README.md

---
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/sequencer2d/web-assets/model_demo.png)

# Sequencer2D: Optimized for Qualcomm Devices

Sequencer2D is a vision transformer model that can classify images from the Imagenet dataset.

This is based on the implementation of Sequencer2D found [here](https://github.com/okojoalg/sequencer).
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/v0.60.0/src/qai_hub_models/models/sequencer2d) 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/sequencer2d/releases/v0.60.0/sequencer2d-onnx-float.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sequencer2d/releases/v0.60.0/sequencer2d-qnn_dlc-float.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sequencer2d/releases/v0.60.0/sequencer2d-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sequencer2d/releases/v0.60.0/sequencer2d-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[Sequencer2D on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/sequencer2d)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/sequencer2d) 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 [Sequencer2D on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/sequencer2d) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Input resolution: 224x224
- Model checkpoint: sequencer2d_s
- Model size (float): 106 MB
- Model size (w8a8): 69.1 MB
- Number of parameters: 27.6M

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| Sequencer2D | ONNX | float | Snapdragon® X2 Elite | 9.833 ms | 2 - 2 MB | NPU
| Sequencer2D | ONNX | float | Snapdragon® X Elite | 17.2 ms | 60 - 60 MB | NPU
| Sequencer2D | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 10.864 ms | 0 - 954 MB | NPU
| Sequencer2D | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 22.66 ms | 0 - 655 MB | NPU
| Sequencer2D | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 16.414 ms | 0 - 4 MB | NPU
| Sequencer2D | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 16.036 ms | 1 - 4 MB | NPU
| Sequencer2D | ONNX | float | Qualcomm® QCS8450 | 22.66 ms | 0 - 655 MB | NPU
| Sequencer2D | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 22.723 ms | 1 - 4 MB | NPU
| Sequencer2D | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 17.2 ms | 60 - 60 MB | NPU
| Sequencer2D | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 9.634 ms | 0 - 736 MB | NPU
| Sequencer2D | ONNX | float | Snapdragon® 8 Elite Mobile | 9.634 ms | 0 - 736 MB | NPU
| Sequencer2D | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 8.05 ms | 0 - 892 MB | NPU
| Sequencer2D | QNN_DLC | float | Snapdragon® X2 Elite | 11.815 ms | 1 - 1 MB | NPU
| Sequencer2D | QNN_DLC | float | Snapdragon® X Elite | 21.605 ms | 1 - 1 MB | NPU
| Sequencer2D | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 14.259 ms | 0 - 2197 MB | NPU
| Sequencer2D | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 25.475 ms | 0 - 815 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 21.122 ms | 1 - 4 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 51.597 ms | 1 - 981 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 20.139 ms | 1 - 3 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® SA8775P | 24.062 ms | 1 - 900 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® SA8650P | 24.062 ms | 1 - 900 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® SA8255P | 24.062 ms | 1 - 900 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® QCS8450 | 25.475 ms | 0 - 815 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 23.998 ms | 1 - 3 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 21.605 ms | 1 - 1 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 12.251 ms | 1 - 1024 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® SA7255P | 51.597 ms | 1 - 981 MB | NPU
| Sequencer2D | QNN_DLC | float | Qualcomm® SA8295P | 25.962 ms | 1 - 656 MB | NPU
| Sequencer2D | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 12.251 ms | 1 - 1024 MB | NPU
| Sequencer2D | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 9.549 ms | 0 - 1186 MB | NPU
| Sequencer2D | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 11.882 ms | 0 - 977 MB | NPU
| Sequencer2D | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 21.126 ms | 0 - 734 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 16.733 ms | 0 - 76 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 37.481 ms | 0 - 796 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 17.436 ms | 0 - 9 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® SA8775P | 19.071 ms | 0 - 795 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® SA8650P | 19.071 ms | 0 - 795 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® SA8255P | 19.071 ms | 0 - 795 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® QCS8450 | 21.126 ms | 0 - 734 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 18.534 ms | 0 - 75 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 9.424 ms | 0 - 784 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® SA7255P | 37.481 ms | 0 - 796 MB | NPU
| Sequencer2D | TFLITE | float | Qualcomm® SA8295P | 20.559 ms | 0 - 485 MB | NPU
| Sequencer2D | TFLITE | float | Snapdragon® 8 Elite Mobile | 9.424 ms | 0 - 784 MB | NPU
| Sequencer2D | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 7.225 ms | 0 - 761 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 13.595 ms | 0 - 947 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 21.905 ms | 0 - 811 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 18.078 ms | 0 - 64 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 36.323 ms | 0 - 862 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 18.834 ms | 0 - 10 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® SA8775P | 20.424 ms | 0 - 862 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® SA8650P | 20.424 ms | 0 - 862 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® SA8255P | 20.424 ms | 0 - 862 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® QCS8450 | 21.905 ms | 0 - 811 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 19.751 ms | 0 - 64 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 10.902 ms | 0 - 802 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® SA7255P | 36.323 ms | 0 - 862 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Qualcomm® SA8295P | 22.914 ms | 0 - 616 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 10.902 ms | 0 - 802 MB | NPU
| Sequencer2D | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 9.088 ms | 0 - 823 MB | NPU

## License
* The license for the original implementation of Sequencer2D can be found
  [here](https://github.com/okojoalg/sequencer/blob/main/LICENSE).

## References
* [Sequencer: Deep LSTM for Image Classification](https://arxiv.org/abs/2205.01972)
* [Source Model Implementation](https://github.com/okojoalg/sequencer)

## 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).