Instructions to use nvidia/AceReason-Nemotron-14B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nvidia/AceReason-Nemotron-14B with Transformers:

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

pipe = pipeline("text-generation", model="nvidia/AceReason-Nemotron-14B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("nvidia/AceReason-Nemotron-14B")
model = AutoModelForCausalLM.from_pretrained("nvidia/AceReason-Nemotron-14B")
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/AceReason-Nemotron-14B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nvidia/AceReason-Nemotron-14B"
# 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/AceReason-Nemotron-14B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/nvidia/AceReason-Nemotron-14B

SGLang

How to use nvidia/AceReason-Nemotron-14B 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/AceReason-Nemotron-14B" \
    --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/AceReason-Nemotron-14B",
		"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/AceReason-Nemotron-14B" \
        --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/AceReason-Nemotron-14B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use nvidia/AceReason-Nemotron-14B with Docker Model Runner:
```
docker model run hf.co/nvidia/AceReason-Nemotron-14B
```

Is it possible to open-source the 2k+ difficult samples from math stage3 separately, as well as the code training data?

by Suu - opened Jun 25, 2025

Discussion

Suu

Jun 25, 2025

Great work! Currently, we are reproducing your approach based on the phased methodology described in your paper. For math stage1 and stage2, we successfully achieved pass@64=61.8 on AIME2024. However, when proceeding to stage3 RL, our performance has remained stagnant at around 62, with no further improvement. Would it be possible for you to open-source the 2k+ difficult samples from math stage3? It would be even better if you could also open-source the code RL training data.

ychenNLP

NVIDIA org Jun 26, 2025

Hi @Suu ,

the 2K+ difficult samples are filtered from the current full dataset based on the model pass rate.
We used the stage2 model to generate 16 responses per problem and keep the questions that have pass rate <= 6/16.

Suu

Jun 27, 2025

I adopted the same approach and observed the reward in training decreasing from 0.2 to 0.05. However, unfortunately, my results only reached 62.6, and it was difficult to improve further.

ychenNLP

NVIDIA org Jul 2, 2025

•

edited Jul 2, 2025

@Suu
In our experiment (DeepSeek-r1-disitlled-7B), stage 3 training (24K length) with hard prompt, the training reward drop from 90% to 20% - pass@1 with average 64 runs improve to 65+ after 300 steps.

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