Magnum-v4-27b-gguf
| Property | Value |
|---|---|
| Parameter Count | 27.2B |
| License | Gemma |
| Architecture | GGUF Format, ChatML-based |
| Training Hardware | 8x H100 GPUs |
| Training Duration | 2 epochs |
What is magnum-v4-27b-gguf?
Magnum-v4-27b-gguf is an advanced language model built on the Gemma 27B architecture, specifically designed to emulate the prose quality of Claude 3 models (Sonnet and Opus). It utilizes the GGUF format for efficient deployment and implements the ChatML conversation format for structured interactions.
Implementation Details
The model was fine-tuned using the Axolotl framework on a diverse collection of 9 high-quality datasets, focusing on instructional, conversational, and creative content. Training was conducted on 8x NVIDIA H100 GPUs provided by Recursal AI/Featherless AI, implementing full-parameter fine-tuning for optimal performance.
- ChatML-based conversation format for consistent interaction patterns
- Comprehensive training on multiple specialized datasets
- Implementation of advanced training techniques including gradient checkpointing and flash attention
- 8192 sequence length with sample packing enabled
Core Capabilities
- High-quality prose generation similar to Claude 3
- Structured conversation handling through ChatML format
- Support for complex instruction following
- Integration with platforms like SillyTavern through specialized templates
- Extensive context window of 8192 tokens
Frequently Asked Questions
Q: What makes this model unique?
The model stands out for its specific focus on replicating Claude 3's prose quality while maintaining efficient deployment through GGUF format. It combines the strengths of Gemma architecture with carefully curated training datasets and specialized conversation handling.
Q: What are the recommended use cases?
The model excels in conversational AI applications, creative writing tasks, and instruction-following scenarios. It's particularly well-suited for applications requiring high-quality prose generation and structured dialogue interactions.
