Moraliane/NekoMix-12B
Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:12BQuant:FP8Ctx Length:32kPublished:Dec 5, 2024Architecture:Transformer0.0K Warm

NekoMix-12B is a 12 billion parameter language model developed by Moraliane, created through a della_linear merge of several pre-trained models including IlyaGusev_saiga_nemo_12b, MarinaraSpaghetti_NemoMix-Unleashed-12B, Vikhrmodels_Vikhr-Nemo-12B-Instruct-R-21-09-24, and TheDrummer_Rocinante-12B-v1.1. With a 32768 token context length, this model is specifically optimized for roleplay scenarios, incorporating both English and Russian language support.

Loading preview...

NekoMix-12B: A Merged Model for Roleplay

NekoMix-12B is a 12 billion parameter language model developed by Moraliane, designed with a focus on roleplay applications. This model was created using the della_linear merge method via mergekit, combining several specialized base models to achieve its unique characteristics.

Key Capabilities & Composition

  • Merged Architecture: Built upon a base of IlyaGusev_saiga_nemo_12b, it integrates MarinaraSpaghetti_NemoMix-Unleashed-12B (a roleplay-oriented model), Vikhrmodels_Vikhr-Nemo-12B-Instruct-R-21-09-24 (for Russian language support and balance), and TheDrummer_Rocinante-12B-v1.1 (to enhance roleplay aspects).
  • Multilingual Support: The merge configuration emphasizes both Russian and English, with specific weights applied to models contributing to each language.
  • Roleplay Optimization: Weights were adjusted during the merge to prioritize roleplay capabilities, particularly through the inclusion and weighting of models known for their roleplay performance.
  • Context Length: Supports a substantial context window of 32768 tokens.

Recommended Usage

For optimal performance, users are advised to experiment with specific presets and sampler settings. The model's creator recommends starting with stock presets like "simple-1" from SillyTavern and specific Parameters_Top(A)Kek settings. Suggested sampler parameters include a Temperature range of 0.7-1.25, TopA at 0.1, and specific DRY settings (0.8, 1.75, 2, 0).

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p