Instructions to use google/shieldgemma-2-4b-it with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use google/shieldgemma-2-4b-it with Transformers:

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

pipe = pipeline("image-text-to-text", model="google/shieldgemma-2-4b-it")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoModelForImageClassification
model = AutoModelForImageClassification.from_pretrained("google/shieldgemma-2-4b-it", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use google/shieldgemma-2-4b-it with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "google/shieldgemma-2-4b-it"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "google/shieldgemma-2-4b-it",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/google/shieldgemma-2-4b-it

SGLang

How to use google/shieldgemma-2-4b-it 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 "google/shieldgemma-2-4b-it" \
    --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": "google/shieldgemma-2-4b-it",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "google/shieldgemma-2-4b-it" \
        --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": "google/shieldgemma-2-4b-it",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use google/shieldgemma-2-4b-it with Docker Model Runner:
```
docker model run hf.co/google/shieldgemma-2-4b-it
```

how do i interpret the results

by cuiyi0326 - opened Mar 19, 2025

Discussion

cuiyi0326

Mar 19, 2025

I'm using the default policies and got the following results. How should i interpret the results? if i understand correctly, the logits tensor shape (3,2) corresponds to 3 default policies and every police has two values yes and no. so my question is why do i always get two [nan, nan]?

ShieldGemma2ImageClassifierOutputWithNoAttention(loss=None, logits=tensor([[ nan, nan],
[ nan, nan],
[35.1562, 53.1972]], device='cuda:0'), hidden_states=None, probabilities=tensor([[ nan, nan],
[ nan, nan],
[1.4618e-08, 1.0000e+00]], device='cuda:0'))

BITDDD

Mar 24, 2025

Use Latest Version Transformers，I get results like this: tensor([[5.3998e-02, 9.4600e-01],
[3.6518e-01, 6.3482e-01],
[1.8238e-08, 1.0000e+00]])，The probability of looking pornographic, dangerous, bloody and violent

merve

Mar 25, 2025

Hello, the policy outputs scores for Yes and No respectively for each policy (policy is provided by text as you put) @BITDDD
@cuiyi0326 can you put your inference code here for me to reproduce?

merve

Mar 26, 2025

Hello again, I built a notebook on how to properly use ShieldGemma 2 with custom policies: https://github.com/merveenoyan/smol-vision/blob/main/ShieldGemma_2_for_Vision_LM_Safety.ipynb

iampelle

May 22, 2025

Use Latest Version Transformers，I get results like this: tensor([[5.3998e-02, 9.4600e-01],
[3.6518e-01, 6.3482e-01],
[1.8238e-08, 1.0000e+00]])，The probability of looking pornographic, dangerous, bloody and violent

Here the order is "dangerous", "sexual", "violence", but I'm not sure. I guess inputing images of the three categories is the only way to find out?

lkv

Google org Sep 23, 2025

Hi @cuiyi0326 ,

Apologies for the late reply, The reason your seeing [nan, nan] for two of the policies is most likely a bug in the older version of the transformers library your are using. This is known issue, and it means the model was unable to compute a valid output for those policies, which is why the values show up as "Not a Number".
The issue has been already identified and fixed in release of Latest version transformers.

Kindly use Latest Version pip install --upgrade transformers and let us know still if you have any concerns will assist you on this.

Thank you.

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