abacusai/Smaug-Humanizer-GGUF

Smaug Humanizer GGUF

GGUF files of abacusai/Smaug-Humanizer, a 12B model that rewrites AI-generated text so that it reads as if a person wrote it, while it keeps the facts, numbers and layout of the input. See the main model card for the training details.

Files

File	Size	Use
Smaug-Humanizer-BF16.gguf	23.8 GB	full precision
Smaug-Humanizer-Q8_0.gguf	12.7 GB	near full quality, fits a 16 GB GPU
Smaug-Humanizer-Q6_K.gguf	9.8 GB	12 GB GPUs
Smaug-Humanizer-Q5_K_M.gguf	8.5 GB	10–12 GB GPUs
Smaug-Humanizer-Q4_K_M.gguf	7.4 GB	8 GB GPUs and laptops

Made with llama.cpp b11899 from the bf16 weights.

Prompt and sampling

Use one user turn with this prompt:

Rewrite the AI-written text below so that it reads as if a person wrote it, in a person's own voice, wording and rhythm.

Keep every fact, number, name and quote. Keep the structure and the language of the text. Output only the rewritten text.

Text:
{text}

Sampling: temperature 1.0, top_p 0.95, top_k 64.

Stop string: add <end_of_turn> as a stop string. The model ends its turn with this text, which is not a special token in the GGUF, so most clients will otherwise print it at the end of the answer.

For texts over about 500 words, split the input at paragraph breaks into chunks of about 500 words, rewrite each chunk, and join them with a blank line.

llama.cpp

Use llama.cpp b11899 or newer (Gemma 4 support). Download from Hugging Face and serve:

llama-server -hf abacusai/Smaug-Humanizer-GGUF:Q4_K_M -ngl 99 -c 8192 --port 8080

Or serve a file you already downloaded:

hf download abacusai/Smaug-Humanizer-GGUF Smaug-Humanizer-Q4_K_M.gguf --local-dir .
llama-server -m Smaug-Humanizer-Q4_K_M.gguf -ngl 99 -c 8192 --port 8080

-ngl 99 puts all layers on the GPU; leave it out to run on the CPU.

curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
  "messages": [{"role": "user", "content": "<prompt above with your text>"}],
  "temperature": 1.0, "top_p": 0.95, "top_k": 64, "stop": ["<end_of_turn>"]
}'

Ollama

ollama run hf.co/abacusai/Smaug-Humanizer-GGUF:Q4_K_M

Add PARAMETER stop "<end_of_turn>" to a Modelfile if the tag shows at the end of the answers.

LM Studio

Search for abacusai/Smaug-Humanizer-GGUF, pick a file, and add <end_of_turn> under Stop Strings in the model settings.

License

Apache 2.0, the same as abacusai/Smaug-Humanizer and its base model, gemma-4-12B. See LICENSE.

Citation

@misc{abacusai2026smaughumanizer,
  title  = {Smaug Humanizer},
  author = {Abacus.AI},
  year   = {2026},
  url    = {https://huggingface.co/abacusai/Smaug-Humanizer}
}