Movie Scripts Converted to Role-Play Scenes

Cinematika V0.1 contains 211 movie scripts converted into a novel-style, multi-character role-play format. The dataset was created by jondurbin using a pipeline of manual regex parsing and LLM augmentation with a custom Mistral-7B fine-tune. It was last updated on Hugging Face in April 2024.

Use Cases

• Train dialogue models on multi-character interactions using the converted role-play scene data.
• Fine-tune large language models for creative writing by learning from the novel-style narrative structure of the scenes.
• Analyze script-to-prose conversion techniques by studying the transformations applied to the original movie scripts.
• Benchmark text style transfer models using the original script and converted role-play data pairs.

Strengths

• Contains 211 distinct movie scripts as a foundation for conversion.
• Data was processed using a custom fine-tuned Mistral-7B model for LLM augmentation.
• Dataset version was released on April 11, 2024.

Limitations

• The dataset is an initial version (V0.1) with a limited scope of 211 movies, pending expansion to a planned ~2400.
• Specific details on data volume, column structure, and sample quality are unavailable from the provided description.
• The conversion methodology may introduce artifacts or biases from the regex parsing and fine-tuned model.

Provenance

Source

Original movie scripts, source unspecified.

Collection Method

Manual regex parsing combined with LLM augmentation using a custom Mistral-7B fine-tune.

Freshness

Last updated 2024-04-11.