Gemma 4 E2B IT (LiteRT-LM)
High-performance, multimodal on-device vision-language model (image-text-to-text) optimized with Google LiteRT-LM.
Model Files & Hardware Tiers
| Hardware Target | File Path | Engine / EP | Size | Description |
|---|---|---|---|---|
| Snapdragon 8 Elite | ||||
| pu/qualcomm/sm8750/gemma-4-E2B-it.litertlm | Hexagon v79 NPU | 2.81 GB | Native NPU for Galaxy S25 series & flagship Android | |
| Qualcomm IoT | ||||
| pu/qualcomm/qcs8275/gemma-4-E2B-it.litertlm | Qualcomm IoT NPU | 3.07 GB | Robotics & edge compute tier | |
| Google Tensor | ||||
| pu/google/tensor_g5/gemma-4-E2B-it.litertlm | Tensor G5 TPU | 2.90 GB | Dedicated TPU for Pixel hardware | |
| Intel Core Ultra | ||||
| pu/intel/ptl/gemma-4-E2B-it.litertlm | Panther Lake NPU | 2.75 GB | On-device PC & laptop NPU | |
| Mobile GPU | gpu/gemma-4-E2B-it-gpu.litertlm | OpenCL / Metal / Vulkan | 1.87 GB | Universal GPU acceleration (Adreno / Mali / Immortalis) |
| CPU Baseline | cpu/gemma-4-E2B-it.litertlm | XNNPACK CPU | 2.41 GB | Universal fallback for all CPU architectures |
Capabilities
- Multimodal: Native visual understanding of photos, documents, and real-world scenes.
- Tool Calling: High-accuracy function calling and structured JSON output.
- English-Only Optimized: Streamlined for responsive assistant interactions.
- Downloads last month
- -