Instructions to use nvidia/Nemotron-Terminal-32B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nvidia/Nemotron-Terminal-32B with Transformers:

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

pipe = pipeline("text-generation", model="nvidia/Nemotron-Terminal-32B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("nvidia/Nemotron-Terminal-32B")
model = AutoModelForCausalLM.from_pretrained("nvidia/Nemotron-Terminal-32B")
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
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use nvidia/Nemotron-Terminal-32B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nvidia/Nemotron-Terminal-32B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nvidia/Nemotron-Terminal-32B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/nvidia/Nemotron-Terminal-32B

SGLang

How to use nvidia/Nemotron-Terminal-32B 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 "nvidia/Nemotron-Terminal-32B" \
    --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": "nvidia/Nemotron-Terminal-32B",
		"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 "nvidia/Nemotron-Terminal-32B" \
        --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": "nvidia/Nemotron-Terminal-32B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use nvidia/Nemotron-Terminal-32B with Docker Model Runner:
```
docker model run hf.co/nvidia/Nemotron-Terminal-32B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Nemotron-Terminal Model Family

Nemotron-Terminal is a family of models specialized for autonomous terminal interaction, fine-tuned from the Qwen3 (8B, 14B, and 32B). Developed by NVIDIA, these models utilize Nemotron-Terminal-Corpus, a large-scale open-source dataset for terminal tasks, to achieve performance that rivals frontier models many times their size.

Model Variants

We release the following variants of the Nemotron-Terminal family:

Nemotron-Terminal-8B
Nemotron-Terminal-14B
Nemotron-Terminal-32B

Performance on Terminal-Bench 2.0

The Nemotron-Terminal family demonstrates profound leaps in capability compared to the Qwen3 baselines across multiple specialized categories.

Model	Size	Base Accuracy	Nemotron-Terminal Accuracy
Nemotron-Terminal-8B	8B	2.47%	13.0%
Nemotron-Terminal-14B	14B	4.04%	20.2%
Nemotron-Terminal-32B	32B	3.37%	27.4%

Usage

The models are trained using the Terminus 2 scaffolding and output a structured JSON format. For evaluation on Terminal Bench 2.0, we encourage using Terminus 2 scaffolding to maintain consistency with training.

Expected Output Format

{
  "analysis": "Analysis of the current terminal state...",
  "plan": "Step-by-step plan for the next command...",
  "commands": [
    {
      "keystrokes": "ls -la\n",
      "duration": 0.1
    }
  ],
  "task_complete": false
}

📜 Citation

If you use this dataset in your research, please cite the following work:

@misc{pi2026dataengineeringscalingllm,
      title={On Data Engineering for Scaling LLM Terminal Capabilities}, 
      author={Renjie Pi and Grace Lam and Mohammad Shoeybi and Pooya Jannaty and Bryan Catanzaro and Wei Ping},
      year={2026},
      eprint={2602.21193},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2602.21193}, 
}