--- library_name: pytorch license: mit tags: - bu_auto - android pipeline_tag: other --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/statetransformer/web-assets/model_demo.png) # StateTransformer: Optimized for Qualcomm Devices StateTransformer is a transformer-based model designed for trajectory prediction in self-driving scenarios. It integrates rasterized map data, agent context, and temporal dynamics to generate accurate future trajectories. 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/v0.64.0/src/qai_hub_models/models/statetransformer) 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 2.50, ONNX Runtime 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/statetransformer/releases/v0.64.0/statetransformer-onnx-float.zip) | TFLITE | float | Universal | | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/statetransformer/releases/v0.64.0/statetransformer-tflite-float.zip) For more device-specific assets and performance metrics, visit **[StateTransformer on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/statetransformer)**. ### Option 2: Export with Custom Configurations Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.64.0/src/qai_hub_models/models/statetransformer) 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 [StateTransformer on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.64.0/src/qai_hub_models/models/statetransformer) for usage instructions. ## Model Details **Model Type:** Model_use_case.driver_assistance **Model Stats:** - Input resolution: 1x224x224x58, 1x224x224x58, 1x4x7 - Model checkpoint: pretrained-mixtral-small - Model size (float): 348 MB - Number of parameters: 90.7M ## Performance Summary | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |---|---|---|---|---|---|--- | StateTransformer | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 189.466 ms | 183 - 4352 MB | NPU | StateTransformer | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 196.612 ms | 168 - 4266 MB | NPU | StateTransformer | ONNX | float | Snapdragon® X2 Elite | 178.504 ms | 278 - 278 MB | NPU | StateTransformer | ONNX | float | Snapdragon® X Elite | 474.056 ms | 278 - 278 MB | NPU | StateTransformer | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 323.258 ms | 100 - 142 MB | NPU | StateTransformer | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 277.911 ms | 105 - 146 MB | NPU | StateTransformer | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 474.056 ms | 278 - 278 MB | NPU | StateTransformer | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 196.612 ms | 168 - 4266 MB | NPU | StateTransformer | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 656.979 ms | 212 - 3148 MB | CPU | StateTransformer | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 913.992 ms | 193 - 4257 MB | CPU | StateTransformer | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 1551.174 ms | 294 - 813 MB | CPU | StateTransformer | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1059.622 ms | 294 - 298 MB | CPU | StateTransformer | TFLITE | float | Qualcomm® SA8775P | 1274.405 ms | 294 - 4057 MB | CPU | StateTransformer | TFLITE | float | Qualcomm® SA8650P | 1274.405 ms | 294 - 4057 MB | CPU | StateTransformer | TFLITE | float | Qualcomm® SA8255P | 1274.405 ms | 294 - 4057 MB | CPU | StateTransformer | TFLITE | float | Qualcomm® QCS8450 | 808.833 ms | 450 - 470 MB | CPU | StateTransformer | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 1309.664 ms | 293 - 812 MB | CPU | StateTransformer | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 656.979 ms | 212 - 3148 MB | CPU | StateTransformer | TFLITE | float | Qualcomm® SA8295P | 947.527 ms | 290 - 3937 MB | CPU ## License * The license for the original implementation of StateTransformer can be found [here](https://github.com/Tsinghua-MARS-Lab/StateTransformer/blob/main/setup.py). ## 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).