Instructions to use nvidia/Nemotron-Labs-Diffusion-8B-Base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nvidia/Nemotron-Labs-Diffusion-8B-Base with Transformers:

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

pipe = pipeline("text-generation", model="nvidia/Nemotron-Labs-Diffusion-8B-Base", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("nvidia/Nemotron-Labs-Diffusion-8B-Base", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use nvidia/Nemotron-Labs-Diffusion-8B-Base with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nvidia/Nemotron-Labs-Diffusion-8B-Base"
# 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-Labs-Diffusion-8B-Base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/nvidia/Nemotron-Labs-Diffusion-8B-Base

SGLang

How to use nvidia/Nemotron-Labs-Diffusion-8B-Base 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-Labs-Diffusion-8B-Base" \
    --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-Labs-Diffusion-8B-Base",
		"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-Labs-Diffusion-8B-Base" \
        --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-Labs-Diffusion-8B-Base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use nvidia/Nemotron-Labs-Diffusion-8B-Base with Docker Model Runner:
```
docker model run hf.co/nvidia/Nemotron-Labs-Diffusion-8B-Base
```

trias702 commited on Jan 20

Commit

456e96b

1 Parent(s): ff07748

Trying to force transformers to use the older causal mask

Browse files

Files changed (1) hide show

modeling_ministral.py +4 -2

modeling_ministral.py CHANGED Viewed

@@ -11,7 +11,7 @@ from transformers.cache_utils import Cache, DynamicCache
 from transformers.generation import GenerationMixin
 # from transformers.integrations import use_kernel_forward_from_hub, use_kernel_func_from_hub, use_kernelized_func
 from transformers.integrations import use_kernel_forward_from_hub
-from transformers.masking_utils import create_causal_mask, create_sliding_window_causal_mask
 from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
 from transformers.modeling_layers import (
     GenericForQuestionAnswering,
@@ -27,6 +27,7 @@ from transformers.utils import TransformersKwargs, auto_docstring, can_return_tu
 # from transformers.utils.generic import maybe_autocast
 from .configuration_ministral_dlm import MinistralDLMConfig
 def rotate_half(x):
     """Rotates half the hidden dims of the input."""
@@ -419,7 +420,8 @@ class Ministral3Model(Ministral3PreTrainedModel):
         if self.training:
             causal_mask = None
-        elif kwargs.get("use_causal_mask", False):
             mask_function = create_causal_mask if self.config.sliding_window is None else create_sliding_window_causal_mask
             causal_mask = mask_function(
                 config=self.config,

 from transformers.generation import GenerationMixin
 # from transformers.integrations import use_kernel_forward_from_hub, use_kernel_func_from_hub, use_kernelized_func
 from transformers.integrations import use_kernel_forward_from_hub
+from transformers.masking_utils import create_causal_mask, create_sliding_window_causal_mask, ALL_MASK_ATTENTION_FUNCTIONS, sdpa_mask_older_torch
 from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
 from transformers.modeling_layers import (
     GenericForQuestionAnswering,
 # from transformers.utils.generic import maybe_autocast
 from .configuration_ministral_dlm import MinistralDLMConfig
+ALL_MASK_ATTENTION_FUNCTIONS._global_mapping['sdpa'] = sdpa_mask_older_torch
 def rotate_half(x):
     """Rotates half the hidden dims of the input."""
         if self.training:
             causal_mask = None
+        #elif kwargs.get("use_causal_mask", False):
+        else:
             mask_function = create_causal_mask if self.config.sliding_window is None else create_sliding_window_causal_mask
             causal_mask = mask_function(
                 config=self.config,