EfficientViT-l2-cls: Optimized for Mobile Deployment
Imagenet classifier and general purpose backbone
EfficientViT is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.
This model is an implementation of EfficientViT-l2-cls found here.
This repository provides scripts to run EfficientViT-l2-cls on Qualcomm® devices. More details on model performance across various devices, can be found here.
Model Details
- Model Type: Model_use_case.image_classification
- Model Stats:
- Model checkpoint: Imagenet
- Input resolution: 224x224
- Number of parameters: 63.7M
- Model size (float): 243 MB
| Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |
|---|---|---|---|---|---|---|---|---|
| EfficientViT-l2-cls | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 24.006 ms | 0 - 206 MB | NPU | EfficientViT-l2-cls.tflite |
| EfficientViT-l2-cls | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 24.602 ms | 1 - 90 MB | NPU | EfficientViT-l2-cls.dlc |
| EfficientViT-l2-cls | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 13.744 ms | 0 - 212 MB | NPU | EfficientViT-l2-cls.tflite |
| EfficientViT-l2-cls | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 15.039 ms | 0 - 99 MB | NPU | EfficientViT-l2-cls.dlc |
| EfficientViT-l2-cls | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 7.215 ms | 0 - 216 MB | NPU | EfficientViT-l2-cls.tflite |
| EfficientViT-l2-cls | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 7.476 ms | 0 - 26 MB | NPU | EfficientViT-l2-cls.dlc |
| EfficientViT-l2-cls | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 7.546 ms | 0 - 179 MB | NPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 9.126 ms | 2 - 208 MB | NPU | EfficientViT-l2-cls.tflite |
| EfficientViT-l2-cls | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 9.732 ms | 1 - 88 MB | NPU | EfficientViT-l2-cls.dlc |
| EfficientViT-l2-cls | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 5.178 ms | 0 - 234 MB | NPU | EfficientViT-l2-cls.tflite |
| EfficientViT-l2-cls | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 5.307 ms | 1 - 117 MB | NPU | EfficientViT-l2-cls.dlc |
| EfficientViT-l2-cls | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 5.477 ms | 0 - 119 MB | NPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 4.15 ms | 0 - 209 MB | NPU | EfficientViT-l2-cls.tflite |
| EfficientViT-l2-cls | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 4.087 ms | 1 - 177 MB | NPU | EfficientViT-l2-cls.dlc |
| EfficientViT-l2-cls | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 4.116 ms | 0 - 171 MB | NPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 3.53 ms | 0 - 206 MB | NPU | EfficientViT-l2-cls.tflite |
| EfficientViT-l2-cls | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 3.27 ms | 1 - 97 MB | NPU | EfficientViT-l2-cls.dlc |
| EfficientViT-l2-cls | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 3.466 ms | 1 - 93 MB | NPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 8.116 ms | 642 - 642 MB | NPU | EfficientViT-l2-cls.dlc |
| EfficientViT-l2-cls | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 7.901 ms | 131 - 131 MB | NPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 25.233 ms | 0 - 170 MB | NPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | w8a16 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | ONNX | 252.44 ms | 82 - 100 MB | CPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | w8a16 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 236.97 ms | 57 - 112 MB | CPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 20.101 ms | 41 - 210 MB | NPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | w8a16 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 14.773 ms | 40 - 188 MB | NPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | w8a16 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 12.654 ms | 40 - 196 MB | NPU | EfficientViT-l2-cls.onnx.zip |
| EfficientViT-l2-cls | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 24.19 ms | 133 - 133 MB | NPU | EfficientViT-l2-cls.onnx.zip |
Installation
Install the package via pip:
pip install "qai-hub-models[efficientvit-l2-cls]"
Configure Qualcomm® AI Hub to run this model on a cloud-hosted device
Sign-in to Qualcomm® AI Hub with your
Qualcomm® ID. Once signed in navigate to Account -> Settings -> API Token.
With this API token, you can configure your client to run models on the cloud hosted devices.
qai-hub configure --api_token API_TOKEN
Navigate to docs for more information.
Demo off target
The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input.
python -m qai_hub_models.models.efficientvit_l2_cls.demo
The above demo runs a reference implementation of pre-processing, model inference, and post processing.
NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).
%run -m qai_hub_models.models.efficientvit_l2_cls.demo
Run model on a cloud-hosted device
In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following:
- Performance check on-device on a cloud-hosted device
- Downloads compiled assets that can be deployed on-device for Android.
- Accuracy check between PyTorch and on-device outputs.
python -m qai_hub_models.models.efficientvit_l2_cls.export
How does this work?
This export script leverages Qualcomm® AI Hub to optimize, validate, and deploy this model on-device. Lets go through each step below in detail:
Step 1: Compile model for on-device deployment
To compile a PyTorch model for on-device deployment, we first trace the model
in memory using the jit.trace and then call the submit_compile_job API.
import torch
import qai_hub as hub
from qai_hub_models.models.efficientvit_l2_cls import Model
# Load the model
torch_model = Model.from_pretrained()
# Device
device = hub.Device("Samsung Galaxy S25")
# Trace model
input_shape = torch_model.get_input_spec()
sample_inputs = torch_model.sample_inputs()
pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
# Compile model on a specific device
compile_job = hub.submit_compile_job(
model=pt_model,
device=device,
input_specs=torch_model.get_input_spec(),
)
# Get target model to run on-device
target_model = compile_job.get_target_model()
Step 2: Performance profiling on cloud-hosted device
After compiling models from step 1. Models can be profiled model on-device using the
target_model. Note that this scripts runs the model on a device automatically
provisioned in the cloud. Once the job is submitted, you can navigate to a
provided job URL to view a variety of on-device performance metrics.
profile_job = hub.submit_profile_job(
model=target_model,
device=device,
)
Step 3: Verify on-device accuracy
To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device.
input_data = torch_model.sample_inputs()
inference_job = hub.submit_inference_job(
model=target_model,
device=device,
inputs=input_data,
)
on_device_output = inference_job.download_output_data()
With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output.
Note: This on-device profiling and inference requires access to Qualcomm® AI Hub. Sign up for access.
Run demo on a cloud-hosted device
You can also run the demo on-device.
python -m qai_hub_models.models.efficientvit_l2_cls.demo --eval-mode on-device
NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).
%run -m qai_hub_models.models.efficientvit_l2_cls.demo -- --eval-mode on-device
Deploying compiled model to Android
The models can be deployed using multiple runtimes:
TensorFlow Lite (
.tfliteexport): This tutorial provides a guide to deploy the .tflite model in an Android application.QNN (
.soexport ): This sample app provides instructions on how to use the.soshared library in an Android application.
View on Qualcomm® AI Hub
Get more details on EfficientViT-l2-cls's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of EfficientViT-l2-cls can be found here.
- The license for the compiled assets for on-device deployment can be found here
References
- EfficientViT: Multi-Scale Linear Attention for High-Resolution Dense Prediction
- Source Model Implementation
Community
- Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
- For questions or feedback please reach out to us.
- Downloads last month
- 109