malikali/CEFR-Aligned-LM

CEFF-Aligned Language Model (CaLM)

This is a model card for the CEFF-Aligned Language Model (CaLM) from the paper: From Tarzan to Tolkien: Controlling the Language Proficiency Level of LLMs for Content Generation

• Paper: https://aclanthology.org/2024.findings-acl.926
• Arxiv: https://arxiv.org/abs/2406.03030
• Github: https://github.com/malik-ali/tarzan2tolkien

Prompt Template

The model text template looks like

<<Summary>>: {summary}
<<CEFR>>: {cefr}
<<Story>>:
{story}
<</Story>>

where you replace {summary} with the summary of the desired story to generate and {cefr} with the desired CEFR level is one of ["A1", "A2", "B1", "B2", "C1", "C2"].

To generate, you can add the summary and target CEFR level and just start generating after the <<Story>>:\n . See the Github repo for examples.