MythoMax-L2-13B-GPTQ
| Property | Value |
|---|---|
| Parameter Count | 13B |
| Model Type | LLaMA2-based |
| License | Other + Meta Llama 2 License |
| Quantization | GPTQ (Multiple Options) |
What is MythoMax-L2-13B-GPTQ?
MythoMax-L2-13B-GPTQ is a quantized version of Gryphe's MythoMax L2 13B model, created through an innovative tensor merging technique combining MythoLogic-L2 and Huginn models. This GPTQ-quantized variant offers multiple compression options while maintaining high performance for both roleplay and creative writing tasks.
Implementation Details
The model implements a sophisticated tensor-based merge technique where 363 unique tensors each received specific ratio treatments. It's available in various GPTQ configurations, including 4-bit and 8-bit versions with different group sizes and Act Order options, allowing users to balance between VRAM usage and inference quality.
- Multiple quantization options (4-bit to 8-bit)
- Group sizes ranging from 32g to 128g
- Act Order optimization available
- Compatible with ExLlama, AutoGPTQ, and Hugging Face TGI
Core Capabilities
- Advanced roleplay interactions
- Creative story writing
- Coherent long-form content generation
- Flexible prompt handling with Alpaca format
- Optimized for both understanding (input) and writing (output)
Frequently Asked Questions
Q: What makes this model unique?
The model uses a pioneering tensor type merge technique that optimizes both input understanding and output generation, resulting in superior performance in both roleplay and creative writing tasks. Each tensor is individually calibrated for optimal performance.
Q: What are the recommended use cases?
The model excels in roleplay scenarios, creative writing, and story generation. It's particularly well-suited for applications requiring both strong comprehension and creative output generation.
