zhiyuanhucs/genshin-llava019-qwen3.5-9b-128k
The zhiyuanhucs/genshin-llava019-qwen3.5-9b-128k model is a 9-billion parameter, behavior-cloning SFT snapshot based on Qwen/Qwen3.5-9B. It is trained with a 128K context length, utilizing sequence packing and specific action/thought special tokens. This model is specialized for tasks related to the Genshin Impact game, integrating LLaVA 0.19 for potential multimodal capabilities, and is optimized for generating game-specific behaviors and responses.
Loading preview...
Overview
This model, genshin-llava019-qwen3.5-9b-128k, is a 9-billion parameter, behavior-cloning SFT (Supervised Fine-Tuning) snapshot developed by zhiyuanhucs. It is built upon the Qwen/Qwen3.5-9B base model and incorporates LLaVA 0.19, suggesting multimodal capabilities, particularly for Genshin Impact-related content. The model was trained with a substantial 128K context length using sequence packing, enabling it to process extensive inputs and generate detailed outputs.
Key Capabilities
- Genshin Impact Specialization: Fine-tuned specifically for tasks and interactions related to the Genshin Impact game, leveraging multiple "Genshin passes" during training.
- Large Context Window: Supports a 128K context length, allowing for processing and generating long, complex sequences of information.
- Behavior Cloning: Utilizes a behavior-cloning SFT approach, indicating its ability to mimic specific interaction patterns or responses.
- Special Tokens: Incorporates unique special tokens like
<|action_start|>,<|action_end|>,<|action_sep|>,<|thought_start|>, and<|thought_end|>for structured output generation, likely for agentic or interactive applications.
Good For
- Developing AI agents or chatbots for Genshin Impact.
- Generating game-specific narratives, dialogues, or actions.
- Research into behavior cloning and large context window applications within specific domains like gaming.