EVA-Tissint-v1.2-14B
| Property | Value |
|---|---|
| Parameter Count | 14.8B |
| Model Type | Text Generation |
| Architecture | Merged Transformer (Qwen2-based) |
| Tensor Type | BF16 |
What is EVA-Tissint-v1.2-14B?
EVA-Tissint-v1.2-14B is an advanced language model created through a sophisticated merge of EVA v0.2 and Tissint v1.2 using mergekit technology. This model represents a careful balance between two powerful base models, with optimized density and weight parameters to enhance performance.
Implementation Details
The model employs a della_linear merge method, utilizing EVA-Qwen2.5-14B-v0.2 as its base. The merge configuration uses precise density and weight parameters: Tissint (density: 0.45, weight: 0.35) and EVA (density: 0.55, weight: 0.65), with additional parameters epsilon: 0.05 and lambda: 1.
- Mergekit-based implementation
- BFloat16 precision for optimal performance
- Balanced parameter distribution between base models
- 128k context window support
Core Capabilities
- Advanced text generation and completion
- Conversational AI applications
- Enhanced context understanding
- Optimized for text-generation-inference
Frequently Asked Questions
Q: What makes this model unique?
This model's uniqueness lies in its carefully calibrated merge of EVA and Tissint models, offering a balance between their respective strengths while maintaining optimal performance through precise density and weight parameters.
Q: What are the recommended use cases?
The model is particularly well-suited for text generation tasks, conversational applications, and scenarios requiring extended context understanding. It's recommended to use the samplers provided in the Tissint model card for optimal results.
