ErtasAI/gemma-4-E2B-it

VISIONConcurrent Unit Cost:1Model Size:5.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 19, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

ErtasAI/gemma-4-E2B-it is a 5.1 billion parameter instruction-tuned multimodal language model developed by Google DeepMind, part of the Gemma 4 family. This model handles text, image, and audio inputs with a 32K token context window, featuring advanced reasoning capabilities and native function-calling support. Optimized for on-device deployment, it excels in tasks like text generation, coding, and multimodal understanding across various platforms.

Loading preview...

Model Overview

ErtasAI/gemma-4-E2B-it is an instruction-tuned variant of Google DeepMind's Gemma 4 E2B model, featuring 5.1 billion parameters and a 32K token context window. As part of the Gemma 4 family, it is a multimodal model capable of processing text, image, and audio inputs, and generating text outputs. The "E" in E2B signifies "effective" parameters, utilizing Per-Layer Embeddings (PLE) for efficient on-device deployment.

Key Capabilities

  • Multimodality: Processes text, images (with variable aspect ratio and resolution), and audio inputs.
  • Reasoning: Includes a built-in reasoning mode for step-by-step thought processes.
  • Long Context: Supports a context window of up to 128K tokens for E2B/E4B models.
  • Function Calling: Native support for structured tool use, enabling agentic workflows.
  • Coding: Enhanced capabilities for code generation, completion, and correction.
  • Optimized for On-Device: Specifically designed for efficient local execution on mobile devices and laptops.
  • Multilingual Support: Pre-trained on over 140 languages with out-of-the-box support for 35+ languages.

What Makes This Model Different?

This model is distinguished by its multimodal capabilities, handling text, image, and audio inputs, a feature not universally present in all LLMs. Its E2B variant is specifically optimized for efficient on-device deployment, making it suitable for environments with limited resources. The Gemma 4 family also introduces configurable thinking modes for enhanced reasoning and native function-calling support, improving its utility for agentic applications. The hybrid attention mechanism, combining local sliding window attention with global attention, allows for processing speed and a low memory footprint without sacrificing deep awareness for complex, long-context tasks.

Should I use this for my use case?

This model is ideal for applications requiring multimodal understanding, especially those targeting on-device deployment or scenarios where efficient local execution is critical. Its strong reasoning, coding, and agentic capabilities make it suitable for complex text generation, code assistance, and interactive AI agents. Developers needing robust multilingual support and the ability to process diverse input types (text, image, audio) will find this model particularly useful. However, for use cases solely focused on text generation without multimodal requirements, other specialized text-only models might be considered.

Finetunes
2 models