MatteoKhan/MistralGemma-7B-Merged

🚀 MistralGemma-Hybrid-7B: A Fusion of Power & Precision

📌 Overview

MistralGemma-Hybrid-7B is an experimental hybrid language model that blends the strengths of Mistral-7B and Gemma-7B using the Spherical Linear Interpolation (slerp) merging technique. Designed to optimize both efficiency and performance, this model offers robust text generation capabilities while leveraging the advantages of both parent models.

🔗 Created by: [Matteo Khan]
🎓 Affiliation: Apprentice at TW3 Partners (Generative AI Research)
📍 License: MIT

🔗 Connect with me on LinkedIn
🔗 Model on Hugging Face

🧠 Model Details

• Model Type: Hybrid Language Model (Merged)
• Parent Models:
  • Mistral-7B-v0.1
  • Gemma-7B
• Merging Technique: Slerp Merge (MergeKit)

🎯 Intended Use

This model is intended for research and experimentation in hybrid model optimization. Potential applications include:

• ✅ Text Generation
• ✅ Conversational AI
• ✅ Creative Writing Assistance
• ✅ Exploration of Model Merging Effects

⚠️ Limitations & Considerations

While MistralGemma-Hybrid-7B offers enhanced capabilities, it also inherits limitations from its parent models:

• ❌ May generate inaccurate or misleading information
• ⚠️ Potential for biased, offensive, or harmful content
• 🔄 Merging may introduce unpredictable behaviors
• 📉 Performance may vary across different tasks

🔬 Merging Process & Configuration

This is not a newly trained model, but rather a merge of existing models using the following configuration:

merge_method: slerp  # Using slerp instead of linear
dtype: float16
models:
  - model: "mistralai/Mistral-7B-v0.1"
    parameters:
      weight: 0.5
  - model: "google/gemma-7b"
    parameters:
      weight: 0.5

parameters:
  normalize: true
  int8_mask: false
  rescale: true  # Helps with different model scales

layers:
  - pattern: ".*"
    layer_range: [0, -1]

📊 No formal evaluation has been conducted yet. Users are encouraged to benchmark and share feedback!

🌍 Environmental Impact

By utilizing model merging rather than training from scratch, MistralGemma-Hybrid-7B significantly reduces computational and environmental costs.

🚀 How to Use

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "YourProfile/MistralGemma-Hybrid-7B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Example usage
prompt = "Write a short story about the future of AI."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=200)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

📝 Citation

@misc{mistralgemma2025,
      title={MistralGemma: A Hybrid Open-Source Language Model},
      author={Your Name},
      year={2025},
      eprint={arXiv:XXXX.XXXXX},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

📩 Feedback & Contact: Reach out via Hugging Face.

🎉 Happy Experimenting! 🚀