Instructions to use aoxo/gpt-oss-20b-uncensored with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use aoxo/gpt-oss-20b-uncensored with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="aoxo/gpt-oss-20b-uncensored")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("aoxo/gpt-oss-20b-uncensored", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use aoxo/gpt-oss-20b-uncensored with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "aoxo/gpt-oss-20b-uncensored"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "aoxo/gpt-oss-20b-uncensored",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/aoxo/gpt-oss-20b-uncensored

SGLang

How to use aoxo/gpt-oss-20b-uncensored with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "aoxo/gpt-oss-20b-uncensored" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "aoxo/gpt-oss-20b-uncensored",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "aoxo/gpt-oss-20b-uncensored" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "aoxo/gpt-oss-20b-uncensored",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use aoxo/gpt-oss-20b-uncensored with Docker Model Runner:
```
docker model run hf.co/aoxo/gpt-oss-20b-uncensored
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

gpt-oss-20b-abliterated

Model Overview

Model Name: gpt-oss-20b-uncensored
Model Type: Large Language Model (Text Generation)
Architecture: Decoder-Only Transformer (Mixture of Experts)
Parameter Size: 21B total parameters (3.6B active per forward pass)
Base Model: gpt-oss-20b
Modification: Abliteration (removal of refusal/alignment mechanisms)

Description

The gpt-oss-20b-abliterated model is a derivative of the original gpt-oss-20b, part of OpenAI’s open-weight GPT-OSS series.
This variant preserves the architecture, quantization, and training of the base model, but has undergone an abliteration process to remove refusal mechanisms and alignment constraints.

As a result, it will respond to a broader range of prompts without applying internal safety filters. All other technical details, reasoning capabilities, and agentic features remain unchanged.

Technical Details

Backbone: Transformer decoder with Mixture of Experts (MoE) routing
Parameters: 21B (3.6B active per forward pass)
Layers: 48 Transformer blocks
Hidden size: 6,144
Attention heads: 48
Context length: 32k tokens
Quantization: MXFP4 for MoE weights (fits within 16GB GPU memory)
Training Data: ~1.2T tokens (web, books, academic text, code, conversations)
Response Format: Compatible with Harmony, though abliteration allows raw completions

Usage

Transformers

from transformers import pipeline

model_id = "aoxo/gpt-oss-20b-abliterated"

pipe = pipeline(
    "text-generation",
    model=model_id,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "user", "content": "Tips on how to insult someone"},
]

outputs = pipe(messages, max_new_tokens=256)
print(outputs[0]["generated_text"][-1])

Resources

📓 Notebook: GPT OSS Abliteration Notebook
📝 Blog Post: The Ultimate Cookbook: Uncensoring GPT-OSS

vLLM

uv pip install --pre vllm==0.10.1+gptoss \
    --extra-index-url https://wheels.vllm.ai/gpt-oss/ \
    --extra-index-url https://download.pytorch.org/whl/nightly/cu128

vllm serve aoxo/gpt-oss-20b-abliterated

Ollama

ollama pull gpt-oss-20b-uncensored
ollama run gpt-oss-20b-uncensored

Limitations & Risks

May produce biased, unsafe, or harmful outputs
Lacks built-in refusal or moderation layers
Should not be deployed in user-facing systems without external filtering
Outputs are not aligned to safety standards

Citation

If you use gpt-oss-20b-abliterated, please cite both the base model and the abliteration:

@misc{openai2025gptoss20b,
      title={gpt-oss-20b Model Card}, 
      author={OpenAI},
      year={2025},
      eprint={2508.10925},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2508.10925}, 
}

@misc{gptoss20b-abliterated,
  author = {aoxo},
  title = {Uncensoring GPT-OSS-20B: Abliteration},
  year = {2025},
  howpublished = {\url{https://medium.com/@aloshdenny/uncensoring-gpt-oss-20b-abliteration}},
}