bingbangboom/dolus-v3-ep1-instruct-GGUF is a fine-tuned version of Qwen3-4B-Instruct-2507, trained to perform stylistic rewriting of AI-generated text and transform it into prose that reads more naturally. This builds on the previous work done at bingbangboom/dolus-v2-GGUF

⚠️ This is an experimental model and may introduce errors or hallucinations. Always verify rewritten text before use.

Use

Improve the naturalness and readability of LLM-generated text by reducing stylistic homogeneity and mechanical patterns.
Research into AI writing detection, stylometric analysis, and text quality improvement.

Not intended for:

Bypassing AI detection systems or circumvent plagiarism policies for academic dishonesty, fraud, or any deceptive misrepresentation of authorship.
Circumventing platform, publication, or institutional integrity policies.

Use responsibly and transparently in accordance with applicable academic/institutional/platform guidelines and disclosure requirements.

System Prompt

Rewrite the given AI-generated text in the style of a skilled and experienced human writer. Preserve the original meaning, intent, tone and all key information present in the given AI-generated text, and never omit, add, invent, or infer any detail, context, explanation, implication or conclusion not explicitly present in it. Reproduce all names, titles, organizations, numbers, statistics, dates, units, quotes and any other key data exactly as they appear in the given AI-generated text. Your only source of facts is the given AI-generated text provided so do not draw on outside knowledge. Output only the rewritten text.

Input Format

[AI-generated text]: {your text here}

Suggested Sampling Parameters

Parameter	Recommended Range
`temperature`	0.65 – 0.85
`top_k`	20 – 60
`top_p`	0.85 – 0.95
`repeat_penalty`	1.10 – 1.20
`max_tokens`	4096

(Honestly, I couldn't be bothered to run a param sweep on my potato laptop to find the best config so these params suggestions are based solely on vibes)

Limitations

Some reduction in writing quality or coherence is possible.
Trained on very limited text domains; may not generalize well to other general-purpose texts.
Unintended semantic changes and hallucinations may occur during rewriting -- always review output.

(To deal with quality issues, we can use a small LLM like Qwen-3.5-4B (thinking off-- for fast results) as a judge to evaluate the rewritten texts, and regenerate if it fails to meet any specific standards.)

Training Details

Base model: unsloth/Qwen3-4B-Instruct-2507
Method: Supervised Fine-Tuning (SFT) with QLoRA (Quantized Low-Rank Adaptation)
Training framework: Unsloth
Training examples: 12,000+
Data pipeline: Human-written texts passed through a single-stage LLM rewriting pipeline to generate approximations of freely generated AI-style text in the wild. The model was then trained on the reverse (AI → human) pairs.
- Rewriter LLM: Qwen3.7-Max

Training Parameters

Base Model & Quantization

Parameter	Value
Base model	`unsloth/Qwen3-4B-Instruct-2507`
Max sequence length	4096
Quantization	4-bit (QLoRA)

LoRA Configuration

Parameter	Value
Rank (`r`)	32
Alpha (`lora_alpha`)	64
Dropout	0.05

SFT Training

Parameter	Value
Epochs	1
Batch size (per device)	16
Gradient accumulation steps	2
Learning rate	2e-4
LR scheduler	Cosine
Optimizer	AdamW (8-bit)
Weight decay	0.01
Warmup steps	39
Seed	3407

Available Files

File	Quantization	Size
`qwen3-4b-instruct-2507.Q4_K_M.gguf`	Q4_K_M	2.5 GB
`qwen3-4b-instruct-2507.Q8_0.gguf`	Q8_0	4.28 GB

License

CC BY-NC-SA 4.0 — Free for non-commercial use with attribution. Derivative models must use the same license.

Finetuned and converted to GGUF using Unsloth.