--- library_name: pytorch license: apache-2.0 tags: - foundation - android pipeline_tag: image-classification --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/siglip2/web-assets/model_demo.png) # SigLIP2: Optimized for Qualcomm Devices SigLIP2 (Sigmoid Loss for Language-Image Pre-training 2) is a vision-language model from Google that computes cosine-similarity scores between images and text prompts. It can be used for zero-shot image classification, image search, and content moderation without any task-specific fine-tuning. This is based on the implementation of SigLIP2 found [here](https://huggingface.co/google/siglip2-base-patch16-224). 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/v0.63.0/src/qai_hub_models/models/siglip2) 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 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/siglip2/releases/v0.63.0/siglip2-onnx-float.zip) | ONNX | w8a16 | Universal | QAIRT 2.50, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/siglip2/releases/v0.63.0/siglip2-onnx-w8a16.zip) | QNN_DLC | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/siglip2/releases/v0.63.0/siglip2-qnn_dlc-float.zip) | QNN_DLC | w8a16 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/siglip2/releases/v0.63.0/siglip2-qnn_dlc-w8a16.zip) | TFLITE | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/siglip2/releases/v0.63.0/siglip2-tflite-float.zip) For more device-specific assets and performance metrics, visit **[SigLIP2 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/siglip2)**. ### Option 2: Export with Custom Configurations Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.63.0/src/qai_hub_models/models/siglip2) 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 [SigLIP2 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.63.0/src/qai_hub_models/models/siglip2) for usage instructions. ## Model Details **Model Type:** Model_use_case.image_classification **Model Stats:** - Image input resolution: 224x224 - Model checkpoint: google/siglip2-base-patch16-224 - Model size (image_encoder) (float): 352 MB - Model size (image_encoder) (w8a16): 92.8 MB - Model size (text_encoder) (float): 1.05 GB - Model size (text_encoder) (w8a16): 461 MB - Text sequence length: 64 ## Performance Summary | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |---|---|---|---|---|---|--- | image_encoder | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 3.338 ms | 1 - 169 MB | NPU | image_encoder | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.189 ms | 0 - 168 MB | NPU | image_encoder | ONNX | float | Snapdragon® X2 Elite | 3.53 ms | 2 - 2 MB | NPU | image_encoder | ONNX | float | Snapdragon® X Elite | 8.735 ms | 182 - 182 MB | NPU | image_encoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 5.814 ms | 1 - 304 MB | NPU | image_encoder | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 12.987 ms | 1 - 294 MB | NPU | image_encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 11.854 ms | 0 - 5 MB | NPU | image_encoder | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 8.332 ms | 0 - 194 MB | NPU | image_encoder | ONNX | float | Qualcomm® QCS8450 | 12.987 ms | 1 - 294 MB | NPU | image_encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 11.318 ms | 0 - 4 MB | NPU | image_encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 8.735 ms | 182 - 182 MB | NPU | image_encoder | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 4.189 ms | 0 - 168 MB | NPU | image_encoder | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 2.116 ms | 0 - 194 MB | NPU | image_encoder | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 2.845 ms | 0 - 192 MB | NPU | image_encoder | ONNX | w8a16 | Snapdragon® X2 Elite | 2.357 ms | 1 - 1 MB | NPU | image_encoder | ONNX | w8a16 | Snapdragon® X Elite | 5.535 ms | 91 - 91 MB | NPU | image_encoder | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 3.683 ms | 0 - 295 MB | NPU | image_encoder | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 7.156 ms | 0 - 285 MB | NPU | image_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 18.752 ms | 2 - 5 MB | NPU | image_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 5.086 ms | 0 - 4 MB | NPU | image_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 5.28 ms | 0 - 98 MB | NPU | image_encoder | ONNX | w8a16 | Qualcomm® QCS8450 | 7.156 ms | 0 - 285 MB | NPU | image_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 5.249 ms | 0 - 4 MB | NPU | image_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 5.535 ms | 91 - 91 MB | NPU | image_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 53.641 ms | 0 - 376 MB | NPU | image_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 6.972 ms | 0 - 256 MB | NPU | image_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 2.845 ms | 0 - 192 MB | NPU | image_encoder | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 6.972 ms | 0 - 256 MB | NPU | image_encoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 3.421 ms | 1 - 166 MB | NPU | image_encoder | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.237 ms | 0 - 163 MB | NPU | image_encoder | QNN_DLC | float | Snapdragon® X2 Elite | 3.903 ms | 1 - 1 MB | NPU | image_encoder | QNN_DLC | float | Snapdragon® X Elite | 9.12 ms | 1 - 1 MB | NPU | image_encoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 5.954 ms | 0 - 253 MB | NPU | image_encoder | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 13.211 ms | 1 - 248 MB | NPU | image_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 12.021 ms | 1 - 4 MB | NPU | image_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 8.494 ms | 1 - 2 MB | NPU | image_encoder | QNN_DLC | float | Qualcomm® QCS8450 | 13.211 ms | 1 - 248 MB | NPU | image_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 11.246 ms | 1 - 3 MB | NPU | image_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 9.12 ms | 1 - 1 MB | NPU | image_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 4.237 ms | 0 - 163 MB | NPU | image_encoder | QNN_DLC | float | Qualcomm® SA8295P | 13.937 ms | 1 - 153 MB | NPU | image_encoder | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 2.46 ms | 0 - 183 MB | NPU | image_encoder | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 3.306 ms | 0 - 179 MB | NPU | image_encoder | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 3.04 ms | 0 - 0 MB | NPU | image_encoder | QNN_DLC | w8a16 | Snapdragon® X Elite | 6.828 ms | 0 - 0 MB | NPU | image_encoder | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 4.255 ms | 0 - 287 MB | NPU | image_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 5.823 ms | 0 - 3 MB | NPU | image_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 6.294 ms | 0 - 2 MB | NPU | image_encoder | QNN_DLC | w8a16 | Qualcomm® SA8650P | 8.664 ms | 0 - 255 MB | NPU | image_encoder | QNN_DLC | w8a16 | Qualcomm® SA8255P | 8.664 ms | 0 - 255 MB | NPU | image_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 6.231 ms | 0 - 3 MB | NPU | image_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 6.828 ms | 0 - 0 MB | NPU | image_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 65.954 ms | 0 - 361 MB | NPU | image_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 7.864 ms | 0 - 262 MB | NPU | image_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 3.306 ms | 0 - 179 MB | NPU | image_encoder | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 7.864 ms | 0 - 262 MB | NPU | image_encoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 3.299 ms | 0 - 169 MB | NPU | image_encoder | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.108 ms | 0 - 165 MB | NPU | image_encoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 5.881 ms | 0 - 252 MB | NPU | image_encoder | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 12.937 ms | 0 - 244 MB | NPU | image_encoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 12.057 ms | 0 - 185 MB | NPU | image_encoder | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 7.95 ms | 0 - 3 MB | NPU | image_encoder | TFLITE | float | Qualcomm® SA8775P | 11.177 ms | 0 - 163 MB | NPU | image_encoder | TFLITE | float | Qualcomm® SA8650P | 11.177 ms | 0 - 163 MB | NPU | image_encoder | TFLITE | float | Qualcomm® SA8255P | 11.177 ms | 0 - 163 MB | NPU | image_encoder | TFLITE | float | Qualcomm® QCS8450 | 12.937 ms | 0 - 244 MB | NPU | image_encoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 10.971 ms | 0 - 184 MB | NPU | image_encoder | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 4.108 ms | 0 - 165 MB | NPU | image_encoder | TFLITE | float | Qualcomm® SA7255P | 400.273 ms | 0 - 11 MB | CPU | image_encoder | TFLITE | float | Qualcomm® SA8295P | 13.752 ms | 0 - 151 MB | NPU | text_encoder | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 2.41 ms | 0 - 144 MB | NPU | text_encoder | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.419 ms | 0 - 144 MB | NPU | text_encoder | ONNX | float | Snapdragon® X2 Elite | 1.406 ms | 1 - 1 MB | NPU | text_encoder | ONNX | float | Snapdragon® X Elite | 2.899 ms | 543 - 543 MB | NPU | text_encoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 2.723 ms | 0 - 255 MB | NPU | text_encoder | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 4.081 ms | 0 - 249 MB | NPU | text_encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 3.771 ms | 0 - 4 MB | NPU | text_encoder | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.526 ms | 0 - 643 MB | NPU | text_encoder | ONNX | float | Qualcomm® QCS8450 | 4.081 ms | 0 - 249 MB | NPU | text_encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 3.515 ms | 0 - 3 MB | NPU | text_encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 2.899 ms | 543 - 543 MB | NPU | text_encoder | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 2.419 ms | 0 - 144 MB | NPU | text_encoder | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 1.235 ms | 0 - 192 MB | NPU | text_encoder | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 1.247 ms | 0 - 190 MB | NPU | text_encoder | ONNX | w8a16 | Snapdragon® X2 Elite | 0.757 ms | 1 - 1 MB | NPU | text_encoder | ONNX | w8a16 | Snapdragon® X Elite | 1.676 ms | 459 - 459 MB | NPU | text_encoder | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.452 ms | 0 - 277 MB | NPU | text_encoder | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 2.223 ms | 0 - 279 MB | NPU | text_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 6.553 ms | 0 - 3 MB | NPU | text_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 2.078 ms | 0 - 4 MB | NPU | text_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.853 ms | 0 - 515 MB | NPU | text_encoder | ONNX | w8a16 | Qualcomm® QCS8450 | 2.223 ms | 0 - 279 MB | NPU | text_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 2.193 ms | 0 - 3 MB | NPU | text_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 1.676 ms | 459 - 459 MB | NPU | text_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 13.139 ms | 0 - 249 MB | NPU | text_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 2.859 ms | 0 - 243 MB | NPU | text_encoder | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 1.247 ms | 0 - 190 MB | NPU | text_encoder | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 2.859 ms | 0 - 243 MB | NPU | text_encoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 2.394 ms | 0 - 139 MB | NPU | text_encoder | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.431 ms | 0 - 139 MB | NPU | text_encoder | QNN_DLC | float | Snapdragon® X2 Elite | 1.695 ms | 0 - 0 MB | NPU | text_encoder | QNN_DLC | float | Snapdragon® X Elite | 3.072 ms | 0 - 0 MB | NPU | text_encoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 2.704 ms | 0 - 251 MB | NPU | text_encoder | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 4.066 ms | 0 - 243 MB | NPU | text_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 3.745 ms | 0 - 3 MB | NPU | text_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.525 ms | 0 - 2 MB | NPU | text_encoder | QNN_DLC | float | Qualcomm® QCS8450 | 4.066 ms | 0 - 243 MB | NPU | text_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 3.52 ms | 0 - 2 MB | NPU | text_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 3.072 ms | 0 - 0 MB | NPU | text_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 2.431 ms | 0 - 139 MB | NPU | text_encoder | QNN_DLC | float | Qualcomm® SA8295P | 4.508 ms | 0 - 129 MB | NPU | text_encoder | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 1.232 ms | 0 - 186 MB | NPU | text_encoder | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 1.276 ms | 0 - 186 MB | NPU | text_encoder | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 0.979 ms | 1 - 1 MB | NPU | text_encoder | QNN_DLC | w8a16 | Snapdragon® X Elite | 1.945 ms | 0 - 0 MB | NPU | text_encoder | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.46 ms | 0 - 268 MB | NPU | text_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 2.063 ms | 0 - 3 MB | NPU | text_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.868 ms | 0 - 2 MB | NPU | text_encoder | QNN_DLC | w8a16 | Qualcomm® SA8650P | 2.216 ms | 0 - 226 MB | NPU | text_encoder | QNN_DLC | w8a16 | Qualcomm® SA8255P | 2.216 ms | 0 - 226 MB | NPU | text_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 2.068 ms | 0 - 2 MB | NPU | text_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 1.945 ms | 0 - 0 MB | NPU | text_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 13.676 ms | 0 - 247 MB | NPU | text_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 2.956 ms | 0 - 241 MB | NPU | text_encoder | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 1.276 ms | 0 - 186 MB | NPU | text_encoder | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 2.956 ms | 0 - 241 MB | NPU | text_encoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 2.414 ms | 0 - 151 MB | NPU | text_encoder | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.406 ms | 0 - 151 MB | NPU | text_encoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 2.741 ms | 0 - 261 MB | NPU | text_encoder | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 4.081 ms | 0 - 246 MB | NPU | text_encoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 3.821 ms | 0 - 544 MB | NPU | text_encoder | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.515 ms | 0 - 2 MB | NPU | text_encoder | TFLITE | float | Qualcomm® SA8775P | 3.688 ms | 0 - 145 MB | NPU | text_encoder | TFLITE | float | Qualcomm® SA8650P | 3.688 ms | 0 - 145 MB | NPU | text_encoder | TFLITE | float | Qualcomm® SA8255P | 3.688 ms | 0 - 145 MB | NPU | text_encoder | TFLITE | float | Qualcomm® QCS8450 | 4.081 ms | 0 - 246 MB | NPU | text_encoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 3.528 ms | 0 - 543 MB | NPU | text_encoder | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 2.406 ms | 0 - 151 MB | NPU | text_encoder | TFLITE | float | Qualcomm® SA7255P | 107.151 ms | 0 - 29 MB | GPU | text_encoder | TFLITE | float | Qualcomm® SA8295P | 4.537 ms | 0 - 132 MB | NPU ## License * The license for the original implementation of SigLIP2 can be found [here](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md). ## References * [SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features](https://arxiv.org/abs/2502.14786) * [Source Model Implementation](https://huggingface.co/google/siglip2-base-patch16-224) ## 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).