Model Overview

Magnolia-Mell-v1-12B is a 12 billion parameter language model developed by grimjim. It was created using the mergekit tool, specifically employing an asymmetric gradient SLERP (Spherical Linear Interpolation) method. This merge combined two base models: grimjim/Magnolia-v3-12B and inflatebot/MN-12B-Mag-Mell-R1.

Key Characteristics

• Merge Method: Utilizes an asymmetric gradient SLERP, applying MN-12B-Mag-Mell-R1 lightly to Magnolia-v3-12B.
• Performance: Tested for narrative text completion, showing fairly high coherence.
• Robustness: Appears to tolerate higher inference temperatures (e.g., temperature=2.0) without significant degradation, which can help mitigate repetition.

Intended Use Cases

This model is particularly well-suited for:

• Narrative Text Completion: Generating coherent and creative long-form stories, descriptions, or other narrative content.
• Creative Writing: Assisting with tasks that require imaginative and consistent text generation.
• Exploratory Generation: Users can experiment with higher temperatures to vary output without immediately losing coherence.