Smaug Humanizer

Smaug Humanizer

Rewrites AI-written text in a human voice while keeping its facts, numbers, names and layout. A 12B model fine-tuned by Abacus.AI from Gemma 4, for writers and editors.

Made by abacusai on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Smaug Humanizer is a 12B text model from Abacus.AI. You give it text that an AI wrote, and it returns the same text reworded so it reads like a person wrote it. Facts, numbers, names and layout stay in place, while the wording, rhythm and voice change.

Who it suits: Writers, editors and content teams who start from AI drafts and want them to sound more natural, and developers building a rewriting step into a tool.

Industry | Who uses it | How they use it | Example
Publishing | Editors | Smooth AI-drafted articles into a more natural voice | A news brief draft reworded for style
Marketing | Content teams | Rewrite AI copy while keeping numbers and names intact | A product description with the same specs
Education | Teachers and tutors | Study how wording and rhythm differ between AI and human text | A classroom comparison of two versions
Software | Developers | Add a rewrite step to a writing tool | A button that rewrites a pasted paragraph

If your text is in another language or runs far beyond 500 words in one piece, you can skip it, since the card lists English and about 500 words per call.

Where to use it:

• Windows, macOS or Linux with Python, for downloading the files
• Python programs that use the transformers library
• Local apps that read GGUF files, through the separate GGUF release the card names

Use case | Who | What they get back
Rewriting a draft | Writers | The same text in a more human voice
Checking facts survive | Editors | A rewrite that keeps every number, name and quote
Batch rewriting | Developers | Paragraph-sized rewrites from a script

How to set it up:

• Python, to run the Hugging Face download tool
• About 23.9 GB of disk space for the BF16 weights, or about 6 GB at 4-bit as an estimate
• A GPU or other hardware that can run a 12B model

1. Install the Hugging Face hub tools with pip.
2. Download the model files with the hf command.
3. Load the model with the transformers library and follow the card's sampling settings.

pip install -U huggingface_hub
hf download abacusai/Smaug-Humanizer

How to use it: You pass in AI-written text, a paragraph or a few at a time, and get the rewritten text back. The card recommends a temperature of 1.0, top_p of 0.95 and top_k of 64.

Tool | What it does
Input | AI-written text of up to about 500 words per call
Output | The text reworded so a person could have written it
temperature | Recommended value is 1.0
top_p | Recommended value is 0.95
top_k | Recommended value is 64

Example: You paste a 400-word product description written by an AI. The model returns the same description with the same facts, numbers, names and paragraph breaks, but with different wording and rhythm.

For newcomers and experts: Start with one short paragraph and compare the output with your original, checking that every number and name survived. Keep the sampling settings from the card.

The model starts from Gemma 4 12B, a pretrained base that is not instruction-tuned. Training had two stages. SFT used 3,676 filtered rewrite pairs, and RL ran 200 steps of GRPO with a reward that favors the human version over the AI draft. Five checks guard numbers, layout, length, script and spelling.

AI-written text goes in
Model rewrites wording and rhythm
Facts, numbers and layout stay
Human-sounding text comes out

Good to know:

• License: Apache 2.0
• Cost to run: none named, though a 12B model needs capable hardware
• Data it reaches: downloads files from Hugging Face, then runs where you load it
• Not for: English-only use is listed, so other languages are not covered
• Not for: long inputs, since the card gives about 500 words per call as the best size

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 12B
• Weights: about 23.9 GB in BF16; about 6 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 1,162
• Likes: 29
• License: apache-2.0
• Languages: en
• Based on: google/gemma-4-12B
• Quantized versions: 4
• Tags: transformers, safetensors, gemma4_unified, image-text-to-text, smaug, abacusai, humanizer, gemma4, grpo, text-generation, conversational

Links

• Model: https://huggingface.co/abacusai/Smaug-Humanizer
• Files: https://huggingface.co/abacusai/Smaug-Humanizer/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download abacusai/Smaug-Humanizer

Then follow the usage in its model card. Nothing in this kit runs by itself.