0xA50C1A1/Qwen3-4B-Instruct-2507-NanoWriter
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Feb 16, 2026License:apache-2.0Architecture:Transformer Open Weights Warm
Qwen3-4B-Instruct-2507-NanoWriter is a lightweight, fast instruction-tuned language model based on the Qwen3 architecture, developed by 0xA50C1A1. This model is specifically fine-tuned using LoRA with RS-LoRA scaling for uncensored storytelling. Its primary use case is generating creative narratives without content restrictions, leveraging techniques like NEFTune for noise injection during training.
Loading preview...
Qwen3-4B-Instruct-2507-NanoWriter Overview
Qwen3-4B-Instruct-2507-NanoWriter is an experimental, lightweight instruction-tuned language model built on the Qwen3 architecture. Developed by 0xA50C1A1, this model's core focus is on uncensored storytelling, aiming to provide a fast and efficient solution for creative writing tasks.
Key Capabilities
- Uncensored Storytelling: Specifically fine-tuned to generate narratives without typical content restrictions, utilizing tools like Heretic for censorship removal during its development.
- Lightweight and Fast: Designed to be a nimble LLM, making it suitable for applications where speed and resource efficiency are crucial.
- LoRA Fine-tuning: Trained using a LoRA (16-bit) method with a rank of 32 and RS-LoRA scaling, enhancing its performance for its specialized task.
- Noise Injection (NEFTune): Incorporates NEFTune (alpha=5) during training, which can help improve model robustness and creativity in generation.
Good for
- Creative Writers: Ideal for authors and content creators seeking an LLM that can assist with generating diverse and unrestricted story content.
- Experimental Narrative Generation: Useful for researchers and developers exploring the boundaries of AI-driven storytelling and content creation.
- Applications Requiring Unfiltered Output: Suitable for use cases where the absence of inherent content filters is a desired characteristic for generated text.