gemma-3-4b-it

Gemma 3 4B is a compact AI model developed by Google DeepMind, offering an excellent balance between size and functionality. Unlike the base 1B version, the 4B model supports multimodal inputs: text, images (with resolution up to 896x896 pixels), and short videos. For example, the model can recognize objects and text in images, such as extracting data from receipts or labels. To process images, it uses the SigLIP visual encoder, which automatically segments large files.

Its innovative architecture and efficient 5:1 ratio of local-to-global attention optimize memory usage while supporting a context window of up to 128K tokens. Gemma 3 4B supports 35 languages, including Russian. The model also includes function calling capabilities, enabling integration with APIs for task automation, such as generating SQL queries or transforming data.

This model is ideal for creating intelligent assistants and for fast document and image processing, making it a perfect choice for developers who need multimodal capabilities without requiring significant computational resources.

Announce Date: 12.03.2025
Parameters: 4B
Context: 132K
Layers: 34, using full attention: 5
Attention Type: Sliding Window Attention
Developer: Google DeepMind
Transformers Version: 4.50.0.dev0
License: gemma

Use our pre-built public endpoints for free to test inference and explore gemma-3-4b-it capabilities. You can obtain an API access token on the token management page after registration and verification.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at support@immers.cloud or send your request to the sales department at sale@immers.cloud.