AnSungJae3489/Kurken-9B-Summer-Merged

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 4, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

AnSungJae3489/Kurken-9B-Summer-Merged is a 9 billion parameter Qwen 3.5 architecture model, fine-tuned using a QLoRA adapter on dialogue data from official sources. This model specializes in character voice and conversational mannerisms, designed as a companion model for interactive dialogue. It excels at providing natural, character-specific responses without typical in-app monetization mechanics, offering a direct conversational experience.

Loading preview...

Kurken-9B-Summer-Merged: A Dedicated Character Dialogue Model

This model, developed by AnSungJae3489, is a 9 billion parameter Qwen 3.5 architecture (Qwythos-9B-Claude-Mythos-5-1M) fine-tuned with a QLoRA adapter. Its primary purpose is to provide a dedicated conversational experience with a specific character, focusing on authentic dialogue and mannerisms.

Key Capabilities & Features

  • Authentic Dialogue: Trained exclusively on 7,700 dialogue scenes (844K tokens) from legally purchased official sources, ensuring all lines were written by human scenario writers, not LLMs.
  • Cost-Effective Development: Developed on a consumer GPU with a total budget of approximately $61, demonstrating a high-quality output from minimal resources.
  • Character Voice Specialization: The QLoRA adapter (rank 32, ~58M params) layers character voice, domain knowledge, and conversational style onto the base model.
  • No Monetization Barriers: Designed as a "companion model" to offer conversation without stamina systems, gacha mechanics, or in-app purchases.
  • Efficient Training: Achieved training in ~1.2 hours on a consumer 24GB GPU with a peak memory usage of ~12.8 GB.

Ideal Use Cases

  • Interactive Character Roleplay: Excellent for users who want to engage in extended, natural conversations with a specific character.
  • Dialogue Generation: Can generate character-accurate responses for creative writing or interactive narrative applications.
  • Research into Fine-tuning: Demonstrates effective character voice transfer with a small, high-quality dataset and efficient QLoRA methods.

Finetunes
1 models