Naija-Petro -- Petroleum Engineering AI

Domain-specific LLM fine-tuned for petroleum engineering on Qwen3-32B.

Overview

• 20,000 synthetic instruction-response pairs (NVIDIA Data Designer)
• QLoRA fine-tuning with Unsloth (2x faster, 70% less VRAM)
• Covers: drilling, reservoir, production, completions, EOR, well testing

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("Shinzmann/naija-petro", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("Shinzmann/naija-petro")

Ollama

ollama run hf.co/Shinzmann/naija-petro-GGUF:Q4_K_M

Training

Param	Value
Base	Qwen3-32B
Method	QLoRA 4-bit
LoRA r / alpha	64 / 128
LR	0.0002
Epochs	2
Samples	~19K train / ~1K eval

Limitations

• Validate outputs with qualified engineers before operational use
• English only; not for general chat