license: gemma
extra_gated_heading: Access Gemma on Hugging Face
extra_gated_prompt: >-
To access Gemma on Hugging Face, you’re required to review and agree to
Google’s usage license. To do this, please ensure you’re logged in to Hugging
Face and click below. Requests are processed immediately.
extra_gated_button_content: Acknowledge license
base_model:
- google/gemma-2-2b-it
litert-community/Gemma2-2B-IT
This model provides a few variants of google/gemma-2-2b-it that are ready for deployment on Android using the LiteRT (fka TFLite) stack and MediaPipe LLM Inference API.
Use the models
Colab
Disclaimer: The target deployment surface for the LiteRT models is Android/iOS/Web and the stack has been optimized for performance on these targets. Trying out the system in Colab is an easier way to familiarize yourself with the LiteRT stack, with the caveat that the performance (memory and latency) on Colab could be much worse than on a local device.
Android
To build the demo app from source, please follow the instructions from the GitHub repository.
Performance
Android
Note that all benchmark stats are from a Samsung S24 Ultra with 1280 KV cache size, 512 tokens prefill, 128 tokens decode.
| Backend | Prefill (tokens/sec) | Decode (tokens/sec) | Time-to-first-token (sec) | Memory (RSS in MB) | Model size (MB) | |
|---|---|---|---|---|---|---|
| dynamic_int8 | CPU | 146 |
11 |
3.9 |
4086 |
2703 |
| dynamic_int8 | GPU | 1052 |
15.6 |
7.6 |
5322 |
2702 |
- Model Size: measured by the size of the .tflite flatbuffer (serialization format for LiteRT models)
- Memory: indicator of peak RAM usage
- The inference on CPU is accelerated via the LiteRT XNNPACK delegate with 4 threads
- Benchmark on CPU is done assuming XNNPACK cache is enabled
- dynamic_int8: quantized model with int8 weights and float activations.