Introducing LiteRT: Google's high-performance runtime for on-device AI, formerly known as TensorFlow Lite. Learn more

LLM Inference guide for Android

The LLM Inference API lets you run large language models (LLMs) completely on-device for Android applications, which you can use to perform a wide range of tasks, such as generating text, retrieving information in natural language form, and summarizing documents. The task provides built-in support for multiple text-to-text large language models, so you can apply the latest on-device generative AI models to your Android apps.

The task supports Gemma-2 2B, the latest in a family of lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models. It also supports the following additional models: Gemma, Phi-2, Falcon-RW-1B and StableLM-3B, along with all models exported through AI Edge.

For more information about the capabilities, models, and configuration options of this task, see the Overview.

Code example

This guide refers to an example of a basic text generation app for Android. You can use the app as a starting point for your own Android app, or refer to it when modifying an existing app. The example code is hosted on GitHub.

Download the code

The following instructions show you how to create a local copy of the example code using the git command line tool.

To download the example code:

Clone the git repository using the following command:

git clone https://github.com/google-ai-edge/mediapipe-samples

Optionally, configure your git instance to use sparse checkout, so you have only the files for the LLM Inference API example app:
```
cd mediapipe
git sparse-checkout init --cone
git sparse-checkout set examples/llm_inference/android
```

After creating a local version of the example code, you can import the project into Android Studio and run the app. For instructions, see the Setup Guide for Android.

Setup

This section describes key steps for setting up your development environment and code projects specifically to use the LLM Inference API. For general information on setting up your development environment for using MediaPipe tasks, including platform version requirements, see the Setup guide for Android.

Dependencies

The LLM Inference API uses the com.google.mediapipe:tasks-genai library. Add this dependency to the build.gradle file of your Android app:

dependencies {
    implementation 'com.google.mediapipe:tasks-genai:0.10.14'
}

Model

The MediaPipe LLM Inference API requires a trained text-to-text language model that is compatible with this task. After downloading a model, install the required dependencies and push the model to the Android device. If you are using a model other than Gemma, you will have to convert the model to a format compatible with MediaPipe.

For more information on available trained models for LLM Inference API, see the task overview Models section.

Download a model

Before initializing the LLM Inference API, download one of the supported models and store the file within your project directory:

Gemma-2 2B: The latest version of Gemma family of models. Part of a family of lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models.
Gemma 2B: Part of a family of lightweight, state-of-the-art open models built from the same research and technology used to create the Gemini models. Well-suited for a variety of text generation tasks, including question answering, summarization, and reasoning.
Phi-2: 2.7 billion parameter Transformer model, best suited for the Question-Answer, chat, and code format.
Falcon-RW-1B: 1 billion parameter causal decoder-only model trained on 350B tokens of RefinedWeb.
StableLM-3B: 3 billion parameter decoder-only language model pre-trained on 1 trillion tokens of diverse English and code datasets.

We recommend using Gemma-2 2B, which is available on Kaggle Models. For more information on the other available models, see the task overview Models section.

Convert model to MediaPipe format

Native model conversion

If you are using an external LLM (Phi-2, Falcon, or StableLM) or a non-Kaggle version of Gemma, use our conversion scripts to format the model to be compatible with MediaPipe.

The model conversion process requires the MediaPipe PyPI package. The conversion script is available in all MediaPipe packages after 0.10.11.

Install and import the dependencies with the following:

$ python3 -m pip install mediapipe

Use the genai.converter library to convert the model:

import mediapipe as mp
from mediapipe.tasks.python.genai import converter

config = converter.ConversionConfig(
  input_ckpt=INPUT_CKPT,
  ckpt_format=CKPT_FORMAT,
  model_type=MODEL_TYPE,
  backend=BACKEND,
  output_dir=OUTPUT_DIR,
  combine_file_only=False,
  vocab_model_file=VOCAB_MODEL_FILE,
  output_tflite_file=OUTPUT_TFLITE_FILE,
)

converter.convert_checkpoint(config)

To convert the LoRA model, the ConversionConfig should specify the base model options as well as additional LoRA options. Notice that since the API only supports LoRA inference with GPU, the backend must be set to 'gpu'.

import mediapipe as mp
from mediapipe.tasks.python.genai import converter

config = converter.ConversionConfig(
  # Other params related to base model
  ...
  # Must use gpu backend for LoRA conversion
  backend='gpu',
  # LoRA related params
  lora_ckpt=LORA_CKPT,
  lora_rank=LORA_RANK,
  lora_output_tflite_file=LORA_OUTPUT_TFLITE_FILE,
)

converter.convert_checkpoint(config)

The converter will output two TFLite flatbuffer files, one for the base model and the other for the LoRA model.

Parameter	Description	Accepted Values
`input_ckpt`	The path to the `model.safetensors` or `pytorch.bin` file. Note that sometimes the model safetensors format are sharded into multiple files, e.g. `model-00001-of-00003.safetensors`, `model-00001-of-00003.safetensors`. You can specify a file pattern, like `model*.safetensors`.	PATH
`ckpt_format`	The model file format.	{"safetensors", "pytorch"}
`model_type`	The LLM being converted.	{"PHI_2", "FALCON_RW_1B", "STABLELM_4E1T_3B", "GEMMA_2B", "GEMMA_7B", "GEMMA-2_2B"}
`backend`	The processor (delegate) used to run the model.	{"cpu", "gpu"}
`output_dir`	The path to the output directory that hosts the per-layer weight files.	PATH
`output_tflite_file`	The path to the output file. For example, "model_cpu.bin" or "model_gpu.bin". This file is only compatible with the LLM Inference API, and cannot be used as a general `tflite` file.	PATH
`vocab_model_file`	The path to the directory that stores the `tokenizer.json` and `tokenizer_config.json` files. For Gemma, point to the single `tokenizer.model` file.	PATH
`lora_ckpt`	The path to the LoRA ckpt of safetensors file that stores the LoRA adapter weight.	PATH
`lora_rank`	An integer representing the rank of LoRA ckpt. Required in order to convert the lora weights. If not provided, then the converter assumes there are no LoRA weights. Note: Only the GPU backend supports LoRA.	Integer
`lora_output_tflite_file`	Output tflite filename for the LoRA weights.	PATH

AI Edge model conversion

If you are using an LLM mapped to a TFLite model through AI Edge, use our bundling script to create a Task Bundle. The bundling process packs the mapped model with additional metadata (e.g., Tokenizer Parameters) needed to run end-to-end inference.

The model bundling process requires the MediaPipe PyPI package. The conversion script is available in all MediaPipe packages after 0.10.14.

Install and import the dependencies with the following:

$ python3 -m pip install mediapipe

Use the genai.bundler library to bundle the model:

import mediapipe as mp
from mediapipe.tasks.python.genai import bundler

config = bundler.BundleConfig(
    tflite_model=TFLITE_MODEL,
    tokenizer_model=TOKENIZER_MODEL,
    start_token=START_TOKEN,
    stop_tokens=STOP_TOKENS,
    output_filename=OUTPUT_FILENAME,
    enable_bytes_to_unicode_mapping=ENABLE_BYTES_TO_UNICODE_MAPPING,
)
bundler.create_bundle(config)

Parameter	Description	Accepted Values
`tflite_model`	The path to the AI Edge exported TFLite model.	PATH
`tokenizer_model`	The path to the SentencePiece tokenizer model.	PATH
`start_token`	Model specific start token. The start token must be present in the provided tokenizer model.	STRING
`stop_tokens`	Model specific stop tokens. The stop tokens must be present in the provided tokenizer model.	LIST[STRING]
`output_filename`	The name of the output task bundle file.	PATH

Push model to the device

Push the content of the output_path folder to the Android device.

$ adb shell rm -r /data/local/tmp/llm/ # Remove any previously loaded models
$ adb shell mkdir -p /data/local/tmp/llm/
$ adb push output_path /data/local/tmp/llm/model_version.bin

Create the task

The MediaPipe LLM Inference API uses the createFromOptions() function to set up the task. The createFromOptions() function accepts values for the configuration options. For more information on configuration options, see Configuration options.

The following code initializes the task using basic configuration options:

// Set the configuration options for the LLM Inference task
val options = LlmInferenceOptions.builder()
        .setModelPATH('/data/local/.../')
        .setMaxTokens(1000)
        .setTopK(40)
        .setTemperature(0.8)
        .setRandomSeed(101)
        .build()

// Create an instance of the LLM Inference task
llmInference = LlmInference.createFromOptions(context, options)

Configuration options

Use the following configuration options to set up an Android app:

Option Name	Description	Value Range	Default Value
`modelPath`	The path to where the model is stored within the project directory.	PATH	N/A
`maxTokens`	The maximum number of tokens (input tokens + output tokens) the model handles.	Integer	512
`topK`	The number of tokens the model considers at each step of generation. Limits predictions to the top k most-probable tokens.	Integer	40
`temperature`	The amount of randomness introduced during generation. A higher temperature results in more creativity in the generated text, while a lower temperature produces more predictable generation.	Float	0.8
`randomSeed`	The random seed used during text generation.	Integer	0
`loraPath`	The absolute path to the LoRA model locally on the device. Note: this is only compatible with GPU models.	PATH	N/A
`resultListener`	Sets the result listener to receive the results asynchronously. Only applicable when using the async generation method.	N/A	N/A
`errorListener`	Sets an optional error listener.	N/A	N/A

Prepare data

The LLM Inference API accepts the following inputs:

prompt (string): A question or prompt.

val inputPrompt = "Compose an email to remind Brett of lunch plans at noon on Saturday."

Run the task

Use the generateResponse() method to generate a text response to the input text provided in the previous section (inputPrompt). This produces a single generated response.

val result = llmInference.generateResponse(inputPrompt)
logger.atInfo().log("result: $result")

To stream the response, use the generateResponseAsync() method.

val options = LlmInference.LlmInferenceOptions.builder()
  ...
  .setResultListener { partialResult, done ->
    logger.atInfo().log("partial result: $partialResult")
  }
  .build()

llmInference.generateResponseAsync(inputPrompt)

Handle and display results

The LLM Inference API returns a LlmInferenceResult, which includes the generated response text.

Here's a draft you can use:

Subject: Lunch on Saturday Reminder

Hi Brett,

Just a quick reminder about our lunch plans this Saturday at noon.
Let me know if that still works for you.

Looking forward to it!

Best,
[Your Name]

LoRA model customization

Mediapipe LLM inference API can be configured to support Low-Rank Adaptation (LoRA) for large language models. Utilizing fine-tuned LoRA models, developers can customize the behavior of LLMs through a cost-effective training process.

LoRA support of the LLM Inference API works for Gemma-2B and Phi-2 models for the GPU backend, with LoRA weights applicable to attention layers only. This initial implementation serves as an experimental API for future developments with plans to support more models and various types of layers in the coming updates.

Prepare LoRA models

Follow the instructions on HuggingFace to train a fine tuned LoRA model on your own dataset with supported model types, Gemma-2B or Phi-2. Gemma-2B and Phi-2 models are both available on HuggingFace in the safetensors format. Since LLM Inference API only supports LoRA on attention layers, only specify attention layers while creating the LoraConfig as following:

# For Gemma-2B
from peft import LoraConfig
config = LoraConfig(
    r=LORA_RANK,
    target_modules=["q_proj", "v_proj", "k_proj", "o_proj"],
)

# For Phi-2
config = LoraConfig(
    r=LORA_RANK,
    target_modules=["q_proj", "v_proj", "k_proj", "dense"],
)

For testing, there are publicly accessible fine-tuned LoRA models which fit LLM Inference API available on HuggingFace. For example, monsterapi/gemma-2b-lora-maths-orca-200k for Gemma-2B and lole25/phi-2-sft-ultrachat-lora for Phi-2.

After training on the prepared dataset and saving the model, you obtain an adapter_model.safetensors file containing the fine-tuned LoRA model weights. The safetensors file is the LoRA checkpoint used in the model conversion.

As the next step, you need convert the model weights into a TensorFlow Lite Flatbuffer using the MediaPipe Python Package. The ConversionConfig should specify the base model options as well as additional LoRA options. Notice that since the API only supports LoRA inference with GPU, the backend must be set to 'gpu'.

import mediapipe as mp
from mediapipe.tasks.python.genai import converter

config = converter.ConversionConfig(
  # Other params related to base model
  ...
  # Must use gpu backend for LoRA conversion
  backend='gpu',
  # LoRA related params
  lora_ckpt=LORA_CKPT,
  lora_rank=LORA_RANK,
  lora_output_tflite_file=LORA_OUTPUT_TFLITE_FILE,
)

converter.convert_checkpoint(config)

The converter will output two TFLite flatbuffer files, one for the base model and the other for the LoRA model.

LoRA model inference

The Web, Android and iOS LLM Inference API are updated to support LoRA model inference. Web supports dynamic LoRA, which can switch different LoRA models during runtime. Android and iOS support static LoRA, which uses the same LoRA weights during the lifetime of the task.

Android supports static LoRA during initialization. To load a LoRA model, users specify the LoRA model path as well as the base LLM.

// Set the configuration options for the LLM Inference task
val options = LlmInferenceOptions.builder()
        .setModelPath('<path to base model>')
        .setMaxTokens(1000)
        .setTopK(40)
        .setTemperature(0.8)
        .setRandomSeed(101)
        .setLoraPath('<path to LoRA model>')
        .build()

// Create an instance of the LLM Inference task
llmInference = LlmInference.createFromOptions(context, options)

To run LLM inference with LoRA, use the same generateResponse() or generateResponseAsync() methods as the base model.