Phi-4-Model-Stock-v2
| Property | Value |
|---|---|
| Author | bunnycore |
| Base Model | unsloth/phi-4 |
| Merge Method | Model Stock |
| Precision | bfloat16 |
| Model URL | Hugging Face |
What is Phi-4-Model-Stock-v2?
Phi-4-Model-Stock-v2 is an advanced language model created through a sophisticated merge of multiple pre-trained models using mergekit. Built on the foundation of unsloth/phi-4, it combines various specialized variants including Phi-4-Stock-RP, SuperThoughts-CoT, and Phi-4-Stock-Ex to create a more capable and versatile model.
Implementation Details
The model utilizes the Model Stock merge method with normalized parameters and bfloat16 precision. It incorporates multiple specialized models, each contributing unique capabilities to the final merged version.
- Base architecture: unsloth/phi-4
- Merged components: Phi-4-Stock-RP, SuperThoughts-CoT-14B-16k-o1-QwQ, Phi-4-Stock-Ex
- Precision: bfloat16
- Normalized parameters for optimal performance
Core Capabilities
- Strong performance on IFEval (0-Shot): 63.75%
- Impressive BBH (3-Shot) results: 54.69%
- MATH Level 5 (4-Shot) capability: 35.88%
- Professional-level MMLU performance: 48.12%
- Average performance across benchmarks: 38.87%
Frequently Asked Questions
Q: What makes this model unique?
The model's uniqueness lies in its careful combination of specialized Phi-4 variants and SuperThoughts-CoT, resulting in strong performance across various benchmarks, particularly in few-shot learning scenarios.
Q: What are the recommended use cases?
Given its strong performance on IFEval and BBH, the model is particularly well-suited for tasks requiring logical reasoning, few-shot learning, and complex problem-solving applications.
