czcheung/Qwen3-4B-Instruct-2507-uncensored-unslop-v2

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 14, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The czcheung/Qwen3-4B-Instruct-2507-uncensored-unslop-v2 is a 4 billion parameter instruction-tuned causal language model, fine-tuned from electroglyph/Qwen3-4B-Instruct-2507-uncensored. This model has undergone a GRPO finetuning process to specifically reduce verbosity and 'slop' in its outputs, offering a more concise and direct writing style. With a context length of 32768 tokens, it is optimized for generating compliant text with a distinct style, making it suitable for applications requiring focused and less verbose responses.

Loading preview...

Overview

This model, czcheung/Qwen3-4B-Instruct-2507-uncensored-unslop-v2, is a 4 billion parameter instruction-tuned variant of the Qwen3-4B-Instruct-2507-uncensored model. Its primary distinction lies in a specialized GRPO (Generative Reinforcement Learning with Policy Optimization) finetuning process aimed at mitigating 'slop' – excessive verbosity and repetitive phrasing – from its generated text. This finetuning was inspired by methods used in electroglyph/gemma-3-4b-it-unslop-GRPO-v3.

Key Capabilities & Differentiators

  • Reduced Verbosity: Significantly less 'slop' compared to its base model, resulting in more concise and direct outputs.
  • Distinct Writing Style: While compliant, it offers a different stylistic output than standard Qwen3 4B 2507 models, influenced by the Gemma writing style from the uncensoring dataset.
  • Uncensored Base: Built upon an uncensored foundation, providing broader response capabilities while refining output quality.
  • GGUF Availability: A UD-Q4_K_XL GGUF version is provided, generated using quant_clone for efficient deployment.

When to Use This Model

This model is particularly well-suited for use cases where:

  • Concise and direct responses are preferred over verbose or 'slop'-filled text.
  • A compliant model with a unique, less generic writing style is desired.
  • Applications benefit from an uncensored base model that has been refined for output quality.
  • Developers need a 4B parameter model with a 32768 token context length that has been specifically optimized for stylistic improvements.

Finetunes
1 models