Welcome to Jetson AGX Thor repository

8.Google Gemma3 visual multimodal large model

Gemma3

Gemma31. Model Size2. Performance3. Using Gemma33.1 Running Gemma33.2 Start a conversation3.3 Visual Function3.4 Ending the ConversationReferences

Gemma is a family of lightweight models built by Google based on Gemini technology. The Gemma 3 model is multimodal (capable of processing text and images), has a 128KB context window, and supports over 140 languages.

1. Model Size

Model	Size
gemma3:1b	815MB
gemma3:4b	3.3GB

2. Performance

Chatbot Arena ELO Score

3. Using Gemma3

3.1 Running Gemma3

Use the run command to run the model. If the model is not already downloaded, it will automatically pull the model from the Ollama model library:


ollama run gemma3:4b

3.2 Start a conversation


xxxxxxxxxx
How to learn a programming language?

Response time depends on hardware configuration, so please be patient!

3.3 Visual Function

test_pic


xxxxxxxxxx
What do you see in this picture? :./test_pic.png
#Using ": + the image path" in the conversation allows the model to use its visual function and interpret the information in the image.

3.4 Ending the Conversation

Use the Ctrl+d shortcut or /bye to end the conversation!

References

Ollama

Official Website: https://ollama.com/

GitHub: https://github.com/ollama/ollama

Gemma3

Ollama Compatible Model: https://ollama.com/library/gemma3

Alibaba Cloud Qwen2.5v visual multimodal large model

Lava visual multimodal large model