Indexnusrefather/Qwen-3.5-2b-roleplay-tuned-v2-ERP-tolerant
Indexnusrefather/Qwen-3.5-2b-roleplay-tuned-v2-ERP-tolerant is a 2.3 billion parameter language model based on the Qwen 3.5 architecture, fine-tuned by Indexnusrefather. This model is specifically optimized for complex roleplay and creative writing tasks, featuring improved punctuation and enhanced tolerance for explicit themes. It is designed for users seeking a smaller model for creative text generation with a focus on roleplaying scenarios.
Loading preview...
Overview
Indexnusrefather/Qwen-3.5-2b-roleplay-tuned-v2-ERP-tolerant is a 2.3 billion parameter model, a fine-tuned version of Qwen 3.5 2b, developed by Indexnusrefather. This iteration builds upon previous versions with an improved training methodology and additional data, primarily focusing on enhancing creative writing capabilities and refining the overall writing style. It maintains a context length of 32768 tokens.
Key Capabilities
- Improved Creative Writing: The model demonstrates enhanced ability in generating creative and engaging text.
- Better Punctuation: It features improved accuracy and consistency in punctuation usage.
- Enhanced ERP Tolerance: The model exhibits increased tolerance for explicit themes and a reduced tendency for refusals in such contexts.
- Potential Instability: Users should be aware of possible instability, as noted by the developer.
Deployment Notes
While direct quantizations were problematic for this specific model, users can achieve high quality with BF16, INT8W8A16, FP8, and NVFP4 formats. LoRA adapter files are provided in the repository for convenient deployment, allowing users to run the model with a base Qwen 3.5 2b model or merge the LoRA weights for a single file. For those requiring a 2B parameter model specifically for roleplay, this model offers a specialized solution, though the developer also suggests their Ministral 3 3b finetune as an alternative for broader use cases.