Llama-MagicalGirl is a 70 billion parameter language model developed by KaraKaraWitch, created by merging several pre-trained Llama-based models using the SCE method. Built upon KaraKaraWitch/Llama-3.X-Workout-70B, this model is designed to meet specific creative requirements, though it can exhibit offensive or dark outputs without a system prompt. It is optimized for creative text generation, with recommended sampling settings to achieve desired outputs.
Loading preview...
Model Overview
Llama-MagicalGirl is a 70 billion parameter language model developed by KaraKaraWitch, constructed through a merge of multiple pre-trained Llama-based models. The model utilizes the SCE (Sparse Component Extraction) merge method, building upon KaraKaraWitch/Llama-3.X-Workout-70B as its base. The merge incorporated models such as SicariusSicariiStuff/Negative_LLAMA_70B, TheDrummer/Nautilus-70B-v0.1, Steelskull/L3.3-Nevoria-R1-70b, and Tarek07/Inception-LLaMa-70B.
Key Characteristics
- Merge Method: Employs the SCE merge method for combining diverse model strengths.
- Base Model: Built on
KaraKaraWitch/Llama-3.X-Workout-70B. - Creative Focus: Designed to fulfill specific creative generation requirements.
- Sampling Recommendations: Optimal performance is achieved with specific sampling settings (Temperature: 1.4, Min P: 0.03) to enhance creativity and control verbosity.
Important Considerations
- System Prompt Requirement: The model can produce offensive or dark content if used without a system prompt. Users are strongly advised to implement a simple system prompt to guide its behavior.
This model aims to provide a unique generative experience, particularly for users seeking specific creative outputs, provided appropriate safety measures like system prompts are in place.
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.