Model Description
SaulM-7B is a specialized model fine-tuned on a corpus of legal, human resources, and workplace policy documents (including federal law, Texas state law, and company handbooks). It is designed to act as the reasoning engine for the "Grace" RAG (Retrieval-Augmented Generation) pipeline.
Its primary function is to receive a complex prompt containing both:
- A set of classification rules.
- A user's workplace scenario query.
- Retrieved context (RAG) from a 3-layer (Federal, State, Company) vector store.
The model then analyzes this information to classify the scenario as "Broken" (a violation), "Not Broken" (compliant), or "Uncertain," providing a detailed JSON-formatted explanation with exact citations from the retrieved context.
This Q4_K_M GGUF version is specifically intended for deployment in environments with limited resources, particularly for CPU-only inference.
Intended Use
The primary intended use for this model is as the core inferencing component of the Grace application. It is not a general-purpose chatbot.
It is designed to be called via an API (like the one built with FastAPI) and expects a highly structured prompt.
Out-of-Scope Uses
- General Chat: The model will perform poorly at general conversation.
- Standalone Inference: The model is not intended to be used without its RAG pipeline. It relies on the provided context to make accurate, citable judgments.
- Legal Advice: This model does not provide legal advice. It is a tool for surfacing relevant policy information and should be used by HR professionals as a reference, not as a substitute for legal counsel.
