ssuki/bi-intent-discovery-mistral

BI Intent Discovery Model

This model is fine-tuned from Mistral-7B-Instruct to perform Business Intelligence (BI) intent discovery tasks.

Model Description

The model analyzes natural language questions about business intelligence data and breaks them down into structured steps for query building. It performs two main phases:

1. Planning Phase: Detects complex questions and breaks them into ordered steps
2. Discovery Phase: Extracts BI concepts (measures, dimensions, timeframes, etc.) from questions

Training Data

• 500 examples of BI questions with structured outputs
• Covers various complexity levels from simple to multi-step queries
• Includes examples with ambiguity handling and unmatched intent capture

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import json

# Load model
model_name = "ssuki/bi-intent-discovery-mistral"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)

# Example usage
question = "Show me total sales by region for the last quarter"
# Format with your prompt template and generate response

Output Format

The model outputs structured JSON containing:

• Intent classification
• Discovery results with measures, dimensions, timeframes
• Unmatched intents for ambiguous terms
• Step-by-step breakdown for complex queries

Training Configuration

• Base Model: Mistral-7B-Instruct-v0.2
• Training Examples: 500
• Epochs: 5
• Learning Rate: 2e-05
• Max Sequence Length: 512