| Property | Value |
|---|---|
| Base Model | Gryphe/MythoMax-L2-13b |
| Model Type | LLaMA Architecture |
| License | LLaMA 2 |
| Quantization Options | 2-bit to 8-bit |
What is MythoMax-L2-13B-GGML?
MythoMax-L2-13B-GGML is a quantized version of the MythoMax L2 13B model, specifically optimized for CPU and GPU inference using the GGML format. It represents a sophisticated merge between MythoLogic-L2 and Huginn models, utilizing an experimental tensor type merge technique for enhanced performance in both roleplay and story writing tasks.
Implementation Details
The model offers multiple quantization levels ranging from 2-bit to 8-bit precision, with file sizes varying from 5.51GB to 13.79GB. Each quantization level provides different trade-offs between model size, RAM usage, and inference quality. The implementation uses various k-quant methods for optimal performance across different use cases.
- Supports multiple quantization formats (q2_K through q8_0)
- Optimized tensor distribution for enhanced coherency
- Customized prompt template for optimal interaction
- Compatible with llama.cpp and various UI implementations
Core Capabilities
- Advanced roleplay and character interaction
- High-quality story writing and narrative generation
- Efficient CPU+GPU inference with various RAM/VRAM configurations
- Support for context lengths up to 4096 tokens with RoPE scaling
Frequently Asked Questions
Q: What makes this model unique?
The model utilizes a unique tensor-type merge technique where 363 tensors have individual ratios applied, resulting in superior performance in both comprehension and generation tasks. It effectively combines MythoLogic-L2's robust understanding with Huginn's writing capabilities.
Q: What are the recommended use cases?
The model excels in roleplay scenarios and creative writing tasks. It's particularly well-suited for applications requiring both strong comprehension and coherent output generation, with various quantization options allowing deployment across different hardware configurations.