Instructions to use vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16 with Transformers:

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

pipe = pipeline("text-generation", model="vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16")
model = AutoModelForCausalLM.from_pretrained("vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16", 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 vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "vcruz305/DeepNemotron-3.5-Lightning-Beta-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": "vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16

SGLang

How to use vcruz305/DeepNemotron-3.5-Lightning-Beta-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 "vcruz305/DeepNemotron-3.5-Lightning-Beta-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": "vcruz305/DeepNemotron-3.5-Lightning-Beta-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 "vcruz305/DeepNemotron-3.5-Lightning-Beta-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": "vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16 with Docker Model Runner:
```
docker model run hf.co/vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

DeepNemotron-3.5-Lightning (DeepLightning fullrun LoRA merged)

Merged checkpoint of the DeepLightning fullrun LoRA trained on nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 (NeMo Automodel, FSDP2, r=16 / alpha=32, exclude *.out_proj, dataset r0b0tlab/deepseek-v4-pro-0813-agentic, 2384 steps, final train loss 0.2096).

The LoRA was merged into the base weights (peft merge_and_unload) so this repo is a standalone dense BF16 model — no PEFT adapter needed at inference.

Architecture

Nemotron-H hybrid: Mamba + MoE (GB10 / DGX Spark class). Query the model with vLLM + --mamba-backend flashinfer for correct Mamba kernel dispatch.
dtype: bfloat16, max_model_len: 8192+.

Serving (recommended)

vllm serve . --trust-remote-code --mamba-backend flashinfer \
  --enable-auto-tool-choice --tool-call-parser hermes \
  --max-model-len 8192 --gpu-memory-utilization 0.85

(vLLM 0.1.dev1+; causal-conv1d / mamba-ssm fused kernels are NOT required when using the flashinfer mamba backend.)

Benchmarks (merged BF16, vLLM, DGX Spark GB10)

sixcat (limit=20, ~180 items), minus tools (harness expects OpenAI tool-call JSON; this Hermes-trained model emits Hermes-XML <tool_call> blocks — a wire-format mismatch, not a capability gap):

Category	Score	n
knowledge	80.0	64/80
math	90.0	18/20
truth	90.0	18/20
instruct	80.0	16/20
code	90.0	18/20
tools*	10.0	2/20
overall	75.6	136/180

hermes-agentic-bench loop_gate (20 tasks): pass_rate 0.45, mean_tools 0.0, n_hit_cap 0, n_parse_fail_tasks 17. As with sixcat tools, the model emits correct Hermes-XML tool calls that the OpenAI-style checker does not parse — parse_ok=False whenever a tool call is present.

*tools/agentic parse failures are a harness wire-format artifact (Hermes-XML vs OpenAI JSON), not a model defect. The model selects the correct tool and args.

Files

model-00001-of-00002.safetensors, model-00002-of-00002.safetensors
config.json, generation_config.json, tokenizer*.json, chat_template.jinja

Downloads last month: 846

Safetensors

Model size

32B params

Tensor type

F32

BF16

Model tree for vcruz305/DeepNemotron-3.5-Lightning-Beta-BF16

Base model

nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16

Adapter

(4)

this model

Adapters

2 models

Quantizations

1 model