sirmyrrh/Kyllima-34B-v1
Kyllima-34B-v1 by sirmyrrh is a 34 billion parameter language model, merged from two Yi 34B-based models using the TIES method, specifically optimized for roleplay and creative writing tasks. It features a 32768 token context length and is uncensored, incorporating elements from Nous Capybara 34B and Bagel DPO. This model is designed for nuanced text generation in creative applications, offering specific prompt formats and settings for optimal performance.
Loading preview...
Kyllima-34B-v1 Overview
Kyllima-34B-v1 is a 34 billion parameter language model developed by sirmyrrh, specifically engineered for roleplay and creative writing. This model is a 50/50 merge of two Yi 34B-based models, utilizing the TIES merge method with TeeZee/Kyllene-34B-v1.1 as its base. It integrates components from models like Nous Capybara 34B and Bagel DPO, aiming for enhanced performance in its target domains.
Key Characteristics
- Architecture: Merged from Yi 34B-based models using
mergekitand the TIES method. - Context Length: Supports a 32768 token context window.
- Specialization: Optimized for generating creative text and engaging in roleplay scenarios.
- Uncensored: Designed to be uncensored, offering flexibility in content generation.
- Prompt Formats: Compatible with metadata, Alpaca-LimaRP, or Vicuna prompt formats.
- Recommended Settings: Optimal performance is achieved with specific settings (e.g., 0.8-1 temperature, 1.1-1.2 repetition penalty).
Use Cases
This model is particularly well-suited for applications requiring:
- Creative Writing: Generating stories, narratives, and other forms of creative text.
- Roleplay: Engaging in detailed and dynamic roleplaying interactions.
Users should note its sensitivity to prompt style and settings, and consider adding specific instructions to the system prompt for best results.