Gemma 4 E2B
Parameters
5.1B
Context Length
128K
Modality
Multimodal
Auxiliary Parameters
400M
Architecture
Dense
KV Cache Rate
35 KB/tok
License
Apache 2.0
Release Date
2 Apr 2026
Knowledge Cutoff
-
API Pricing (per 1M)
Self-hosted only
Hardware & System Requirements
Precision
Context
Workload Memory: 5.15 GB
INT4 · 8K context
Base Weights: 2.37 GB
KV Cache: 0.27 GB
Runtime & Graphs: 2.50 GB
Minimum Required Hardware
1x NVIDIA L4 24GB
5.1 / 21.6 GB usable
1x RTX 3060 12GB
5.1 / 10.8 GB usable
Apple M4 24GB
5.1 / 18.0 GB usable
Determine requirements for your specific workload:
Open in VRAM Calculator
Runtime requirements
Checkpoint format
FP16
CUDA
>= 11.8
PyTorch
>= 2.1
Transformers
>= 5.5.0.dev0
| Architecture | Precision | Attention kernel |
|---|---|---|
Blackwell, Hopper | Native FP16 | FlashAttention-3 |
Ada Lovelace, Ampere | Native FP16 | FlashAttention-2 |
Architecture Diagram
Evaluation Benchmarks
Rank
#200
| Benchmark | Score | Rank |
|---|---|---|
Professional Knowledge | 0.6 | 55 |
Graduate-Level QA | 0.434 | 112 |
Coding Index | 0.07 | 151 |
Intelligence Index | auto 0.08 Standard 0.07 | 261 288 |
Rankings
Overall Rank
#200
Coding Rank
#161
About Gemma 4 E2B
Gemma 4 E2B is an ultra-efficient model with 2.3B effective parameters (5.1B with Per-Layer Embeddings) designed for mobile and IoT devices. Supports text, image, and audio input with 128K context window, delivering frontier capabilities on edge devices with near-zero latency and offline operation. Features built-in reasoning mode and native function calling for agentic workflows.
Specifications
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
8
Key-Value Heads
1
Attention Head Dimension
256
Position Embedding
ROPE
RoPE Theta
10,000
Sliding Window Attention
Yes
Sliding Window Size
512
Sliding Window Ratio
83.3%
Normalization
RMS Normalization
Activation Function
GELU
Dimensions
Base Precision
FP16
Auxiliary Parameters
400M
Hidden Dimension Size
6,144
Number of Layers
35
FFN Intermediate Size (Dense)
6,144
Tokenizer
Vocabulary Size
262,144
About Gemma 4
Gemma 4 is Google DeepMind's most advanced open model family, built from Gemini 3 research and technology. Featuring both Dense and Mixture-of-Experts (MoE) architectures, these multimodal models handle text, images, and audio (on smaller variants), with context windows up to 256K tokens. Designed for frontier-level performance across reasoning, coding, and agentic workflows, Gemma 4 delivers unprecedented intelligence-per-parameter from mobile devices to enterprise servers. Released under Apache 2.0 license.