Latest Content
Gemma 4 E2B IT icon

Gemma 4 E2B IT

NVIDIA
Gemma 4 E2B IT is a multimodal dense transformer model designed for efficient on-device reasoning, coding, and agentic workflows. It features 2.3B effective parameters, around 5.1B including embeddings, with 35 layers, 1,536 hidden size, and 8 attention heads. The model uses hybrid attention with 512 token sliding window and global layers, supporting up to 128K context with proportional RoPE scaling. It integrates Per-Layer Embeddings for efficiency and supports text, image, video, and audio.
TypeOmni Model
CapabilitiesText Generation, Instruction Following, Reasoning, Mathematical Reasoning+7 more
Release Date02 April, 2026
Links
Blog|HF Model Card
LicenseApache 2.0

Inference Instructions

Deploy and run this model on NVIDIA B200 GPUs using the command below. Copy the command to get started with inference.

CONSOLE
docker run --gpus all 
 --shm-size 128g 
 -p 8000:8000 
 -v ~/.cache/huggingface:/root/.cache/huggingface 
 -e HF_TOKEN='YOUR_HF_TOKEN' 
 --ipc=host 
 lmsysorg/sglang:cu13-gemma4 
 python3 -m sglang.launch_server 
 --model-path google/gemma-4-E2B-it 
 --host 0.0.0.0 
 --port 8000 
 --max-prefill-tokens 65536 
 --tool-call-parser gemma4 
 --reasoning-parser gemma4 
 --max-running-requests 1024 
 --tp 4 
 --mem-fraction-static 0.95 
 --trust-remote-code
Note

Use the lmsysorg/sglang:gemma4 image (or later) for CUDA 12.9.

Model Benchmarks

Each model was tested with a fixed input size and total token volume while increasing concurrency to measure serving performance under load.

ITL vs Concurrency

Time to First Token

Throughput Scaling

Total Tokens/sec vs Avg TTFT

Vultr Cloud GPU

NVIDIA HGX B200

Deploy NVIDIA B200 on Vultr Cloud GPU

Deploy NVIDIA B200 instancesContact Sales