gemma-4-E4B-it

Gemma‑4‑E4B‑it is the second largest dense model in the new family of open LLMs from Google, offering solid performance with extremely economical resource consumption. The model uses the innovative Per‑Layer Embeddings (PLE) technique, which fundamentally changes the approach to building small language models. In standard transformers, each token receives a single embedding vector that passes through all network layers. PLE works differently: each of the 42 decoder layers gets its own small embedding per token. These embeddings are stored in large tables (the total model size reaches 8 billion parameters), but during inference only the effective part — 4.5 billion — is active.

This architecture allows the E4B model to achieve performance comparable to models two to three times larger. According to community feedback, E4B confidently surpasses Gemma‑3 27B in several tasks, even though its effective size is 12 times smaller. The model supports a context window of 128 thousand tokens and uses hybrid attention with a sliding window of 512 tokens. A key difference between E4B and the larger 31B model is built‑in audio support (an encoder of ~300M parameters), which makes the model universal — it can simultaneously process text, images, and sound.

Developers position E4B as a model for complex local tasks. It is ideally suited for use on high‑performance laptops, powerful mobile devices, and embedded systems. Thanks to the Apache 2.0 licence, the model can be freely fine‑tuned and integrated into commercial products that operate under tight memory constraints.

For the developers’ usage recommendations for the model, please refer to this link: https://ai.google.dev/gemma/docs/core/model_card_4?hl=en

Announce Date: 02.03.2026
Parameters: 8B
Context: 132K
Layers: 42, using full attention: 4, using no attention: 18
Attention Type: Sliding Window Attention
Developer: Google DeepMind
Transformers Version: 5.5.0.dev0
vLLM Version: gemma4
License: Apache 2.0

Use our pre-built public endpoints for free to test inference and explore gemma-4-E4B-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.