fine-Tuning Process
A developer is creating a single training example to fine-tune a language model for tool use. They have a user's prompt and the ideal final response that relies on external information. Arrange the following steps in the correct chronological order to create the final annotated data point for the training set.