Hugging Face's logo

qualcomm
/
VIT

Image Classification
PyTorch
LiteRT
backbone
android
Model card Files Community

VIT: Optimized for Mobile Deployment

Imagenet classifier and general purpose backbone

VIT 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 VIT found here.

This repository provides scripts to run VIT 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: 86.6M
    • Model size (float): 330 MB
    • Model size (w8a16): 86.2 MB
    • Model size (w8a8): 83.2 MB
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
VIT float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 43.891 ms 0 - 315 MB NPU VIT.tflite
VIT float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 45.176 ms 0 - 324 MB NPU VIT.dlc
VIT float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 17.891 ms 0 - 321 MB NPU VIT.tflite
VIT float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 21.31 ms 0 - 316 MB NPU VIT.dlc
VIT float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 12.696 ms 0 - 28 MB NPU VIT.tflite
VIT float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 13.809 ms 0 - 32 MB NPU VIT.dlc
VIT float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 16.043 ms 0 - 315 MB NPU VIT.tflite
VIT float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 16.876 ms 1 - 324 MB NPU VIT.dlc
VIT float SA7255P ADP Qualcomm® SA7255P TFLITE 43.891 ms 0 - 315 MB NPU VIT.tflite
VIT float SA7255P ADP Qualcomm® SA7255P QNN_DLC 45.176 ms 0 - 324 MB NPU VIT.dlc
VIT float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 13.166 ms 0 - 24 MB NPU VIT.tflite
VIT float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 13.828 ms 0 - 31 MB NPU VIT.dlc
VIT float SA8295P ADP Qualcomm® SA8295P TFLITE 20.073 ms 0 - 307 MB NPU VIT.tflite
VIT float SA8295P ADP Qualcomm® SA8295P QNN_DLC 19.819 ms 1 - 327 MB NPU VIT.dlc
VIT float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 12.733 ms 0 - 26 MB NPU VIT.tflite
VIT float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 13.893 ms 0 - 30 MB NPU VIT.dlc
VIT float SA8775P ADP Qualcomm® SA8775P TFLITE 16.043 ms 0 - 315 MB NPU VIT.tflite
VIT float SA8775P ADP Qualcomm® SA8775P QNN_DLC 16.876 ms 1 - 324 MB NPU VIT.dlc
VIT float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 12.72 ms 0 - 13 MB NPU VIT.tflite
VIT float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 13.859 ms 0 - 28 MB NPU VIT.dlc
VIT float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 13.522 ms 1 - 22 MB NPU VIT.onnx
VIT float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 8.994 ms 0 - 319 MB NPU VIT.tflite
VIT float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 9.587 ms 38 - 370 MB NPU VIT.dlc
VIT float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 9.342 ms 1 - 336 MB NPU VIT.onnx
VIT float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 8.262 ms 0 - 319 MB NPU VIT.tflite
VIT float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 8.024 ms 1 - 314 MB NPU VIT.dlc
VIT float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 7.694 ms 1 - 320 MB NPU VIT.onnx
VIT float Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 16.174 ms 1116 - 1116 MB NPU VIT.dlc
VIT float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 14.917 ms 171 - 171 MB NPU VIT.onnx
VIT w8a16 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 64.987 ms 0 - 189 MB NPU VIT.dlc
VIT w8a16 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 51.328 ms 0 - 211 MB NPU VIT.dlc
VIT w8a16 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 25.883 ms 0 - 48 MB NPU VIT.dlc
VIT w8a16 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 22.734 ms 0 - 189 MB NPU VIT.dlc
VIT w8a16 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN_DLC 202.108 ms 0 - 1721 MB NPU VIT.dlc
VIT w8a16 SA7255P ADP Qualcomm® SA7255P QNN_DLC 64.987 ms 0 - 189 MB NPU VIT.dlc
VIT w8a16 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 26.026 ms 0 - 48 MB NPU VIT.dlc
VIT w8a16 SA8295P ADP Qualcomm® SA8295P QNN_DLC 37.113 ms 0 - 212 MB NPU VIT.dlc
VIT w8a16 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 25.998 ms 0 - 49 MB NPU VIT.dlc
VIT w8a16 SA8775P ADP Qualcomm® SA8775P QNN_DLC 22.734 ms 0 - 189 MB NPU VIT.dlc
VIT w8a16 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 26.042 ms 2 - 49 MB NPU VIT.dlc
VIT w8a16 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 157.686 ms 462 - 586 MB NPU VIT.onnx
VIT w8a16 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 19.309 ms 0 - 195 MB NPU VIT.dlc
VIT w8a16 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 125.213 ms 621 - 779 MB NPU VIT.onnx
VIT w8a16 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 15.916 ms 0 - 187 MB NPU VIT.dlc
VIT w8a16 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 106.961 ms 489 - 619 MB NPU VIT.onnx
VIT w8a16 Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 27.125 ms 335 - 335 MB NPU VIT.dlc
VIT w8a16 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 176.132 ms 923 - 923 MB NPU VIT.onnx
VIT w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 24.362 ms 0 - 49 MB NPU VIT.tflite
VIT w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 30.363 ms 0 - 164 MB NPU VIT.dlc
VIT w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 12.786 ms 0 - 57 MB NPU VIT.tflite
VIT w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 16.066 ms 0 - 229 MB NPU VIT.dlc
VIT w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 12.219 ms 0 - 93 MB NPU VIT.tflite
VIT w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 10.54 ms 0 - 27 MB NPU VIT.dlc
VIT w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 12.509 ms 0 - 49 MB NPU VIT.tflite
VIT w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 9.612 ms 0 - 163 MB NPU VIT.dlc
VIT w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) TFLITE 80.889 ms 2 - 43 MB NPU VIT.tflite
VIT w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN_DLC 78.712 ms 0 - 408 MB NPU VIT.dlc
VIT w8a8 SA7255P ADP Qualcomm® SA7255P TFLITE 24.362 ms 0 - 49 MB NPU VIT.tflite
VIT w8a8 SA7255P ADP Qualcomm® SA7255P QNN_DLC 30.363 ms 0 - 164 MB NPU VIT.dlc
VIT w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 12.222 ms 0 - 44 MB NPU VIT.tflite
VIT w8a8 SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 10.582 ms 0 - 25 MB NPU VIT.dlc
VIT w8a8 SA8295P ADP Qualcomm® SA8295P TFLITE 14.639 ms 0 - 49 MB NPU VIT.tflite
VIT w8a8 SA8295P ADP Qualcomm® SA8295P QNN_DLC 16.427 ms 0 - 169 MB NPU VIT.dlc
VIT w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 12.225 ms 0 - 104 MB NPU VIT.tflite
VIT w8a8 SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 10.564 ms 0 - 26 MB NPU VIT.dlc
VIT w8a8 SA8775P ADP Qualcomm® SA8775P TFLITE 12.509 ms 0 - 49 MB NPU VIT.tflite
VIT w8a8 SA8775P ADP Qualcomm® SA8775P QNN_DLC 9.612 ms 0 - 163 MB NPU VIT.dlc
VIT w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 12.199 ms 0 - 113 MB NPU VIT.tflite
VIT w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 10.556 ms 0 - 23 MB NPU VIT.dlc
VIT w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 156.848 ms 462 - 577 MB NPU VIT.onnx
VIT w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 8.675 ms 0 - 55 MB NPU VIT.tflite
VIT w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 7.0 ms 0 - 164 MB NPU VIT.dlc
VIT w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 127.23 ms 487 - 644 MB NPU VIT.onnx
VIT w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 7.175 ms 0 - 54 MB NPU VIT.tflite
VIT w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 6.083 ms 0 - 162 MB NPU VIT.dlc
VIT w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 103.704 ms 495 - 618 MB NPU VIT.onnx
VIT w8a8 Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 11.686 ms 412 - 412 MB NPU VIT.dlc
VIT w8a8 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 180.062 ms 924 - 924 MB NPU VIT.onnx

Installation

Install the package via pip:

pip install qai-hub-models

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.vit.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.vit.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.vit.export

Profiling Results
------------------------------------------------------------
VIT
Device                          : cs_8275 (ANDROID 14)                  
Runtime                         : TFLITE                                
Estimated inference time (ms)   : 43.9                                  
Estimated peak memory usage (MB): [0, 315]                              
Total # Ops                     : 1579                                  
Compute Unit(s)                 : npu (1579 ops) gpu (0 ops) cpu (0 ops)

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.vit import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S24")

# 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.vit.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.vit.demo -- --eval-mode on-device

Deploying compiled model to Android

The models can be deployed using multiple runtimes:

  • TensorFlow Lite (.tflite export): This tutorial provides a guide to deploy the .tflite model in an Android application.

  • QNN (.so export ): This sample app provides instructions on how to use the .so shared library in an Android application.

View on Qualcomm® AI Hub

Get more details on VIT's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

  • The license for the original implementation of VIT can be found here.
  • The license for the compiled assets for on-device deployment can be found here

References

Community

Downloads last month
104
Inference Providers NEW
Image Classification
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support