summerMC/matutake

Model Overview

summerMC/matutake is a 1.5 billion parameter model developed by summerMC, built upon the summerMC/Sakura base model. It underwent full Supervised Fine-Tuning (SFT) using the summerMC/v-Fable dataset, which consists of Fable-style trace data. The primary goal of matutake is to adapt the Fable-style and response tendencies to the Sakura base, focusing on natural Japanese conversation, reasoning, and role-play.

Key Capabilities

• Natural Japanese Conversation: Excels in general chat, following instructions, and generating longer conversational responses.
• Enhanced Reasoning & Explanation: Capable of providing step-by-step explanations, summarizing, comparing, and generating extended logical responses.
• Fable-style & Roleplay: Designed to reproduce Fable-style conversational patterns, role-play responses, and atmosphere-driven dialogue.
• Robust Instruction Following: Improves adherence to Japanese prompts for information organization and text generation.

Intended Use Cases

This model is ideal for general dialogue and generation tasks, including casual Japanese chat, generating responses based on specific instructions, and creating character-driven or atmospheric conversations. It is particularly strong in maintaining conversational flow, producing coherent explanations, and adapting to various roles or contexts. The model prioritizes naturalness, atmosphere, and extended response generation over benchmark score optimization.

Limitations

matutake is intended for research and experimental purposes. It does not guarantee factual accuracy, may produce plausible but incorrect explanations, and can exhibit inconsistencies in very long responses. Its Fable-style bias might lead to overly emotional or dramatic outputs, and it may generate verbose, chain-of-thought-like content. Human verification is crucial for any critical applications.