prompts

Prompts

We save pre-defined prompts is a simple JSON format.

The keys correspond to the names of arguments expected by llm_inference.py. For example:

{
    "prompt_prefix": "Rewrite the complex sentence with simple sentence(s). Keep the meaning same, but make it simpler.",
    "prompt_suffix": "Complex: {input}\nSimple:",
    "prompt_template": "Complex: {complex}\nSimple: {simple}",
    "example_separator": "\n\n",
    "prompt_format": "prefix_initial",
    "source_field": "complex",
    "target_field": "simple"
}

To use a pre-defined prompt, simply pass the path of the prompt.json file to inference.py with the arugment --prompt_json.

Note, even though LangChain allows you to load pre-defined prompts, we found that doing it this way allows for more flexibility.

• p0.json is similar to the prompt used in Feng et al., 2023
• p1.json is similar to p1.json but presents the examples in a more 'predict-the-next-token' format.
• p2.json is similar to the prompt used by Maddela et al., 2022 and is based on the instructions provided to crowd workers to create the ASSET reference simplificationsAlva-Manchego et al., 2021.