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

Libraries

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

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

pipe = pipeline("text-generation", model="openai/gpt-oss-20b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("openai/gpt-oss-20b")
model = AutoModelForCausalLM.from_pretrained("openai/gpt-oss-20b", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
AMD Developer Cloud
Local Apps Settings

vLLM

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

Install from pip and serve model

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

Use Docker

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

SGLang

How to use openai/gpt-oss-20b 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 "openai/gpt-oss-20b" \
    --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": "openai/gpt-oss-20b",
		"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 "openai/gpt-oss-20b" \
        --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": "openai/gpt-oss-20b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use openai/gpt-oss-20b with Docker Model Runner:
```
docker model run hf.co/openai/gpt-oss-20b
```

triton_kernels and multiprocessing

#74

by piotrm0 - opened Aug 7, 2025

Discussion

piotrm0

Aug 7, 2025

Has anyone been able to share the quantized model across multiple processes in python (i.e. using multiprocessing or torch.multiprocessing)? I'm myself getting issues with pickling/unpickling related to triton_kernels:

ModuleNotFoundError: No module named 'triton_kernels_b8dc79e809df14c1'
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/opt/conda/lib/python3.12/multiprocessing/spawn.py", line 122, in spawn_main
    exitcode = _main(fd, parent_sentinel)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.12/multiprocessing/spawn.py", line 132, in _main
    self = reduction.pickle.load(from_parent)

Note that I'm able to use the model fine from the main process and can share other models on CUDA with child processes (via spawn) but not this one. Anyone seen this? (and better yet, have a solution?)

reach-vb

OpenAI org Aug 8, 2025

•

edited Aug 8, 2025

We packaged the triton_kernels better, if you use the main, it should be a much better experience.
https://github.com/huggingface/transformers/pull/39926

piotrm0

Aug 16, 2025

Thank you for looking into it. I got the latest versions of transformers and kernels from github but am still seeing the issue. Here is a minimal example to test:

from transformers import AutoModelForCausalLM
from torch.multiprocessing import get_context
import torch
import os

MODEL_NAME = "openai/gpt-oss-20b"
DEVICE = "cuda"

spawn_context = get_context("spawn")


def use_model(model):
    print(os.getpid(), model.device)
    print(
        model.generate(
            input_ids=torch.tensor([[0, 1, 2]], device=model.device), max_new_tokens=1))


if __name__ == "__main__":
    model = AutoModelForCausalLM.from_pretrained(MODEL_NAME,
                                                 torch_dtype="auto",
                                                 device_map=DEVICE)

    use_model(model)
    spawn_context.Process(target=use_model, args=(model, )).start()

Here are my versions (I removed triton_kernels just in case it was interfering in some way):

kernels                                  0.9.0.dev0
transformers                             4.56.0.dev0
triton                                   3.4.0
# at these commits (main branch head as of Sat Aug 16):
  - Installing transformers (4.56.0.dev0 cd22550)
  - Installing kernels (0.9.0.dev0 1caa4c1)

Are you able to reproduce?

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment