mistral-7b-instruct-tuned

Model Card for mistral-7b-instruct-sft

Abstract

This repository hosts a fine‑tuned Mistral 7B‑Instruct model that leverages parameter‑efficient LoRA adaptation via the Unsloth framework. The model is adapted for email‑security tasks using a curated phishing‑email training dataset and demonstrates state‑of‑the‑art performance (94.9 % accuracy, 93.9 % precision, 96.1 % recall).

Model Details

Model Description

• Developed by: Montimage
• Model type: Large Language Model (LLM)
• Language(s): English
• License: Apache‑2.0
• Finetuned from model: unsloth/mistral-7b-instruct-v0.3-bnb-4bit

Model Sources

• Repository: https://huggingface.co/nosadaniel/mistral-7b-instruct-tuned

How to Get Started with the Model

You can download the model using the Hugging Face Space.

Locally using Ollama

1. Install Ollama – https://ollama.com
2. Pull the base model:

   ollama pull unsloth/mistral-7b-instruct-v0.3-bnb-4bit
   

3. Create a merged‑model manifest (no file extension) with the following content and place it in a folder of your choice:

   FROM mistral-7b-instruct-v0.3-bnb-4bit
   ADAPTER /path/to/your/downloaded/adapter
   

   Then run:

   ollama create merged-model --from ./merged-model
   

4. Run the merged model:

   ollama run merged-model
   

Training Details

Training Data

• Dataset: Phishing Email Training Dataset
• Link: https://huggingface.co/datasets/nosadelian/phishing-email-training-dataset

Training Procedure

• Fine‑tuning method: LoRA (Low‑Rank Adaptation) via Unsloth
• Training regime: fp16 mixed precision
• Epochs: 3 (full dataset)
• Learning rate: 2e‑4
• Batch size: 32

Model Comparison Table (selected row for this model)

Comparison with Base Model (mistral:7b): The fine‑tuned model achieves substantially higher accuracy (94.9 % vs 84.0 %), recall (96.1 % vs 72.7 %), and overall quality metrics, while maintaining comparable precision, demonstrating the effectiveness of LoRA fine‑tuning for email‑phishing detection.

Model Performance Analysis – mistral‑7b‑instruct‑sft

• Total Responses: 256
• Accuracy: 94.9 % (243/256)
• Valid Responses: 100 % (256/256)
• Average Confidence: 0.921

Classification Metrics

Confusion Matrix

Additional Metrics

• False Positive Rate: 6.2 %
• False Negative Rate: 3.9 %
• Negative Predictive Value: 96.0 %
• Matthews Correlation Coefficient: 0.899

Performance Insights

• ✅ High Precision – Low false‑positive rate, fostering user trust.
• ✅ High Recall – Catches the vast majority of phishing attempts, enhancing security.
• ✅ Excellent F1‑Score – Well‑balanced precision and recall.
• ✅ Strong MCC – Strong overall correlation between predictions and ground truth.

Citation

BibTeX:

@dataset{mistral-7b-instruct-sft,
  title={Fine‑tuned Mistral 7B‑Instruct model for Email Phishing Detection},
  author={Montimage, Nosadaniel, Luong89},
  year={2025},
  publisher={Montimage}
}

APA:

Montimage, Nosakhare Daniel Ahanor, & Luong89. (2025). Fine‑tuned Mistral 7B‑Instruct model for Email Phishing Detection. Montimage.

Model Card Authors

Montimage Email Security Research Division AI/ML Engineering Team Cybersecurity Domain Experts

Framework versions

• PEFT 0.17.1