Instructions to use google/gemma-3-12b-it with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use google/gemma-3-12b-it with Transformers:

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

pipe = pipeline("image-text-to-text", model="google/gemma-3-12b-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 AutoProcessor, AutoModelForImageTextToText

processor = AutoProcessor.from_pretrained("google/gemma-3-12b-it")
model = AutoModelForImageTextToText.from_pretrained("google/gemma-3-12b-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?"}
        ]
    },
]
inputs = processor.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(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use google/gemma-3-12b-it with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "google/gemma-3-12b-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/gemma-3-12b-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/gemma-3-12b-it

SGLang

How to use google/gemma-3-12b-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/gemma-3-12b-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/gemma-3-12b-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/gemma-3-12b-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/gemma-3-12b-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/gemma-3-12b-it with Docker Model Runner:
```
docker model run hf.co/google/gemma-3-12b-it
```

Image-to-text (OCR) functionality omits top-most line for recognition / output

by snowboarder04 - opened Mar 13, 2025

Discussion

snowboarder04

Mar 13, 2025

•

edited Mar 13, 2025

I tried uploading a PNG containing handwritten notes using LM Studio and it returned the text but in almost all cases the top line of text from the document is missing, especially when near a possible margin.

I have tried several docs, with varying results which are unique to this model.

Other more purpose-trained models, such as olmocr-7b DO recognise / output the topmost line.

Unlimitedlevel19

Mar 15, 2025

spongebob7

Mar 23, 2025

Same here. It omits either the first top line or the bottom line.

Gjjgbj

Apr 26, 2025

•

edited Apr 26, 2025

BalakrishnaCh

Google org Oct 28, 2025

Hi,

Apologies for the late replay, thanks for reaching out to us. The above issue might be because of any of the following reasons.

Fixed Resolution and Resizing: The Gemma 3 vision encoder operates at a fixed resolution (896 X 896 pixels). When you upload a high-resolution PNG, especially one with a non-square aspect ratio, the image must be resized, scaled, or cropped to fit this square input size.
Crop Heuristics (Loss of Edge Detail): The pre-processing pipeline (even with techniques like Pan & Scan) can use heuristics to identify the main content and crop aggressively. If your handwritten notes are placed right against the margin of the PNG canvas, this automatic cropping or resizing process can mistake the edge-bound text for unimportant border information and discard it.

Recommended Workaround:

Since the issue is with the input-handling of the gemma-3-12b-it model , the most reliable fix is to modify your input image:

Add a Wide Margin: Before uploading the PNG to LM Studio, open it in an image editor and add a generous white border/margin (e.g 10-20% of the image height) to the top, bottom, and sides.
Objective: By placing the handwritten text clearly away from the physical edge of the PNG file, you prevent the model's internal cropping/resizing from accidentally cutting off the first line.

Thanks.

Fradeck619

Dec 1, 2025

•

edited Dec 1, 2025

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