NASA Research Assistant (Ollama Model)

Model Description

Model Details

• Base Model: Microsoft Phi-3 Mini 3.8B
• Training Data: NASA research papers, mission reports, and scientific publications
• Model Size: ~3.8B parameters
• License: MIT
• Format: GGUF (compatible with Ollama, llama.cpp)

Training Data Sources

• NASA research papers from PubMed Central (3,938 documents processed)
• NASA Taskbook project descriptions
• NASA Open Science Data Repository
• Space biology and microgravity research
• Astronaut health and safety studies
• Planetary science publications

Capabilities

• Question Answering: Detailed responses about space science topics
• Research Summarization: Condensing complex scientific papers
• Technical Explanations: Breaking down aerospace concepts
• Mission Analysis: Discussing NASA missions and findings
• Scientific Accuracy: Trained on peer-reviewed research

Usage

With Ollama

Example Queries

• "What are the main health risks for astronauts on long-duration missions?"
• "Explain the effects of microgravity on plant growth"
• "Summarize recent findings about space radiation exposure"
• "How does the ISS maintain its orbit?"

Model Performance

• Scientific accuracy in space-related topics
• Detailed explanations suitable for researchers and students
• Proper citation of NASA research when applicable
• Clear communication of complex concepts

Limitations

• Knowledge cutoff based on training data (up to 2024)
• Primarily focused on NASA and US space research
• May not have complete coverage of very recent developments
• Should not be used for mission-critical decisions without verification

Training Process

1. Data Collection: Gathered NASA research papers and publications
2. Data Processing: Cleaned and chunked scientific texts
3. RAG Integration: Built retrieval-augmented generation system
4. Fine-tuning: Applied LoRA fine-tuning on instruction pairs
5. Optimization: Configured for scientific accuracy and detail

Technical Specifications

• Context Length: 4096 tokens
• Temperature: 0.7 (balanced creativity/accuracy)
• Top-p: 0.9
• Quantization: Q4_K_M (recommended for most use cases)

Citation

Contact

Acknowledgments

• NASA for providing open access to research data
• Microsoft for the Phi-3 base model
• The open-source community for tools and libraries