Instructions to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with Transformers:

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

pipe = pipeline("text-generation", model="SerialKicked/GPT-OSS-20B-Heretic-GGUF")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("SerialKicked/GPT-OSS-20B-Heretic-GGUF", dtype="auto")

llama-cpp-python

How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="SerialKicked/GPT-OSS-20B-Heretic-GGUF",
	filename="gpt-oss-20b-heretic-IQ4_NL.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL
# Run inference directly in the terminal:
llama-cli -hf SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL
# Run inference directly in the terminal:
llama-cli -hf SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL
# Run inference directly in the terminal:
./llama-cli -hf SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL
# Run inference directly in the terminal:
./build/bin/llama-cli -hf SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL

Use Docker

docker model run hf.co/SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL

LM Studio
Jan

vLLM

How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SerialKicked/GPT-OSS-20B-Heretic-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SerialKicked/GPT-OSS-20B-Heretic-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL

SGLang

How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF 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 "SerialKicked/GPT-OSS-20B-Heretic-GGUF" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SerialKicked/GPT-OSS-20B-Heretic-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "SerialKicked/GPT-OSS-20B-Heretic-GGUF" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SerialKicked/GPT-OSS-20B-Heretic-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with Ollama:
```
ollama run hf.co/SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL
```

Unsloth Studio new

How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for SerialKicked/GPT-OSS-20B-Heretic-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for SerialKicked/GPT-OSS-20B-Heretic-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for SerialKicked/GPT-OSS-20B-Heretic-GGUF to start chatting

Pi new

How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL

Run Hermes

hermes

Docker Model Runner
How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with Docker Model Runner:
```
docker model run hf.co/SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL
```

Lemonade

How to use SerialKicked/GPT-OSS-20B-Heretic-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull SerialKicked/GPT-OSS-20B-Heretic-GGUF:IQ4_NL

Run and chat with the model

lemonade run user.GPT-OSS-20B-Heretic-GGUF-IQ4_NL

List all available models

lemonade list

Quantized versions of coder3101/gpt-oss-20b-heretic. This is a new abliteration method that doesn't seem to be as damaging to the base model as the previous methods were.

WARNING: Due to the nature of GPT-OSS, and its MXFP4 weight format, alterations (tuning, abliteration, quant) have to be done very carefully. The person who made the abliteration pass with Heretic erased the MXFP4 format (maybe a requirement for Heretic, maybe not, I have no idea). This means that my quantization (and any other for this model) is an approximation of an approximation of GPT-OSS's weights.

As a result, do not expect a quality comparable to the official non-abliterated model. This is a toy, not a production-ready model.

The repo includes the following quantized versions (Q8 or MXFP4 recommended, the other variants are not stable):

Okay Models

Q8_0: This should be decent-enough. Allows for 8-12K context on 24GB VRAM.
MXFP4: This is NOT a fix to the problem, and will behave worse than Q8, but it's lightweight and seems more stable than IQ4_NL (didn't test much, ngl)

Experimental Models (they do weird shit)

Q5_1: Technically it's Q5_K_M but in practice it's Q5_1 with Q8 attention layers. Unstable, not recommended.
IQ4_NL: Should work, but will be worse than what IQ4_NL would be on other models, while using a bit more VRAM. Unstable, not recommended.

Why No Qx_K_x Models?!

Simply put, K-type quantization schemes are not compatible with GPT-OSS's architecture, at least not without serious voodoo magic. 90% of the layers can't use those K-type formats, so llama.cpp's quantization tool automatically converts them to what works, usually Q8 (or Q5_1).

That's why when you look for quantized versions of GPT-OSS, you'll see tons of repos with supposedly different quants but with the same filesize everywhere.

Downloads last month: 30

GGUF

Model size

21B params

Architecture

gpt-oss

Hardware compatibility

4-bit

5-bit

8-bit

View +1 variant

Model tree for SerialKicked/GPT-OSS-20B-Heretic-GGUF

Base model

openai/gpt-oss-20b

Finetuned

coder3101/gpt-oss-20b-heretic

Quantized

(1)

this model