OGAI-8x7B: Oil & Gas AI Model for Drilling Process Optimization

Model Description

OGAI-8x7B is a LoRA fine-tuned Mixtral-8x7B model, engineered for oil and gas engineering applications. This version has been optimized for drilling process automation, technical document understanding, and engineering problem-solving.

The model is a core component of GainEnergy's Upstrima AI Platform, which provides pragmatic AI agents, advanced workflows, and retrieval-augmented generation (RAG)-enhanced document processing.

Technical Architecture

Base Model Specifications

• Architecture: Mixtral-8x7B Sparse Mixture of Experts (SMoE)
• Parameters: 8 experts with 7B parameters each (46.7B total parameters, 12.9B active per token)
• Context Length: Extended to 32,768 tokens for handling technical documentation
• Attention Mechanism: Sliding Window Attention with specialized oil & gas technical vocabulary

Fine-tuning Approach

• Method: Low-Rank Adaptation (LoRA) with rank 16
• Training Dataset: 2.8M specialized oil & gas engineering datapoints (800K drilling-specific)
• Hardware: Trained on 16x NVIDIA A100 80GB GPUs
• Training Time: 2,800 GPU hours
• Special Features: Enhanced numerical precision for engineering calculations with reduced hallucination rates

Performance Optimizations

• Quantization: Custom 4-bit and 8-bit quantization schemes preserving numerical accuracy
• Inference Speed: Optimized KV cache management for real-time drilling advisory
• Memory Footprint: Reduced to 12GB with 4-bit quantization while maintaining 95%+ calculation accuracy
• Speculative Decoding: Implemented for 2.7x faster response generation with technical queries

Deployment-Optimized Versions

For flexible deployment options, we provide quantized versions of this model:

• 4-bit Quantized Version (NF4) – Lower memory usage with efficient performance.
• 8-bit Quantized Version (Int8) – Balanced model size and precision.
• GGUF Version for llama.cpp – Optimized for CPU-based inference and edge computing.

Using the GGUF Model with llama.cpp

To run OGAI-8x7B using llama.cpp, use the following command:

./server \
  --gguf-file-name ogai-8x7b-q4_k_m.gguf \
  --repo-slug GainEnergy/OGAI-8x7b-Q4_K_M-GGUF \
  --np 8

Deployment Options

This model supports one-click deployment to various platforms directly from the Hugging Face Hub:

Hugging Face Inference Endpoints

Click the "Deploy" button and select "Inference Endpoints" to deploy on Hugging Face's managed infrastructure.

Local Deployment with vLLM

python -m vllm.entrypoints.openai.api_server \
  --model GainEnergy/ogai-8x7b \
  --tensor-parallel-size 2

OGAI Model Family & Expansion Roadmap

OGAI-8x7B is part of the OGAI Model Family, designed for various energy-sector applications.

Model Name	Base Model	Fine-Tuning	Domain Focus
OGAI 3.1 Engineer	Custom Framework	Full fine-tuning	AI-powered Oil & Gas Engineering Assistance
OGAI-Quantum	Hybrid AI-Quantum	Hybrid Fine-Tuning	Reservoir Simulation & Seismic Data Processing
OGAI-R1	TinyR1-32B	Fine-tuned	Engineering AI Reasoning & Logical Analysis
OGMOE	Mixtral-8x7B MoE	Fine-tuned	Mixture of Experts (MoE) for Drilling Optimization
OGAI-8x22B	Mixtral-8x22B	Full fine-tuning	High-Performance LLM for Engineering Reasoning
OGAI-8x7B	Mixtral-8x7B	LoRA fine-tuning	Drilling Optimization, Well Planning & Document RAG

How to Use

Run Inference in Python

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "GainEnergy/ogai-8x7b"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto")

prompt = "Calculate the required casing depth for a well in a high-pressure formation."
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Citing OGAI-8x7B

@article{ogai8x7b2025,
  title={OGAI-8x7B: An AI Model for Oil & Gas Drilling Engineering},
  author={GainEnergy AI Team},
  year={2025},
  publisher={Hugging Face Models}
}