AeroMind-Phi4-mini

AeroMind-Phi4-mini : GGUF

This model was finetuned and converted to GGUF format using Unsloth.

Example usage:

• For text only LLMs: llama-cli -hf Kami574/AeroMind-Phi4-mini --jinja
• For multimodal models: llama-mtmd-cli -hf Kami574/AeroMind-Phi4-mini --jinja

Available Model files:

• Phi-4-mini-instruct.Q8_0.gguf

Ollama

An Ollama Modelfile is included for easy deployment. This was trained 2x faster with Unsloth <img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>

---

license: apache-2.0 base_model: unsloth/Phi-4-mini-instruct tags:

• aerospace
• engineering
• physics
• phi-4-mini
• slm
• fine-tuned language:
• en pipeline_tag: text-generation datasets:
• aerospace-qna-v6

---

🚀 Phi-4 Mini Aerospace Engineering Q&A Engine (v6)

This model is a fine-tuned version of Phi-4 Mini optimized for high-fidelity aerospace engineering assistance. It was trained on a custom synthetic dataset generating multi-turn, physics-grounded, and mathematically rigorous engineering dialogues across 30 aerospace subdomains.

---

🛠 Model Overview

• Base Architecture: microsoft/Phi-4-mini-instruct
• Dataset Version: V6 (Production SLM Fine-Tuning Ready)
• Format: Multi-Turn JSONL / ChatML with System Prompt Injection
• Train / Val Split: 90% Training / 10% Validation

---

📊 Dataset Breakdown by Subdomain

The dataset generation engine executes 1,000 global production iterations distributed uniformly across all 30 aerospace subdomains, yielding ~33–34 multi-turn conversation instances per subdomain (each containing 1 to 3 turn-pairs per instance).

Note: Exact final count depends on length-validation filtering and duplicate detection performed during post-generation cleaning.

---

⚙️ Key Dataset Features (V6 Engine)

1. Mandatory System Prompts: Injected directly into every instance during training to reinforce role adherence.
2. Adversarial Physics Traps: Integrates domain-matched variables (beta_val, crack_length, combustion_instability, etc.) to evaluate physical reasoning.
3. Positive Code Auditor Paradigm: Outputs verified, fully operational Python code snippets with architectural explanations rather than exhibiting buggy inputs.
4. Context Window Safety: Length-validation filtering applied to prevent high-token context OOM exceptions during fine-tuning.

---

💡 System Prompt

All training examples enforce the following system prompt:

You are a highly advanced aerospace engineering AI assistant. You provide mathematically rigorous, physics-grounded, and safety-critical responses across all aerospace subdomains.