Hugging Face's logo

qualcomm
/
Gemma-4-E2B-it

Text Generation
PyTorch
llm
vlm
generative_ai
android
Model card Files
xet
 Community

Gemma-4-E2B-it: Optimized for Qualcomm Devices

Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on small models) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages.

This is based on the implementation of Gemma-4-E2B-it found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.

Deploying Gemma-4-E2B-it on-device

Follow the GenieX quickstart to install GenieX and deploy the model on a target device.

Getting Started

There are two ways to deploy this model on your device:

Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

Runtime Precision Chipset SDK Versions Download
GENIEX_LLAMACPP q4_0 Universal Download

For more device-specific assets and performance metrics, visit Gemma-4-E2B-it on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:

  • Custom weights (e.g., fine-tuned checkpoints)
  • Custom input shapes
  • Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for Gemma-4-E2B-it on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.text_generation

Model Stats:

  • Model architecture: Mixture-of-Experts (MoE) Transformer with Per-Layer Expert Selection and selective routing.
  • Supported languages: Multilingual (trained on 140+ languages)
  • TTFT: Time To First Token is the time it takes to generate the first response token. This is expressed as a range because it varies based on the length of the prompt.
  • Response Rate: Rate of response generation after the first response token.

Performance Summary

Model Runtime Precision Chipset Context Length Response Rate (tokens per second) Time To First Token (range, seconds)
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Gen 5 Mobile 512 35.450218 0.5510827500000001 - 2.2043310000000003
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Gen 5 Mobile 512 35.028285 0.583708 - 2.334832
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Gen 5 Mobile 512 17.606799 0.09366424999999999 - 0.37465699999999996
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Gen 5 Mobile 4096 24.948986 1.10109603125 - 35.235073
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Gen 5 Mobile 4096 22.172419 1.41503134375 - 45.281003
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Gen 5 Mobile 4096 18.185303 0.129888375 - 4.156428
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Mobile 512 30.421424 0.6692315 - 2.676926
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Mobile 512 30.226744 0.65932325 - 2.637293
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Mobile 512 26.660068 0.12724075 - 0.508963
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Mobile 4096 22.975215 1.23248246875 - 39.439439
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Mobile 4096 19.17611 1.52781575 - 48.890104
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® 8 Elite Mobile 4096 20.615236 0.18424428125 - 5.895817
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X2 Elite 512 37.335026 0.171274 - 0.685096
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X2 Elite 512 36.671678 0.17170575 - 0.686823
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X2 Elite 512 35.068599 0.090643 - 0.362572
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X2 Elite 4096 38.550303 0.26170965625000003 - 8.374709000000001
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X2 Elite 4096 39.138586 0.26236340625000004 - 8.395629000000001
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X2 Elite 4096 25.860731 0.11250709375 - 3.600227
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X Elite 512 27.336969 0.35564425 - 1.422577
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X Elite 512 19.991204 0.39397775 - 1.575911
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X Elite 512 18.725014 0.22572075 - 0.902883
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X Elite 4096 14.995314 0.6033723125 - 19.307914
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X Elite 4096 20.251114 0.5856200625 - 18.739842
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Snapdragon® X Elite 4096 16.824594 0.2757331875 - 8.823462
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Qualcomm® QCS9075 512 23.348938 0.90453275 - 3.618131
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Qualcomm® QCS9075 512 23.427159 0.90761 - 3.63044
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Qualcomm® QCS9075 512 13.47721 0.2507545 - 1.003018
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Qualcomm® QCS9075 4096 18.616736 1.2610428125000002 - 40.353370000000005
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Qualcomm® QCS9075 4096 18.688134 1.25764 - 40.24448
Gemma-4-E2B-it GENIEX_LLAMACPP q4_0 Qualcomm® QCS9075 4096 12.228339 0.3025998125 - 9.683194

License

  • The license for the original implementation of Gemma-4-E2B-it can be found here.

References

Community

Usage and Limitations

This model may not be used for or in connection with any of the following applications:

  • Accessing essential private and public services and benefits;
  • Administration of justice and democratic processes;
  • Assessing or recognizing the emotional state of a person;
  • Biometric and biometrics-based systems, including categorization of persons based on sensitive characteristics;
  • Education and vocational training;
  • Employment and workers management;
  • Exploitation of the vulnerabilities of persons resulting in harmful behavior;
  • General purpose social scoring;
  • Law enforcement;
  • Management and operation of critical infrastructure;
  • Migration, asylum and border control management;
  • Predictive policing;
  • Real-time remote biometric identification in public spaces;
  • Recommender systems of social media platforms;
  • Scraping of facial images (from the internet or otherwise); and/or
  • Subliminal manipulation
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
Text Generation
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support