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

LLM Inference guide for iOS

The LLM Inference API lets you run large language models (LLMs) completely on-device for iOS 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 iOS apps.

The task supports the following variants of Gemma: Gemma-2 2B, Gemma 2B, and Gemma 7B. Gemma is 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 external models: Phi-2, Falcon-RW-1B and StableLM-3B.

In addition to the supported models, users can use Google's AI Edge Torch to export PyTorch models into multi-signature LiteRT (tflite) models, which are bundled with tokenizer parameters to create Task Bundles that are compatible with the LLM Inference API.

You can see this task in action with the MediaPipe Studio demo. For more information about the capabilities, models, and configuration options of this task, see the Overview.

Code example

The MediaPipe Tasks example code is a basic implementation of an LLM Inference API app for iOS. You can use the app as a starting point for your own iOS app, or refer to it when modifying an existing app. The LLM Inference API 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/ios/
```

After creating a local version of the example code, you can install the MediaPipe task library, open the project using Xcode and run the app. For instructions, see the Setup Guide for iOS.

Setup

This section describes key steps for setting up your development environment and code projects to use 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 iOS.

Dependencies

LLM Inference API uses the MediaPipeTasksGenai library, which must be installed using CocoaPods. The library is compatible with both Swift and Objective-C apps and does not require any additional language-specific setup.

For instructions to install CocoaPods on macOS, refer to the CocoaPods installation guide. For instructions on how to create a Podfile with the necessary pods for your app, refer to Using CocoaPods.

Add the MediaPipeTasksGenai pod in the Podfile using the following code:

target 'MyLlmInferenceApp' do
  use_frameworks!
  pod 'MediaPipeTasksGenAI'
  pod 'MediaPipeTasksGenAIC'
end

If your app includes unit test targets, refer to the Set Up Guide for iOS for additional information on setting up your Podfile.

Model

The MediaPipe LLM Inference API task requires a trained model that is compatible with this task. For more information on available trained models for LLM Inference API, see the task overview Models section.

Download a model

Download a model and add it to your project directory using Xcode. For instructions on how to add files to your Xcode project, refer to Managing files and folders in your Xcode project.

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.

In addition to the supported models, you can use Google's AI Edge Torch to export PyTorch models into multi-signature LiteRT (tflite) models. For more information, see Torch Generative converter for PyTorch models.

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

The LLM Inference API is compatible with two categories types of models, some of which require model conversion. Use the table to identify the required steps method for your model.

	Models	Conversion method	Compatible platforms	File type
Supported models	Gemma 2B, Gemma 7B, Gemma-2 2B, Phi-2, StableLM, Falcon	MediaPipe	Android, iOS, web	.bin
Other PyTorch models	All PyTorch LLM models	AI Edge Torch Generative library	Android, iOS	.task

We are hosting the converted .bin files for Gemma 2B, Gemma 7B, and Gemma-2 2B on Kaggle. These models can be directly deployed using our LLM Inference API. To learn how you can convert other models, see the Model Conversion section.

Create the task

You can create the LLM Inference API task by calling one of its initializers. The LlmInference(options:) initializer sets values for the configuration options.

If you don't need a LLM Inference API initialized with customized configuration options, you can use the LlmInference(modelPath:) initializer to create a LLM Inference API with the default options. For more information about configuration options, see Configuration Overview.

The following code demonstrates how to build and configure this task.

import MediaPipeTasksGenai

let modelPath = Bundle.main.path(forResource: "model",
                                      ofType: "bin")

let options = LlmInferenceOptions()
options.baseOptions.modelPath = modelPath
options.maxTokens = 1000
options.topk = 40
options.temperature = 0.8
options.randomSeed = 101

let llmInference = try LlmInference(options: options)

Configuration options

This task has the following configuration options for iOS apps:

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

Prepare data

LLM Inference API works with text data. The task handles the data input preprocessing, including tokenization and tensor preprocessing.

All preprocessing is handled within the generateResponse(inputText:) function. There is no need for additional preprocessing of the input text beforehand.

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

Run the task

To run the LLM Inference API, use the generateResponse(inputText:) method. The LLM Inference API returns the possible categories for the input text.

let result = try LlmInference.generateResponse(inputText: inputPrompt)

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

let resultStream =  LlmInference.generateResponseAsync(inputText: inputPrompt)

do {
  for try await partialResult in resultStream {
    print("\(partialResult)")
  }
  print("Done")
}
catch {
  print("Response error: '\(error)")
}

Handle and display results

The LLM Inference API returns 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 all Gemma variants 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 or Phi-2. Gemma-2 2B, 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
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.

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

import MediaPipeTasksGenai

let modelPath = Bundle.main.path(forResource: "model",
                                      ofType: "bin")
let loraPath= Bundle.main.path(forResource: "lora_model",
                                      ofType: "bin")
let options = LlmInferenceOptions()
options.modelPath = modelPath
options.maxTokens = 1000
options.topk = 40
options.temperature = 0.8
options.randomSeed = 101
options.loraPath = loraPath

let llmInference = try LlmInference(options: options)

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