Few-Shot Intent Recognition Prompts for LLM Tuning

A collection of datasets reformatted into prompts for few-shot intent recognition research with large language models. The dataset, created by author 'qgyd2021', was last updated on May 14, 2024. Subsets like {dataset_name}_prompt are dynamically generated from source data and template subsets, resulting in varied outputs.

Use Cases

• Fine-tuning LLMs for intent classification using dynamically generated prompt templates like '{dataset_name}_prompt'.
• Benchmarking few-shot learning performance on intent recognition tasks using the provided prompt examples and intent labels.
• Studying prompt engineering strategies for tasks where prompts may be truncated, using the dataset's guidelines and template structure.
• Training models to map user queries (e.g., 'will i be okay on the gym...') to specific intents (e.g., 'Is it safe to go to the gym indoors if I'm vaccinated?').

Strengths

• Includes practical guidelines for designing prompts resilient to truncation during LLM training.
• Provides example prompts with clear intent labels and user query text for model demonstration.

Limitations

• Specific dataset size, row count, and column details are unknown, making scalability assessment difficult.
• The dynamic generation of prompt subsets leads to non-deterministic outputs, which may affect reproducibility.

Provenance

Source

Hugging Face user qgyd2021.

Collection Method

Collected from existing intent recognition datasets and reformatted into prompt templates.

Freshness

Last updated on 2024-05-14.