o-unrestricted v1.0

Publication status: the prepared GGUF upload is pending. Until it is available here, use the reproducible model-preparation scripts in the linked GitHub repository.

o is an independent chat application powered by a Qwen3.8-27B RVN derivative. It is not affiliated with OpenAI and is not an OpenAI model or leak. “Unrestricted” is the application label; it is not a guarantee about model behavior.

The source is 0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF, based on Tim Rohrbaugh's ARA derivative of Qwen3.8-27B. Attribution and the Apache 2.0 license are retained.

Property	Value
Parameters	27B, as reported by upstream
Quantization	RVN Q4_K_M multilingual
GGUF architecture	qwen35
Weight size	16,547,400,224 bytes (15.41 GiB)
Source context metadata	262,144 tokens
Pilot serving context	4,096 tokens
Display name	o-unrestricted
License	Apache-2.0; see UPSTREAM-LICENSE

The only GGUF change is the display name in general.name; tensors, tokenizer, architecture, and embedded chat template remain unchanged. SOURCE.json pins the source revision and both checksums. The preparation script verifies the upstream checksum and unchanged tensor payload.

Run

Install the Hugging Face CLI and Ollama. Downloading requires space for the 15.41 GiB file. Ollama also maintains its own model store.

hf download Poopbutt1234/o-unrestricted --local-dir .
shasum -a 256 -c o-unrestricted-v1.gguf.sha256
# Start Ollama first if its app/service is not already running:
ollama serve
# In another terminal, in the downloaded directory:
ollama create o-unrestricted -f Modelfile
ollama run o-unrestricted

The name is local after creation; it is not a public Ollama registry model. The application source includes a streaming web UI, Docker Compose deployment, and reproducible model download/preparation scripts.

Use a GPU with at least 24 GB memory for this Q4 pilot. A 16 GiB Mac does not have adequate headroom for this 15.41 GiB model plus the OS, KV cache, and runtime buffers. Model context capacity does not imply that this deployment can serve the full context window.

The Modelfile supplies a concise, direct o persona. It instructs the model to say “I am o” when asked for its name, and to explain the underlying Qwen derivative when asked about provenance. The upstream embedded template contains no default “I am Qwen” or “You are Qwen” system identity. The original chat template remains embedded unchanged.

For redistribution, include this card, the Modelfile, SOURCE.json, UPSTREAM-LICENSE, and the renamed GGUF with its generated SHA-256 sidecar. Publish under a namespace you control.