Anoopsingh53/ISRO-SpaceAI-7B-Instruct
ISRO-SpaceAI-7B-Instruct by Anoopsingh53 is a 7.61 billion parameter instruction-tuned foundation language model based on the Qwen 2.5 7B Instruct architecture, featuring a 32,768 token context length. This model is specifically designed for scientific reasoning and multi-spectral telemetry analysis across heliophysics, oceanography, and planetary observation. It excels in domains like ISRO Aditya-L1 Heliophysics, CalCOFI/Oceansat-3 Marine Oceanography, Sentinel-1 SAR Microwave Radar Floods, and NASA Kepler Exoplanetary Photometry, demonstrating strong domain adaptation with low perplexity on scientific test sets.
Loading preview...
ISRO-SpaceAI-7B-Instruct: India's First Multi-Domain Space AI
ISRO-SpaceAI-7B-Instruct is an open-weights, 7.61 billion parameter foundation language model developed by Anoopsingh53. Built on the Qwen 2.5 7B Instruct architecture, it is specialized for scientific reasoning and multi-spectral telemetry analysis across diverse space and earth science domains. The model was trained using 4-bit NormalFloat (NF4) QLoRA, with unquantized full IEEE FP16 weight safe-merging, and features a native context length of 32,768 tokens.
Key Capabilities & Specialization
- Domain-Specific Expertise: Purpose-built for analyzing data from ISRO Aditya-L1 Heliophysics, CalCOFI/Oceansat-3 Marine Oceanography, Sentinel-1 SAR Microwave Radar Floods, and NASA Kepler Exoplanetary Photometry.
- Multi-Scale Scientific Reasoning: Bridges disciplines from sub-nanometer solar EUV spectral flux to deep-sea CTD hydrographic profiles and exoplanetary transit light curves.
- Empirical Benchmarking: Achieves strong domain adaptation with perplexity values below 11 and exact next-token accuracy between 53-60% on specialized test sets for oceanography, heliophysics, and astrophysics.
- Robust Architecture: Utilizes a 28-layer auto-regressive decoder-only dense transformer with Grouped-Query Attention and Rotary Position Embedding (RoPE) with $\theta = 1,000,000$.
Ideal Use Cases
- Scientific Data Analysis: Analyzing complex datasets from space missions and earth observation.
- Research & Development: Supporting research in heliophysics, oceanography, and exoplanetary science.
- Educational Tools: Developing tools for understanding multi-spectral telemetry and scientific phenomena.
- Geospatial AI Pipelines: Integration into projects requiring multimodal AI for atmospheric composition and oceanographic sonification.