Hugging Face MarinaraSpaghetti/Nemomix-v4.0-12B is a 12 billion parameter language model, part of the Nemomix series, merged using the della_linear method with Mistral-Nemo-Base-2407 as its base. This model integrates the intelligence of Instruct Nemo with the enhanced prose of various roleplaying fine-tunes, making it particularly effective for generating nuanced and engaging text. It is optimized for scenarios requiring both smart instruction following and high-quality narrative output, with a context length of 32768 tokens.
Nemomix-v4.0-12B Overview
Nemomix-v4.0-12B is the latest iteration in the Nemomix series by MarinaraSpaghetti, a 12 billion parameter language model designed to combine the "smartness" of base Instruct Nemo models with the "better prose" from various roleplaying fine-tunes. This version is presented as the most effective merge in the series to date.
Key Capabilities
- Enhanced Prose and Roleplaying: Specifically merged to improve narrative quality and roleplaying capabilities, drawing from models like Intervitens' mini-magnum-12b-v1.1, Invisietch's Atlantis-v0.1-12B, and NeverSleepHistorical's lumi-nemo-e2.0.
- Instruction Following: Built upon a Mistral Instruct base, ensuring strong adherence to given instructions.
- Optimized for Specific Settings: Recommended temperature of 0.35, with suggestions for higher temperatures (1.0-1.2) when combined with increased Min P (0.01-0.1) and base DRY settings of 0.8/1.75/2/0.
- Merge Method: Created using the
della_linearmerge method, withmistralaiMistral-Nemo-Base-2407as the foundational model.
Good For
- Use cases requiring a balance of intelligent instruction processing and high-quality, engaging text generation, particularly in creative writing or roleplaying contexts.
- Developers looking for a model that excels in producing nuanced and well-written responses while maintaining strong instructional adherence.