Zual/MPropositioneur-V1: Atomic Proposition Extraction

MPropositioneur-V1, developed by Luc Pommeret at LISN (CNRS), is a compact language model built upon the Qwen3-0.6B architecture. Its core function is to atomize complex sentences or passages, breaking them down into a list of simple, independent, and semantically faithful atomic propositions.

Key Capabilities & Features

• Atomic Proposition Extraction: Specializes in deconstructing text into its most basic declarative units.
• Distillation Training: Optimized through distillation to achieve its specialized task.
• Multilingual Support: Trained to handle multiple languages, including French and English.
• JSON Output: Generates propositions as a JSON list of strings, ["p1", "p2", ...], for easy programmatic use.
• Specific Prompt Format: Utilizes a clear prompt structure: <|im_start|>user\nAtomize: {text}<|im_end|>\n<|im_start|>assistant\n.

Ideal Use Cases

This model is particularly well-suited for applications requiring fine-grained information processing:

• Retrieval-Augmented Generation (RAG): Improves retrieval quality by indexing atomic propositions instead of larger text chunks.
• Open Information Extraction (OpenIE): Facilitates more precise extraction of facts and relationships.
• Text Simplification and Discourse Analysis: Aids in breaking down complex information for easier understanding and analysis.