Midnight-Miqu-70B-v1.5
Property | Value |
---|---|
Parameter Count | 70B |
Model Type | Language Model (LLM) |
Architecture | DARE Linear Merge |
License | Other (Personal Use Only) |
Paper | DARE Paper |
What is Midnight-Miqu-70B-v1.5?
Midnight-Miqu-70B-v1.5 is an advanced language model created through a DARE linear merge of Midnight-Miqu-70B-v1.0 and Tess-70B-v1.6. It's specifically optimized for creative writing and roleplaying scenarios, maintaining the quality of its predecessor while incorporating improvements from the Tess model. The model supports up to 32K context length and demonstrates strong performance in narrative generation and character interaction.
Implementation Details
The model utilizes the DARE (Dynamic Auxiliary Representation Enhancement) linear merge method, implemented in FP16 precision. It achieves notable benchmark scores, including 61.18% on IFEval (0-Shot) and 38.54% on BBH (3-Shot), demonstrating its capabilities across various tasks.
- Supports Quadratic Sampling with recommended smoothing factor of 0.2
- Implements Min-P sampling for improved output quality
- Optimized for 32K context with alpha_rope=1
- Available in multiple quantization formats (GGUF, GPTQ, EXL2)
Core Capabilities
- Advanced creative writing and storytelling
- Character-based roleplaying with consistent persona maintenance
- Detailed sensory and environmental descriptions
- Uncensored output with content flexibility
- Long-context handling up to 32K tokens
Frequently Asked Questions
Q: What makes this model unique?
The model combines the creative strengths of Midnight-Miqu with improvements from Tess-70B, specifically optimized for storytelling and roleplaying while maintaining consistent character personas and generating rich, detailed narratives.
Q: What are the recommended use cases?
The model excels in creative writing, character-based roleplaying, and storytelling applications. It's particularly suited for scenarios requiring detailed environmental descriptions and character interactions, while maintaining contextual awareness and narrative consistency.