Hugging Face's logo

qualcomm
/
Sequencer2D

Image Classification
PyTorch
LiteRT
android
Model card Files
xet
Community

Sequencer2D: Optimized for Mobile Deployment

Imagenet classifier and general purpose backbone

sequencer2d is a vision transformer model that can classify images from the Imagenet dataset.

This model is an implementation of Sequencer2D found here.

This repository provides scripts to run Sequencer2D 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: sequencer2d_s
    • Input resolution: 224x224
    • Number of parameters: 27.6M
    • Model size (float): 106 MB
    • Model size (w8a16): 69.1 MB
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
Sequencer2D float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 116.219 ms 0 - 538 MB NPU Sequencer2D.tflite
Sequencer2D float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 89.881 ms 1 - 594 MB NPU Sequencer2D.dlc
Sequencer2D float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 57.652 ms 0 - 416 MB NPU Sequencer2D.tflite
Sequencer2D float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 68.099 ms 0 - 433 MB NPU Sequencer2D.dlc
Sequencer2D float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 59.971 ms 0 - 80 MB NPU Sequencer2D.tflite
Sequencer2D float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 42.334 ms 0 - 89 MB NPU Sequencer2D.dlc
Sequencer2D float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) ONNX 61.945 ms 0 - 50 MB NPU Sequencer2D.onnx.zip
Sequencer2D float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 61.928 ms 0 - 539 MB NPU Sequencer2D.tflite
Sequencer2D float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 44.001 ms 0 - 587 MB NPU Sequencer2D.dlc
Sequencer2D float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 60.11 ms 0 - 76 MB NPU Sequencer2D.tflite
Sequencer2D float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 41.816 ms 0 - 87 MB NPU Sequencer2D.dlc
Sequencer2D float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 63.251 ms 0 - 38 MB NPU Sequencer2D.onnx.zip
Sequencer2D float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 44.179 ms 0 - 544 MB NPU Sequencer2D.tflite
Sequencer2D float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 30.463 ms 1 - 1026 MB NPU Sequencer2D.dlc
Sequencer2D float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 53.55 ms 7 - 32 MB NPU Sequencer2D.onnx.zip
Sequencer2D float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 45.568 ms 0 - 547 MB NPU Sequencer2D.tflite
Sequencer2D float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 22.335 ms 1 - 582 MB NPU Sequencer2D.dlc
Sequencer2D float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 54.391 ms 8 - 32 MB NPU Sequencer2D.onnx.zip
Sequencer2D float Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 44.533 ms 469 - 469 MB NPU Sequencer2D.dlc
Sequencer2D float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 25.592 ms 3 - 3 MB NPU Sequencer2D.onnx.zip
Sequencer2D w8a8 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 82.161 ms 0 - 491 MB NPU Sequencer2D.tflite
Sequencer2D w8a8 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 40.075 ms 0 - 387 MB NPU Sequencer2D.tflite
Sequencer2D w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 43.482 ms 0 - 65 MB NPU Sequencer2D.tflite
Sequencer2D w8a8 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) ONNX 62.202 ms 127 - 255 MB NPU Sequencer2D.onnx.zip
Sequencer2D w8a8 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 45.82 ms 0 - 489 MB NPU Sequencer2D.tflite
Sequencer2D w8a8 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) ONNX 261.522 ms 19 - 40 MB CPU Sequencer2D.onnx.zip
Sequencer2D w8a8 RB5 (Proxy) Qualcomm® QCS8250 (Proxy) ONNX 271.335 ms 16 - 48 MB CPU Sequencer2D.onnx.zip
Sequencer2D w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 43.548 ms 0 - 63 MB NPU Sequencer2D.tflite
Sequencer2D w8a8 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 57.675 ms 123 - 254 MB NPU Sequencer2D.onnx.zip
Sequencer2D w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 32.596 ms 0 - 497 MB NPU Sequencer2D.tflite
Sequencer2D w8a8 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 55.021 ms 161 - 2609 MB NPU Sequencer2D.onnx.zip
Sequencer2D w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 31.918 ms 0 - 487 MB NPU Sequencer2D.tflite
Sequencer2D w8a8 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 36.684 ms 163 - 1165 MB NPU Sequencer2D.onnx.zip
Sequencer2D w8a8 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 52.765 ms 232 - 232 MB NPU Sequencer2D.onnx.zip

Installation

Install the package via pip:

pip install "qai-hub-models[sequencer2d]"

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.sequencer2d.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.sequencer2d.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.sequencer2d.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.sequencer2d 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.sequencer2d.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.sequencer2d.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 Sequencer2D's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

  • The license for the original implementation of Sequencer2D can be found here.
  • The license for the compiled assets for on-device deployment can be found here

References

Community

Downloads last month
14
Inference Providers NEW
Image Classification
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support