| Property | Value |
|---|---|
| Model Size | 8B parameters |
| Base Architecture | Llama 3.1 |
| Author | Nexesenex |
| Hugging Face | Model Repository |
| Average Benchmark Score | 23.45 |
What is Llama_3.1_8b_Dolermed_V1.01?
Llama_3.1_8b_Dolermed_V1.01 is a specialized merged language model that combines medical expertise with general language capabilities. Built using the Model Stock merge method, it integrates Llama-3.1-MedIT-SUN-8B and Hermes-3-Llama-3.1-8B-lorablated with Dolphin3.0-Llama3.1-8B as the base model.
Implementation Details
The model employs a sophisticated merge methodology using mergekit, with specific configuration parameters including bfloat16 dtype and normalized weights. The merge process maintains equal weighting (1.0) for both constituent models, ensuring balanced capabilities.
- Uses Model Stock merge methodology
- Implements bfloat16 precision
- Features normalized weights
- Employs union-based tokenizer configuration
Core Capabilities
- IFEval (0-Shot): 50.87% accuracy
- BBH (3-Shot): 31.71% performance
- MATH Level 5 (4-Shot): 13.44% accuracy
- Professional MMLU (5-shot): 28.56% accuracy
- Specialized medical knowledge integration
Frequently Asked Questions
Q: What makes this model unique?
This model's uniqueness lies in its balanced integration of medical expertise from MedIT-SUN with the general capabilities of Hermes-3, all built on the Dolphin3.0 base. The normalization and precise merge methodology ensure consistent performance across different domains.
Q: What are the recommended use cases?
The model is particularly well-suited for medical and healthcare applications while maintaining strong general-purpose capabilities. It shows notable strength in inference tasks (50.87% on IFEval) and professional knowledge testing (28.56% on MMLU-PRO).