Zual/MPropositionneur-V2
Zual/MPropositionneur-V2 is a compact 0.8 billion parameter language model, based on Qwen3-0.6B, developed by Luc Pommeret at LISN (CNRS). It is specifically fine-tuned for atomic proposition extraction, designed to decompose complex sentences into simple, independent statements. This model excels at improving information granularity for tasks like RAG, Open Information Extraction, and text simplification. It supports multiple languages, including French and English, for proposition generation.
Loading preview...
MPropositionneur-V2: Atomic Proposition Extraction Model
MPropositionneur-V2 is a small, specialized language model (0.8B parameters) built upon the Qwen3-0.6B architecture. Developed by Luc Pommeret at LISN (CNRS), its core function is to perform atomic proposition extraction through distillation. This process involves breaking down complex sentences or passages into a list of simple, independent, and semantically faithful statements.
Key Capabilities
- Atomic Proposition Generation: Decomposes intricate text into fundamental, decontextualized assertions.
- Multilingual Support: Trained to handle multiple languages, with a focus on French and English.
- JSON Output: Provides extracted propositions as a structured JSON list of strings.
- Prompt Format: Utilizes a specific prompt format (
<|im_start|>user\nAtomize: {text}<|im_end|>\n<|im_start|>assistant\n) for consistent input.
Good for
- Retrieval-Augmented Generation (RAG): Enhances RAG systems by indexing atomic propositions instead of larger text chunks, leading to more granular and precise retrieval.
- Open Information Extraction (OpenIE): Facilitates the extraction of structured information from unstructured text.
- Text Simplification and Discourse Analysis: Aids in simplifying complex content and analyzing the underlying propositions within discourse.