trashpanda-org/MS-24B-Instruct-Mullein-v0

MS-24B-Instruct-Mullein-v0 Overview

MS-24B-Instruct-Mullein-v0 is a 24 billion parameter instruction-tuned language model developed by trashpanda-org, built upon the unsloth/Mistral-Small-24B-Instruct-2501 base and merged with the MS-24B-Mullein-v0 model using the TIES method. This instruct variant is noted for its improved character and scenario portrayal, offering a more controlled output compared to its base version, which is described as more "unhinged." The model leverages a diverse set of datasets for its training, including Allura's Sugarquill 10k, estrogen's floyd-instruct, Gryphe's Sonnet3.5 RP, kalo's Opus-22k, Norquinal's OpenCAI, and Dampfinchen's Creative Writing Multiturn, among others.

Key Capabilities

• Enhanced Character Portrayal: Excels at accurately portraying characters and their personas within narratives.
• Scenario Generation: Capable of generating detailed and consistent scenarios.
• Creative Writing: Strong performance in creative writing tasks, including nuanced interactions and narrative development.
• Instruction Following: Designed to follow instructions effectively, leading to more predictable and refined outputs.

Good For

• Roleplay and Interactive Fiction: Ideal for applications requiring detailed character and scenario generation.
• Narrative Development: Suitable for creative writing, story generation, and developing complex plotlines.
• Content Generation: Useful for generating engaging and contextually rich textual content where character consistency is important.