| --- |
| library_name: pytorch |
| license: other |
| tags: |
| - bu_auto |
| - android |
| pipeline_tag: image-classification |
|
|
| --- |
| |
|  |
|
|
| # EfficientFormer: Optimized for Qualcomm Devices |
|
|
| EfficientFormer is a vision transformer model that can classify images from the Imagenet dataset. |
|
|
| This is based on the implementation of EfficientFormer found [here](https://github.com/snap-research/EfficientFormer). |
| This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/efficientformer) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary). |
|
|
| Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device. |
|
|
| ## Getting Started |
| There are two ways to deploy this model on your device: |
|
|
| ### Option 1: Download Pre-Exported Models |
|
|
| Below are pre-exported model assets ready for deployment. |
|
|
| | Runtime | Precision | Chipset | SDK Versions | Download | |
| |---|---|---|---|---| |
| | ONNX | float | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientformer/releases/v0.59.0/efficientformer-onnx-float.zip) |
| | ONNX | w8a16 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientformer/releases/v0.59.0/efficientformer-onnx-w8a16.zip) |
| | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientformer/releases/v0.59.0/efficientformer-qnn_dlc-float.zip) |
| | QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientformer/releases/v0.59.0/efficientformer-qnn_dlc-w8a16.zip) |
| | TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/efficientformer/releases/v0.59.0/efficientformer-tflite-float.zip) |
|
|
| For more device-specific assets and performance metrics, visit **[EfficientFormer on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/efficientformer)**. |
|
|
|
|
| ### Option 2: Export with Custom Configurations |
|
|
| Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/efficientformer) Python library to compile and export the model with your own: |
| - Custom weights (e.g., fine-tuned checkpoints) |
| - Custom input shapes |
| - Target device and runtime configurations |
|
|
| This option is ideal if you need to customize the model beyond the default configuration provided here. |
|
|
| See our repository for [EfficientFormer on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/efficientformer) for usage instructions. |
|
|
| ## 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 |
| |
| ## Performance Summary |
| | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
| |---|---|---|---|---|---|--- |
| | EfficientFormer | ONNX | float | Snapdragon® X2 Elite | 0.636 ms | 2 - 2 MB | NPU |
| | EfficientFormer | ONNX | float | Snapdragon® X Elite | 1.382 ms | 25 - 25 MB | NPU |
| | EfficientFormer | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 0.922 ms | 0 - 80 MB | NPU |
| | EfficientFormer | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 5.091 ms | 1 - 83 MB | NPU |
| | EfficientFormer | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.318 ms | 0 - 137 MB | NPU |
| | EfficientFormer | ONNX | float | Qualcomm® QCS8450 | 5.091 ms | 1 - 83 MB | NPU |
| | EfficientFormer | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 1.765 ms | 1 - 3 MB | NPU |
| | EfficientFormer | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 1.382 ms | 25 - 25 MB | NPU |
| | EfficientFormer | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 0.695 ms | 0 - 44 MB | NPU |
| | EfficientFormer | ONNX | float | Snapdragon® 8 Elite Mobile | 0.695 ms | 0 - 44 MB | NPU |
| | EfficientFormer | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.601 ms | 0 - 44 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Snapdragon® X2 Elite | 0.564 ms | 1 - 1 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Snapdragon® X Elite | 1.371 ms | 13 - 13 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 0.896 ms | 0 - 91 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 1.783 ms | 0 - 92 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 5.14 ms | 0 - 3 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.33 ms | 0 - 16 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Qualcomm® QCS8450 | 1.783 ms | 0 - 92 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 1.914 ms | 0 - 3 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 1.371 ms | 13 - 13 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 0.604 ms | 0 - 61 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 0.604 ms | 0 - 61 MB | NPU |
| | EfficientFormer | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.53 ms | 0 - 71 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Snapdragon® X2 Elite | 0.872 ms | 1 - 1 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Snapdragon® X Elite | 1.677 ms | 1 - 1 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 1.039 ms | 0 - 77 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 5.579 ms | 0 - 78 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 4.922 ms | 1 - 39 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.475 ms | 1 - 2 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® SA8775P | 2.059 ms | 1 - 43 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® SA8650P | 2.059 ms | 1 - 43 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® SA8255P | 2.059 ms | 1 - 43 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® QCS8450 | 5.579 ms | 0 - 78 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 1.9 ms | 3 - 5 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 1.677 ms | 1 - 1 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 0.786 ms | 0 - 44 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® SA7255P | 4.922 ms | 1 - 39 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Qualcomm® SA8295P | 4.008 ms | 0 - 39 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 0.786 ms | 0 - 44 MB | NPU |
| | EfficientFormer | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.645 ms | 1 - 45 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 0.864 ms | 0 - 0 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Snapdragon® X Elite | 1.785 ms | 0 - 0 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.086 ms | 0 - 80 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8275 | 3.226 ms | 0 - 57 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.553 ms | 0 - 2 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® SA8775P | 1.908 ms | 0 - 59 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® SA8650P | 1.908 ms | 0 - 59 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® SA8255P | 1.908 ms | 0 - 59 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 1.734 ms | 0 - 2 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 1.785 ms | 0 - 0 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 6.989 ms | 0 - 178 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 1.696 ms | 0 - 62 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 0.708 ms | 0 - 51 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Qualcomm® SA7255P | 3.226 ms | 0 - 57 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 0.708 ms | 0 - 51 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.618 ms | 0 - 66 MB | NPU |
| | EfficientFormer | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1.696 ms | 0 - 62 MB | NPU |
| | EfficientFormer | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.037 ms | 0 - 98 MB | NPU |
| | EfficientFormer | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 5.535 ms | 0 - 96 MB | NPU |
| | EfficientFormer | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 4.864 ms | 0 - 52 MB | NPU |
| | EfficientFormer | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.477 ms | 0 - 6 MB | NPU |
| | EfficientFormer | TFLITE | float | Qualcomm® SA8775P | 2.065 ms | 0 - 55 MB | NPU |
| | EfficientFormer | TFLITE | float | Qualcomm® SA8650P | 2.065 ms | 0 - 55 MB | NPU |
| | EfficientFormer | TFLITE | float | Qualcomm® SA8255P | 2.065 ms | 0 - 55 MB | NPU |
| | EfficientFormer | TFLITE | float | Qualcomm® QCS8450 | 5.535 ms | 0 - 96 MB | NPU |
| | EfficientFormer | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 1.872 ms | 0 - 27 MB | NPU |
| | EfficientFormer | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 0.767 ms | 0 - 58 MB | NPU |
| | EfficientFormer | TFLITE | float | Qualcomm® SA7255P | 4.864 ms | 0 - 52 MB | NPU |
| | EfficientFormer | TFLITE | float | Qualcomm® SA8295P | 3.977 ms | 0 - 48 MB | NPU |
| | EfficientFormer | TFLITE | float | Snapdragon® 8 Elite Mobile | 0.767 ms | 0 - 58 MB | NPU |
| | EfficientFormer | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.652 ms | 0 - 53 MB | NPU |
| |
| ## 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). |
| |
| ## 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). |
| |
