Instructions to use nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 with Transformers:

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

pipe = pipeline("text-generation", model="nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", trust_remote_code=True)
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
Local Apps Settings

vLLM

How to use nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16"
# 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/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16

SGLang

How to use nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 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/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16" \
    --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/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16",
		"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/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16" \
        --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/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 with Docker Model Runner:
```
docker model run hf.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
```

suhara commited on Dec 15, 2025

Commit

4a31ff1

verified ·

1 Parent(s): 0d1698d

Upload modeling_nemotron_h.py

Browse files

Files changed (1) hide show

modeling_nemotron_h.py +4 -3

modeling_nemotron_h.py CHANGED Viewed

@@ -623,8 +623,8 @@ class NemotronHMamba2Mixer(nn.Module):
             hidden_states = hidden_states.reshape(batch_size, seq_len, -1, self.head_dim).float()
             B = B.reshape(batch_size, seq_len, -1, self.ssm_state_size).float()
             C = C.reshape(batch_size, seq_len, -1, self.ssm_state_size).float()
-            B = B.repeat(1, 1, self.num_heads // self.n_groups, 1)
-            C = C.repeat(1, 1, self.num_heads // self.n_groups, 1)
             pad_size = (self.chunk_size - seq_len % self.chunk_size) % self.chunk_size
             D_residual = self.D[..., None] * pad_tensor_by_size(hidden_states, pad_size)
@@ -852,7 +852,8 @@ class NemotronHMOE(nn.Module):
                 final_hidden_states.index_add_(0, token_indices, weighted_output)
             else:
                 # Local empty expert: no-op compute that still marks params as used.
-                dummy_out = expert(torch.zeros_like(hidden_states[0]).unsqueeze(0).to(final_hidden_states.dtype))
                 final_hidden_states = final_hidden_states + dummy_out
         # in original deepseek, the output of the experts are gathered once we leave this module

             hidden_states = hidden_states.reshape(batch_size, seq_len, -1, self.head_dim).float()
             B = B.reshape(batch_size, seq_len, -1, self.ssm_state_size).float()
             C = C.reshape(batch_size, seq_len, -1, self.ssm_state_size).float()
+            B = B.repeat_interleave(self.num_heads // self.n_groups, dim=2, output_size=self.num_heads)
+            C = C.repeat_interleave(self.num_heads // self.n_groups, dim=2, output_size=self.num_heads)
             pad_size = (self.chunk_size - seq_len % self.chunk_size) % self.chunk_size
             D_residual = self.D[..., None] * pad_tensor_by_size(hidden_states, pad_size)
                 final_hidden_states.index_add_(0, token_indices, weighted_output)
             else:
                 # Local empty expert: no-op compute that still marks params as used.
+                expert_dtype = expert.down_proj.weight.dtype
+                dummy_out = expert(torch.zeros_like(hidden_states[0]).unsqueeze(0).to(expert_dtype))
                 final_hidden_states = final_hidden_states + dummy_out
         # in original deepseek, the output of the experts are gathered once we leave this module