Reka-Flash-3-21B-Reasoning-Uncensored-MAX-NEO-Imatrix-GGUF
| Property | Value |
|---|---|
| Author | DavidAU |
| Base Model | Reka-Flash-3-21B |
| Context Length | 128k |
| Model Hub | Hugging Face |
What is Reka-Flash-3-21B-Reasoning-Uncensored-MAX-NEO-Imatrix-GGUF?
This is an enhanced version of RekaAI's Reka-Flash 3 model, specifically optimized for reasoning and problem-solving tasks. The model features "Neo Imatrix" augmentation and "Maxed out" quantization to improve overall performance while maintaining strong reasoning capabilities even at lower quantization levels.
Implementation Details
The model implements several key optimizations: BF16 output tensor for all quants, custom Neo Imatrix dataset for improved function and instruction following, and augmented quantization that enables up to 50% faster problem-solving compared to non-augmented versions.
- Uses reasoning tags for operation (<reasoning> and </reasoning>)
- Supports multiple quantization levels (IQ1-Q8)
- Features 128k context window
- Operates without requiring system prompts
Core Capabilities
- Enhanced reasoning and problem-solving abilities
- Strong performance even at lower quantization levels
- Stable temperature/reasoning across wide ranges
- Uncensored output generation
- Faster solution generation compared to standard models
- Improved instruction following and world knowledge through Neo Imatrix
Frequently Asked Questions
Q: What makes this model unique?
The model combines Neo Imatrix dataset augmentation with maxed quantization, enabling superior reasoning capabilities and faster problem-solving while maintaining strong performance even at lower quant levels. It's particularly notable for its temperature stability and uncensored output.
Q: What are the recommended use cases?
The model excels at reasoning tasks, creative writing, problem-solving, and general text generation. It's particularly effective for cases requiring complex thinking or creative solutions, with recommended quants IQ3s/IQ4XS/IQ4NL for creative uses and q5s/q6/q8 for general usage.
