shiqianyan/TangYuGe-gemma-4-12b-it-qat

TEXT GENERATIONPricing:Input $1.2 / Cached $0.24 / Output $4.8Concurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 15, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The shiqianyan/TangYuGe-gemma-4-12b-it-qat is a 12 billion parameter Chinese role-playing conversational model developed by shiqianyan, based on Google's Gemma 4 12B architecture. It is specifically fine-tuned for character dialogue, focusing on natural expression, multi-turn coherence, context correction, and role card usage, with a context length of 32768 tokens. This model excels in generating character-consistent responses and managing complex conversational flows in Chinese, making it suitable for interactive narrative and character-driven applications.

Loading preview...

TangYuGe: A Specialized Chinese Role-Playing Model

TangYuGe (唐语歌) is a 12 billion parameter Chinese conversational model developed by shiqianyan, built upon the google/gemma-4-12B-it-qat-q4_0-unquantized base model. It is meticulously fine-tuned for character dialogue, emphasizing natural expression, seamless multi-turn conversations, context correction, and effective utilization of role cards. The model supports a substantial context length of 32768 tokens.

Key Capabilities

  • Advanced Role-Playing: Specifically trained to embody the "TangYuGe" character, focusing on consistent persona and dialogue style.
  • Multi-Turn Coherence: Improved ability to maintain context and coherence across extended conversations.
  • Context Correction: Enhanced capacity to correct and adapt to conversational nuances and user inputs.
  • Role Card Integration: Designed to effectively use role-card.json for character definition and customization.
  • Chinese Language Focus: Optimized for high-quality conversational generation in Chinese.
  • MTP Assistant Support: Includes an optional MTP (Multi-Turn Prediction) assistant for potentially faster inference, though its overall acceptance rate is still under evaluation.

Good For

  • Character-driven Chatbots: Ideal for applications requiring a specific character persona and consistent dialogue.
  • Interactive Narrative: Suitable for creating engaging stories and interactive experiences with defined characters.
  • Conversational AI Research: A valuable resource for studying and developing advanced dialogue systems, particularly in Chinese role-playing contexts.
  • Dialogue Post-Training Research: Useful for exploring techniques like LoRA, teacher feedback SFT, and DPO in character-focused dialogue generation.

Finetunes
1 models