--- library_name: pytorch license: other tags: - android pipeline_tag: image-classification --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientformer/web-assets/model_demo.png) # EfficientFormer: Optimized for Mobile Deployment ## Imagenet classifier and general purpose backbone EfficientFormer is a vision transformer model that can classify images from the Imagenet dataset. This model is an implementation of EfficientFormer found [here](https://github.com/snap-research/EfficientFormer). This repository provides scripts to run EfficientFormer on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/efficientformer). ### Model Details - **Model Type:** Model_use_case.image_classification - **Model Stats:** - Model checkpoint: efficientformer_l1_300d - Input resolution: 224x224 - Number of parameters: 12.3M - Model size (float): 46.9 MB - Model size (w8a16): 12.2 MB | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | EfficientFormer | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 4.918 ms | 0 - 48 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.tflite) | | EfficientFormer | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 5.024 ms | 1 - 34 MB | NPU | [EfficientFormer.dlc](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.dlc) | | EfficientFormer | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 3.826 ms | 0 - 56 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.tflite) | | EfficientFormer | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 5.662 ms | 0 - 43 MB | NPU | [EfficientFormer.dlc](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.dlc) | | EfficientFormer | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 1.528 ms | 0 - 157 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.tflite) | | EfficientFormer | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 1.63 ms | 0 - 11 MB | NPU | [EfficientFormer.dlc](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.dlc) | | EfficientFormer | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 5.185 ms | 1 - 13 MB | NPU | [EfficientFormer.onnx.zip](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.onnx.zip) | | EfficientFormer | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 2.096 ms | 0 - 49 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.tflite) | | EfficientFormer | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 2.304 ms | 1 - 33 MB | NPU | [EfficientFormer.dlc](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.dlc) | | EfficientFormer | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 1.072 ms | 0 - 61 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.tflite) | | EfficientFormer | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 1.113 ms | 1 - 42 MB | NPU | [EfficientFormer.dlc](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.dlc) | | EfficientFormer | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 3.603 ms | 0 - 46 MB | NPU | [EfficientFormer.onnx.zip](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.onnx.zip) | | EfficientFormer | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.814 ms | 0 - 54 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.tflite) | | EfficientFormer | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 0.832 ms | 1 - 40 MB | NPU | [EfficientFormer.dlc](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.dlc) | | EfficientFormer | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 2.616 ms | 0 - 44 MB | NPU | [EfficientFormer.onnx.zip](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.onnx.zip) | | EfficientFormer | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 1.923 ms | 103 - 103 MB | NPU | [EfficientFormer.dlc](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.dlc) | | EfficientFormer | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 5.397 ms | 25 - 25 MB | NPU | [EfficientFormer.onnx.zip](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer.onnx.zip) | | EfficientFormer | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 1.666 ms | 0 - 27 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.tflite) | | EfficientFormer | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 0.843 ms | 0 - 51 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.tflite) | | EfficientFormer | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 0.741 ms | 0 - 59 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.tflite) | | EfficientFormer | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 7.688 ms | 21 - 48 MB | NPU | [EfficientFormer.onnx.zip](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.onnx.zip) | | EfficientFormer | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 0.992 ms | 0 - 27 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.tflite) | | EfficientFormer | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 4.166 ms | 0 - 37 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.tflite) | | EfficientFormer | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | ONNX | 35.184 ms | 12 - 27 MB | CPU | [EfficientFormer.onnx.zip](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.onnx.zip) | | EfficientFormer | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 31.712 ms | 0 - 75 MB | GPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.tflite) | | EfficientFormer | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 30.835 ms | 13 - 23 MB | CPU | [EfficientFormer.onnx.zip](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.onnx.zip) | | EfficientFormer | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.52 ms | 0 - 42 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.tflite) | | EfficientFormer | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 5.36 ms | 25 - 86 MB | NPU | [EfficientFormer.onnx.zip](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.onnx.zip) | | EfficientFormer | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 0.41 ms | 0 - 30 MB | NPU | [EfficientFormer.tflite](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.tflite) | | EfficientFormer | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 4.366 ms | 0 - 44 MB | NPU | [EfficientFormer.onnx.zip](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.onnx.zip) | | EfficientFormer | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 6.483 ms | 25 - 25 MB | NPU | [EfficientFormer.onnx.zip](https://huggingface.co/qualcomm/EfficientFormer/blob/main/EfficientFormer_w8a8.onnx.zip) | ## Installation Install the package via pip: ```bash pip install "qai-hub-models[efficientformer]" ``` ## Configure Qualcomm® AI Hub to run this model on a cloud-hosted device Sign-in to [Qualcomm® AI Hub](https://app.aihub.qualcomm.com/) 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. ```bash qai-hub configure --api_token API_TOKEN ``` Navigate to [docs](https://app.aihub.qualcomm.com/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. ```bash python -m qai_hub_models.models.efficientformer.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.efficientformer.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. ```bash python -m qai_hub_models.models.efficientformer.export ``` ## How does this work? This [export script](https://aihub.qualcomm.com/models/efficientformer/qai_hub_models/models/EfficientFormer/export.py) leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) 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. ```python import torch import qai_hub as hub from qai_hub_models.models.efficientformer 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. ```python 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. ```python 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](https://myaccount.qualcomm.com/signup). ## Run demo on a cloud-hosted device You can also run the demo on-device. ```bash python -m qai_hub_models.models.efficientformer.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.efficientformer.demo -- --eval-mode on-device ``` ## Deploying compiled model to Android The models can be deployed using multiple runtimes: - TensorFlow Lite (`.tflite` export): [This tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a guide to deploy the .tflite model in an Android application. - QNN (`.so` export ): This [sample app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html) provides instructions on how to use the `.so` shared library in an Android application. ## View on Qualcomm® AI Hub Get more details on EfficientFormer's performance across various devices [here](https://aihub.qualcomm.com/models/efficientformer). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of EfficientFormer can be found [here](https://github.com/snap-research/EfficientFormer?tab=License-1-ov-file#readme). * The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf) ## References * [Rethinking Vision Transformers for MobileNet Size and Speed](https://arxiv.org/abs/2212.08059) * [Source Model Implementation](https://github.com/snap-research/EfficientFormer) ## Community * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI. * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).