SADIM-54M: Stellar Intelligence Framework

SADIM-54M is a specialized AI model trained on the Gaia Data Release 3 (DR3) catalog. It bridges the gap between massive raw astronomical observations and actionable scientific insights.

Research Objectives

• Galactic Archaeology: Identifying stellar streams and ancient structures.

• Big Data Optimization: Providing an AI-ready interface for 1B+ records.

• Scalability: Real-time stellar analysis for future space surveys.

1. Technical Feature Map

The model is designed to process 13 fundamental astronomical parameters:

Feature Name	Description
source_id	Unique Gaia DR3 Identifier
ra / dec	Celestial Equatorial Coordinates
l / b	Galactic Coordinates (Disk Alignment)
pmra / pmdec	Kinematics (Proper Motion Velocity)
d_pc	Distance in Parsecs (1/parallax)
x, y, z	3D Heliocentric Cartesian Mapping
abs_m	Absolute Magnitude (Intrinsic Brightness)
bp_rp	Color Index (Temperature Indicator)

2. When to use the Model vs. the Dataset?

• Use the SADIM-54M Model: For fast inference, predicting missing stellar properties, or automating the classification of new astronomical data.
• Use the Gaia-DR3 Dataset: For deep-dive research, historical record querying, or training your own custom neural networks.
• Dataset Link: samfatnassi/gaia-dr3

3. Quick Start (Python)

Since the dataset contains over 1 Billion records, we recommend using Streaming Mode:

from datasets import load_dataset
from transformers import AutoModel

# 1. Access the Data
dataset = load_dataset("samfatnassi/gaia-dr3", split="train", streaming=True)

# 2. Load the Model
model = AutoModel.from_pretrained("KilmaAI/SADIM-54M")

# Fetch a sample star
sample_star = next(iter(dataset))
print(f"Analyzing Star ID: {sample_star['source_id']}")