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

Libraries

How to use trl-lib/gpt-oss-20b-chess with Transformers:

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

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

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("trl-lib/gpt-oss-20b-chess")
model = AutoModelForCausalLM.from_pretrained("trl-lib/gpt-oss-20b-chess", 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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use trl-lib/gpt-oss-20b-chess with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/trl-lib/gpt-oss-20b-chess

SGLang

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

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

gpt-oss-20b-chess

gpt-oss-20b fine-tuned with TRL on 500,000 Lichess puzzles from trl-lib/chess-puzzles, then quantized back to MXFP4, so it serves like the original model (14 GB). It was trained and evaluated with the TRL example sft_gpt_oss_chess.

Given a puzzle in the prompt format of the dataset (the board, its FEN and the legal moves), the model answers with the winning move in UCI notation, without reasoning.

Model	Test puzzles solved	Puzzle rating
gpt-oss-20b, low reasoning effort	14.8%	764
gpt-oss-20b-chess	69.9%	1813

Both models are measured on the 1,000 test puzzles of the dataset. A puzzle counts as solved on its first move, so the puzzle rating is higher than the model would get on Lichess, where the whole solution is required.

vllm serve trl-lib/gpt-oss-20b-chess

Downloads last month: -

Safetensors

Model size

21B params

Tensor type

F32

BF16

Model tree for trl-lib/gpt-oss-20b-chess

Base model

openai/gpt-oss-20b

Quantized

(259)

this model

trl-lib
/

gpt-oss-20b-chess

gpt-oss-20b-chess

Model tree for trl-lib/gpt-oss-20b-chess

Dataset used to train trl-lib/gpt-oss-20b-chess