Hugging Face's logo

qualcomm
/
ConvNext-Base

Image Classification
PyTorch
LiteRT
ONNX
android
Model card Files Community

ConvNext-Base: Optimized for Mobile Deployment

Imagenet classifier and general purpose backbone

ConvNextBase 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 ConvNext-Base found here.

This repository provides scripts to run ConvNext-Base 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: 88.6M
    • Model size (float): 338 MB
    • Model size (w8a16): 88.7 MB
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
ConvNext-Base float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 41.584 ms 0 - 283 MB NPU ConvNext-Base.tflite
ConvNext-Base float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 42.312 ms 1 - 281 MB NPU ConvNext-Base.dlc
ConvNext-Base float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 18.282 ms 0 - 295 MB NPU ConvNext-Base.tflite
ConvNext-Base float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 20.47 ms 0 - 300 MB NPU ConvNext-Base.dlc
ConvNext-Base float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 7.451 ms 0 - 28 MB NPU ConvNext-Base.tflite
ConvNext-Base float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 8.253 ms 0 - 28 MB NPU ConvNext-Base.dlc
ConvNext-Base float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 11.422 ms 0 - 284 MB NPU ConvNext-Base.tflite
ConvNext-Base float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 12.128 ms 1 - 281 MB NPU ConvNext-Base.dlc
ConvNext-Base float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 7.483 ms 0 - 34 MB NPU ConvNext-Base.tflite
ConvNext-Base float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 8.283 ms 0 - 25 MB NPU ConvNext-Base.dlc
ConvNext-Base float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 7.666 ms 0 - 399 MB NPU ConvNext-Base.onnx
ConvNext-Base float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 5.561 ms 0 - 295 MB NPU ConvNext-Base.tflite
ConvNext-Base float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 6.003 ms 1 - 290 MB NPU ConvNext-Base.dlc
ConvNext-Base float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 5.511 ms 1 - 297 MB NPU ConvNext-Base.onnx
ConvNext-Base float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 4.839 ms 0 - 282 MB NPU ConvNext-Base.tflite
ConvNext-Base float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 5.395 ms 1 - 286 MB NPU ConvNext-Base.dlc
ConvNext-Base float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 4.359 ms 1 - 284 MB NPU ConvNext-Base.onnx
ConvNext-Base float Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 8.87 ms 1390 - 1390 MB NPU ConvNext-Base.dlc
ConvNext-Base float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 7.876 ms 176 - 176 MB NPU ConvNext-Base.onnx
ConvNext-Base w8a16 QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 14.431 ms 0 - 130 MB NPU ConvNext-Base.dlc
ConvNext-Base w8a16 QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 9.619 ms 0 - 146 MB NPU ConvNext-Base.dlc
ConvNext-Base w8a16 QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 5.96 ms 0 - 33 MB NPU ConvNext-Base.dlc
ConvNext-Base w8a16 QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 6.226 ms 0 - 130 MB NPU ConvNext-Base.dlc
ConvNext-Base w8a16 RB3 Gen 2 (Proxy) Qualcomm® QCS6490 (Proxy) QNN_DLC 39.227 ms 0 - 206 MB NPU ConvNext-Base.dlc
ConvNext-Base w8a16 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 5.951 ms 0 - 34 MB NPU ConvNext-Base.dlc
ConvNext-Base w8a16 Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 239.386 ms 521 - 915 MB NPU ConvNext-Base.onnx
ConvNext-Base w8a16 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 4.158 ms 0 - 138 MB NPU ConvNext-Base.dlc
ConvNext-Base w8a16 Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 201.977 ms 683 - 1468 MB NPU ConvNext-Base.onnx
ConvNext-Base w8a16 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 3.842 ms 0 - 130 MB NPU ConvNext-Base.dlc
ConvNext-Base w8a16 Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 157.297 ms 687 - 1377 MB NPU ConvNext-Base.onnx
ConvNext-Base w8a16 Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 6.626 ms 521 - 521 MB NPU ConvNext-Base.dlc
ConvNext-Base w8a16 Snapdragon X Elite CRD Snapdragon® X Elite ONNX 213.519 ms 918 - 918 MB NPU ConvNext-Base.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.convnext_base.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.convnext_base.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.convnext_base.export

Profiling Results
------------------------------------------------------------
ConvNext-Base
Device                          : cs_8275 (ANDROID 14)                 
Runtime                         : TFLITE                               
Estimated inference time (ms)   : 41.6                                 
Estimated peak memory usage (MB): [0, 283]                             
Total # Ops                     : 598                                  
Compute Unit(s)                 : npu (598 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.convnext_base 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.convnext_base.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.convnext_base.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 ConvNext-Base's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

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

References

Community

Downloads last month
41
Inference Providers NEW
Image Classification
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support