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

Libraries

How to use nvidia/NFT-32B with Transformers:

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

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

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("nvidia/NFT-32B")
model = AutoModelForCausalLM.from_pretrained("nvidia/NFT-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/NFT-32B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nvidia/NFT-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/NFT-32B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/nvidia/NFT-32B

SGLang

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

Docker Model Runner
How to use nvidia/NFT-32B with Docker Model Runner:
```
docker model run hf.co/nvidia/NFT-32B
```

Haoxiang-Wang commited on Jul 15, 2025

Commit

1b7ed31

verified ·

1 Parent(s): f8b463e

Update README.md

Browse files

Files changed (1) hide show

README.md +4 -4

README.md CHANGED Viewed

@@ -133,7 +133,7 @@ Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated sys
 ## Training Method
-![NFT Method](../assets/method_NFT.jpg)
 The NFT training pipeline consists of three main components:
@@ -147,7 +147,7 @@ The NFT training pipeline consists of three main components:
 L_NFT(θ) = r[-log(π_θ⁺(a|q) / π(a|q))] + (1-r)[-log((1 - r_q * (π_θ⁺(a|q) / π(a|q))) / (1-r_q))]
 ```
-![Policy Distribution](../assets/distribution_NFT.jpg)
 ## Training Datasets
@@ -176,7 +176,7 @@ NFT-32B is evaluated on 6 mathematical reasoning benchmarks:
 ## Performance
-![Performance Comparison](../assets/main_compare_NFT.jpg)
 NFT-32B achieves state-of-the-art performance among supervised learning methods for mathematical reasoning:
@@ -192,7 +192,7 @@ NFT-32B achieves state-of-the-art performance among supervised learning methods
 Notably, NFT-32B performs similarly to DAPO (59.2% vs 59.9%) while using a simpler supervised learning approach.
-![Validation Curves](../assets/val_acc_curve_NFT.jpg)
 ## Usage

 ## Training Method
+![NFT Method](./assets/method_NFT.jpg)
 The NFT training pipeline consists of three main components:
 L_NFT(θ) = r[-log(π_θ⁺(a|q) / π(a|q))] + (1-r)[-log((1 - r_q * (π_θ⁺(a|q) / π(a|q))) / (1-r_q))]
 ```
+![Policy Distribution](./assets/distribution_NFT.jpg)
 ## Training Datasets
 ## Performance
+![Performance Comparison](./assets/main_compare_NFT.jpg)
 NFT-32B achieves state-of-the-art performance among supervised learning methods for mathematical reasoning:
 Notably, NFT-32B performs similarly to DAPO (59.2% vs 59.9%) while using a simpler supervised learning approach.
+![Validation Curves](./assets/val_acc_curve_NFT.jpg)
 ## Usage