Model Overview
monsoon-nlp/nyrkr-joker-llama is a specialized language model fine-tuned from LLaMa2-7b-hf (7 billion parameters) to analyze and explain humor in New Yorker cartoon captions. Developed by monsoon-nlp, this model takes a cartoon description and its caption, then generates an explanation of how the joke works and why it's funny.
Key Capabilities
- Humor Explanation: Generates detailed explanations for New Yorker cartoon captions, linking the caption to the provided scene description.
- Contextual Understanding: Demonstrates an understanding of situational humor and comedic timing within the context of a cartoon.
- Fine-tuned on Specific Data: Leverages a dataset of 1.2k New Yorker caption contest entries for its specialized task.
Technical Details
The model was fine-tuned using QLoRA on a V100 GPU, merging adapter weights from checkpoint step=160. It utilizes a specific prompt format to guide its joke explanations, focusing on the relationship between the scene, caption, and the comedic element.
Ideal Use Cases
This model is particularly well-suited for:
- Humor Analysis: Research into the mechanics of humor and joke structures.
- Content Generation: Assisting in the creation of humorous content by providing insights into comedic elements.
- Educational Tools: Developing applications that help users understand and appreciate complex humor.