Instructions to use nvidia/Nemotron-Cascade-2-30B-A3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nvidia/Nemotron-Cascade-2-30B-A3B with Transformers:

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

pipe = pipeline("text-generation", model="nvidia/Nemotron-Cascade-2-30B-A3B", 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/Nemotron-Cascade-2-30B-A3B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("nvidia/Nemotron-Cascade-2-30B-A3B", trust_remote_code=True, 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use nvidia/Nemotron-Cascade-2-30B-A3B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nvidia/Nemotron-Cascade-2-30B-A3B"
# 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-Cascade-2-30B-A3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/nvidia/Nemotron-Cascade-2-30B-A3B

SGLang

How to use nvidia/Nemotron-Cascade-2-30B-A3B 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-Cascade-2-30B-A3B" \
    --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-Cascade-2-30B-A3B",
		"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-Cascade-2-30B-A3B" \
        --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-Cascade-2-30B-A3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use nvidia/Nemotron-Cascade-2-30B-A3B with Docker Model Runner:
```
docker model run hf.co/nvidia/Nemotron-Cascade-2-30B-A3B
```

Nemotron Cascade mutates nested path literals in tool arguments even with explicit literal-preservation instructions and temperature=0.

#23

by joonsoo-me - opened Mar 31

Discussion

joonsoo-me

Mar 31

Nemotron Cascade Tool-Calling Issue Report

It can emit valid tool calls for simple flat schemas, but it does not reliably preserve exact path literals inside nested tool arguments.

The failure persisted after:

strengthening the Jinja template with explicit literal-preservation language
adding a negative example
switching sampling to temperature: 0

The result is a model-side fidelity problem for agentic tool use, especially for file/repo operations where exact paths must not drift.

Environment

Host under test: dgx spark gb10
Runtime: llama.cpp
llama.cpp commit: 6861f6509a2b
Model: Nemotron-Cascade-2-30B-A3B-MXFP4_MOE_F16.gguf - almost same for automation job
Chat template: nemotron-opencode-custom-v0.1.jinja - keep base template and added more functions to augment tool-calling
Reasoning: off
Tested sampling:
- test 1 : temperature: 0.35
- test 2 : temperature: 0

What Worked

Flat tool calling is functional.

Test	Result
Flat `calculator` tool call	`finish_reason=tool_calls`
Flat tool round-trip	Correct final answer returned

Example:

calculator(128 * 37)

Observed output:

finish_reason=tool_calls
The product is 4736

What Failed

Nested object arguments do not preserve exact path literals.

Reproduction Case

Tool schema conceptually included a nested object like:

{
  "type": "object",
  "properties": {
    "path": { "type": "string" },
    "options": {
      "type": "object",
      "properties": {
        "recursive": { "type": "boolean" },
        "include_hidden": { "type": "boolean" }
      },
      "required": ["recursive"]
    }
  },
  "required": ["path", "options"]
}

Prompt intent:

Use the search_files tool. Preserve exact path literals character-for-character.

Expected behavior:

/home/username1234/llama.cpp

Observed behavior (critical):

/home/username1235/llama.cpp

After re-running with temperature: 0, the drift still occurred:

/home/username1234/llama.cpp
→
/home/username1236/llama.cpp

Impact

This is safe enough for tool selection, but not safe for exact-path automation .

Affected workflows:

file search
file edit
repository navigation
any agent flow that must preserve literal paths, IDs, or filenames exactly

Jinja Hardening Already Tried

The following template-level changes were already tested and did not fix the issue:

stronger explicit literal-preservation instructions
negative example for path copying
generic OpenCode-style tool-call formatting
non-reasoning default
temperature: 0

Conclusion

The model is capable of basic tool calling, but nested tool arguments are not reliable enough for path-sensitive agentic use .

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