MFANNv0.25
| Property | Value |
|---|---|
| Parameter Count | 8.03B |
| License | LLaMA 3.1 |
| Tensor Type | F32 |
| Context Length | 8192 tokens |
| Language | English |
What is MFANNv0.25?
MFANNv0.25 is an 8.03B parameter language model based on the LLaMA 3.1 architecture, developed by netcat420. This model features a unique dual-personality system, offering both a standard helpful assistant mode and an experimental "SATANN" mode for cybersecurity operations.
Implementation Details
The model implements advanced transformer architecture with specific sampling parameters optimized for performance. It operates with a maximum context length of 8192 tokens and uses F32 tensor precision for calculations. The implementation includes carefully tuned parameters such as temperature: 1, top p: 1, top k: 50, and a repeat penalty of 1.19 over 69 tokens.
- Built on LLaMA 3.1 architecture with safetensor implementation
- Supports batch processing with 128 prompt batch size
- Configurable GPU layer offloading (32 layers for vulkan)
- Minimum probability threshold of 0.03
Core Capabilities
- Dual-mode operation with distinct personality profiles
- Long-context understanding (8K tokens)
- Text generation and conversational abilities
- Specialized cyber operations support in SATANN mode
- Compatible with text-generation-inference endpoints
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its dual-mode operation, allowing it to switch between a standard helpful assistant and a specialized cybersecurity-focused mode (SATANN), making it versatile for different use cases.
Q: What are the recommended use cases?
The model is suitable for general conversational AI applications in standard mode, while the SATANN mode is designed for cybersecurity research and testing scenarios. Users should be mindful of ethical considerations when utilizing the specialized mode.
