humanizer-GGUF

humanizer-GGUF

Rewrites AI-written drafts in English and Chinese so they read like a person wrote them, keeping numbers, dates, names and quotes. A 12B model in GGUF files for running locally.

Made by jialinyyzz 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: humanizer is a 12B model, in its second version, that rewrites drafts written by AI in English and Chinese. The goal is wording that reads as if a person wrote it, while numbers, dates, names and quotes stay as they were. This repository holds its GGUF files, from a large 8-bit file down to small 2-bit ones.

Who it suits: Writers, editors and students who start from an AI draft and want plainer, more natural wording, and who are willing to proofread the result.

Industry | Who uses it | How they use it | Example
Content teams | Editors | Smooth AI-assisted drafts before review | A blog draft reworded, with the facts left alone
Education | Students | Rework a first draft in their own voice and check it | An essay draft rewritten, then proofread by the student
Business writing | Marketers and analysts | Turn stiff generated copy into natural prose | A product update that keeps its figures and names
Translation work | Bilingual writers | Rewrite English or Chinese drafts locally | A Chinese draft reworded without leaving the machine

Skip it if you write in other languages, since the card covers English and Chinese only.

Where to use it:

• The author's desktop app for Mac and Windows, which fetches the files for you
• A local runner that opens GGUF files, on Windows, macOS or Linux

Use case | Who | What they get back
Rewrite a draft | Writers | A new version with the facts kept
See the changes | Editors | A side-by-side view with new wording highlighted, in the desktop app
Pick a size for your memory | Hobbyists | A file from 5.6 GB up to 12.7 GB

How to set it up: Pick the file that matches your memory. The 8-bit file is recommended with 32 GB or more, the 6-bit file suits 16 GB, and the smaller files suit machines with less.

• Python and pip, if you download by command
• Enough memory for the chosen file, from 8.0 GB of peak use on the smallest listed to 13.7 GB for the 8-bit file
• Or just the desktop app, which downloads the files itself

1. Install the Hugging Face hub tool.
2. Download the recommended 8-bit file together with the prompt format file.

pip install -U huggingface_hub
hf download jialinyyzz/humanizer-GGUF humanizer-12b-Q8_0.gguf prompt_format.json

How to use it: In your own runner, load the GGUF file and use the prompt format file that came with it.

Tool | What it does
hf download | Fetches the model file and the prompt format file
prompt_format.json | Holds the prompt format the card refers to
Desktop app | Downloads the files and shows draft and rewrite side by side

Example: You paste an AI-written product update that contains figures, a date and a customer quote. The model returns a version in plainer, more natural wording, with those figures, the date and the quote unchanged.

For newcomers and experts: If you are new to this, start with the desktop app, paste a short draft and compare the two columns. Always read the rewrite before you use it.

If you know the field, the three smallest kinds of file got extra training so they imitate the full bf16 model, which served as their teacher. Sensitive tensors got more bits. The card reports that at 3.9 GB the 2-bit file picks the same next token as bf16 87 percent of the time, against 70 percent for a standard build.

Paste an AI-written draft
The model rewrites it
Numbers, dates and names stay
You proofread the result

Good to know: The card's results are the maker's own. It reports 11 of 210 English rewrites flagged as AI by one detector, and 376 of 420 with no factual problem. Check your school's, publisher's or employer's rules about AI-assisted text.

Title: File size in GB
Q8_0: 12.7 GB
Q6_K: 10.0 GB
Q4_K_M: 7.6 GB
IQ3_XXS-QAT: 5.6 GB
2-bit: 3.9 GB

• Smaller files: the Q3-size file needs proofreading of numbers and names, since English can show a few more fact slips
• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the files you download from Hugging Face, plus the drafts you feed a local runner
• Not for: languages other than English and Chinese

On Hugging Face, 13 November 2026

• Task: text generation
• Library: gguf
• Parameters: 11.9B
• Weights: about 6 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 26,382
• Likes: 65
• License: apache-2.0
• Languages: en, zh
• Based on: jialinyyzz/humanizer
• Tags: gguf, llama.cpp, quantized, imatrix, humanizer, text-rewriting, gemma4, text-generation, conversational

Links

• Model: https://huggingface.co/jialinyyzz/humanizer-GGUF
• Files: https://huggingface.co/jialinyyzz/humanizer-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download jialinyyzz/humanizer-GGUF

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