---
library_name: transformers
license: apache-2.0
license_link: https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507/blob/main/LICENSE
pipeline_tag: text-generation
---

This is a GRPO finetune to remove slop from this model: [Qwen3-4B-Instruct-2507-uncensored](https://huggingface.co/electroglyph/Qwen3-4B-Instruct-2507-uncensored)

It's not perfect, there are some before and after examples below.

I used the same method (mostly) as this model: [gemma-3-4b-it-unslop-GRPO-v3](https://huggingface.co/electroglyph/gemma-3-4b-it-unslop-GRPO-v3)

Note: This is *not* an RP tune, it's a compliant model with a different style from regular Qwen3 4B 2507.

My uncensoring dataset was generated by Gemma 3 27B abliterated model, which added a lot of Gemma writing style to this model.

It also added some Gemma style slop, which this finetune has helped mitigate.

I've uploaded a UD-Q4_K_XL GGUF with settings that I grabbed from Unsloth's quant using my lil utility: [quant_clone](https://github.com/electroglyph/quant_clone)

Here are some pics of before and after output with the slop highlighted and total at the bottom:

Prompt = "write a short story about a gothic romance, it should be around 500 words long"

Before:

![before](./orig1.png)

![before](./orig2.png)

![before](./orig3.png)

After:

![before](./new1.png)

![before](./new2.png)

![before](./new3.png)