sophosympatheia/Midnight-Rose-70B-v1.0
Midnight-Rose-70B-v1.0 is a 69 billion parameter DARE TIES merge model created by sophosympatheia, built upon Llama 2 and incorporating Tulu-2-DPO, Lzlv, and Opus models. This model is specifically optimized for creative writing, roleplaying, and storytelling, aiming to produce lengthy and coherent output. It features a 32768 token context length and is designed for uncensored generative text tasks.
Loading preview...
Midnight-Rose-70B-v1.0: A Creative Writing and Roleplaying Powerhouse
Midnight-Rose-70B-v1.0 is a 69 billion parameter language model developed by sophosympatheia, engineered through a DARE TIES merge. It combines the strengths of several prominent models, including allenai/tulu-2-dpo-70b, lizpreciatior/lzlv_70b_fp16_hf, and dreamgen/opus-v0.5-70b, further enhanced by integrating specific LoRAs like Airoboros-L2 and fiction.live-Kimiko-V2. This model is a successor to previous merges like Rogue Rose and Aurora Nights, offering improved performance in its target domains.
Key Capabilities & Features
- Optimized for Creative Writing: Designed to generate lengthy, coherent, and imaginative text, making it ideal for storytelling.
- Roleplaying Excellence: Excels in interactive roleplaying scenarios, maintaining character consistency and narrative flow.
- Uncensored Output: Provides uncensored content generation, offering flexibility for diverse creative applications.
- High Context Length: Supports a 32768 token context window, allowing for detailed and extended interactions.
- Advanced Sampling Recommendations: Benefits from specific sampling techniques like Quadratic Sampling and Min-P for enhanced output quality.
Ideal Use Cases
Midnight-Rose-70B-v1.0 is particularly well-suited for:
- Interactive Storytelling: Generating dynamic and engaging narratives.
- Character-driven Roleplay: Creating immersive experiences with consistent character personas.
- Creative Content Generation: Producing long-form creative text where detailed and imaginative output is desired.
Users are encouraged to experiment with system prompts and sampling settings to fine-tune the model's behavior for their specific needs.