--- library_name: pytorch license: mit pipeline_tag: image-classification tags: - foundation - android --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/openai_clip/web-assets/model_demo.png) # OpenAI-Clip: Optimized for Mobile Deployment ## Multi-modal foundational model for vision and language tasks like image/text similarity and for zero-shot image classification Contrastive Language-Image Pre-Training (CLIP) uses a ViT like transformer to get visual features and a causal language model to get the text features. Both the text and visual features can then be used for a variety of zero-shot learning tasks. This model is an implementation of OpenAI-Clip found [here](https://github.com/openai/CLIP/). This repository provides scripts to run OpenAI-Clip on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/openai_clip). ### Model Details - **Model Type:** Image classification - **Model Stats:** - Model checkpoint: ViT-B/16 - Image input resolution: 224x224 - Text context length: 77 - Number of parameters (CLIPTextEncoder): 76.0M - Model size (CLIPTextEncoder): 290 MB - Number of parameters (CLIPImageEncoder): 115M - Model size (CLIPImageEncoder): 437 MB | Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | CLIPTextEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 5.704 ms | 0 - 2 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.tflite) | | CLIPTextEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 4.73 ms | 0 - 23 MB | FP16 | NPU | [OpenAI-Clip.so](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.so) | | CLIPTextEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 35.21 ms | 0 - 131 MB | FP16 | NPU | [OpenAI-Clip.onnx](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.onnx) | | CLIPTextEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 4.08 ms | 0 - 195 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.tflite) | | CLIPTextEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 3.379 ms | 0 - 72 MB | FP16 | NPU | [OpenAI-Clip.so](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.so) | | CLIPTextEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 25.241 ms | 0 - 536 MB | FP16 | NPU | [OpenAI-Clip.onnx](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.onnx) | | CLIPTextEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 3.991 ms | 0 - 109 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.tflite) | | CLIPTextEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 2.754 ms | 0 - 65 MB | FP16 | NPU | Use Export Script | | CLIPTextEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 23.811 ms | 0 - 315 MB | FP16 | NPU | [OpenAI-Clip.onnx](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.onnx) | | CLIPTextEncoder | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 5.664 ms | 0 - 1004 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.tflite) | | CLIPTextEncoder | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 4.838 ms | 0 - 1 MB | FP16 | NPU | Use Export Script | | CLIPTextEncoder | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 5.761 ms | 0 - 2 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.tflite) | | CLIPTextEncoder | SA8255 (Proxy) | SA8255P Proxy | QNN | 4.832 ms | 0 - 1 MB | FP16 | NPU | Use Export Script | | CLIPTextEncoder | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 5.675 ms | 0 - 2 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.tflite) | | CLIPTextEncoder | SA8775 (Proxy) | SA8775P Proxy | QNN | 4.864 ms | 0 - 1 MB | FP16 | NPU | Use Export Script | | CLIPTextEncoder | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 5.702 ms | 0 - 2 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.tflite) | | CLIPTextEncoder | SA8650 (Proxy) | SA8650P Proxy | QNN | 4.893 ms | 0 - 2 MB | FP16 | NPU | Use Export Script | | CLIPTextEncoder | SA8295P ADP | SA8295P | TFLITE | 7.762 ms | 0 - 87 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.tflite) | | CLIPTextEncoder | SA8295P ADP | SA8295P | QNN | 6.779 ms | 0 - 6 MB | FP16 | NPU | Use Export Script | | CLIPTextEncoder | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 6.582 ms | 0 - 167 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.tflite) | | CLIPTextEncoder | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 5.291 ms | 0 - 68 MB | FP16 | NPU | Use Export Script | | CLIPTextEncoder | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 5.229 ms | 0 - 0 MB | FP16 | NPU | Use Export Script | | CLIPTextEncoder | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 38.327 ms | 127 - 127 MB | FP16 | NPU | [OpenAI-Clip.onnx](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPTextEncoder.onnx) | | CLIPImageEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 36.525 ms | 0 - 3 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.tflite) | | CLIPImageEncoder | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 27.189 ms | 0 - 51 MB | FP16 | NPU | [OpenAI-Clip.so](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.so) | | CLIPImageEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 29.37 ms | 0 - 666 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.tflite) | | CLIPImageEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 20.849 ms | 0 - 170 MB | FP16 | NPU | [OpenAI-Clip.so](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.so) | | CLIPImageEncoder | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 116.721 ms | 1 - 3571 MB | FP16 | NPU | [OpenAI-Clip.onnx](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.onnx) | | CLIPImageEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 25.793 ms | 0 - 461 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.tflite) | | CLIPImageEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 19.399 ms | 0 - 172 MB | FP16 | NPU | Use Export Script | | CLIPImageEncoder | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 109.572 ms | 1 - 2716 MB | FP16 | NPU | [OpenAI-Clip.onnx](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.onnx) | | CLIPImageEncoder | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 36.934 ms | 0 - 2 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.tflite) | | CLIPImageEncoder | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 21.964 ms | 1 - 2 MB | FP16 | NPU | Use Export Script | | CLIPImageEncoder | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 36.858 ms | 0 - 2 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.tflite) | | CLIPImageEncoder | SA8255 (Proxy) | SA8255P Proxy | QNN | 22.523 ms | 1 - 2 MB | FP16 | NPU | Use Export Script | | CLIPImageEncoder | SA8775 (Proxy) | SA8775P Proxy | TFLITE | 37.002 ms | 0 - 3 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.tflite) | | CLIPImageEncoder | SA8775 (Proxy) | SA8775P Proxy | QNN | 22.519 ms | 1 - 2 MB | FP16 | NPU | Use Export Script | | CLIPImageEncoder | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 37.107 ms | 1 - 3 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.tflite) | | CLIPImageEncoder | SA8650 (Proxy) | SA8650P Proxy | QNN | 22.501 ms | 1 - 2 MB | FP16 | NPU | Use Export Script | | CLIPImageEncoder | SA8295P ADP | SA8295P | TFLITE | 42.408 ms | 0 - 359 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.tflite) | | CLIPImageEncoder | SA8295P ADP | SA8295P | QNN | 26.489 ms | 1 - 6 MB | FP16 | NPU | Use Export Script | | CLIPImageEncoder | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 36.678 ms | 0 - 547 MB | FP16 | NPU | [OpenAI-Clip.tflite](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.tflite) | | CLIPImageEncoder | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 31.003 ms | 0 - 165 MB | FP16 | NPU | Use Export Script | | CLIPImageEncoder | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 161.714 ms | 188 - 188 MB | FP16 | NPU | [OpenAI-Clip.onnx](https://huggingface.co/qualcomm/OpenAI-Clip/blob/main/CLIPImageEncoder.onnx) | ## Installation This model can be installed as a Python package via pip. ```bash pip install "qai-hub-models[openai_clip]" ``` ## 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.openai_clip.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.openai_clip.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.openai_clip.export ``` ``` Profiling Results ------------------------------------------------------------ CLIPTextEncoder Device : Samsung Galaxy S23 (13) Runtime : TFLITE Estimated inference time (ms) : 5.7 Estimated peak memory usage (MB): [0, 2] Total # Ops : 660 Compute Unit(s) : NPU (658 ops) CPU (2 ops) ------------------------------------------------------------ CLIPImageEncoder Device : Samsung Galaxy S23 (13) Runtime : TFLITE Estimated inference time (ms) : 36.5 Estimated peak memory usage (MB): [0, 3] Total # Ops : 659 Compute Unit(s) : NPU (659 ops) ``` ## How does this work? This [export script](https://aihub.qualcomm.com/models/openai_clip/qai_hub_models/models/OpenAI-Clip/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.openai_clip import CLIPTextEncoder,CLIPImageEncoder # Load the model text_encoder_model = CLIPTextEncoder.from_pretrained() image_encoder_model = CLIPImageEncoder.from_pretrained() # Device device = hub.Device("Samsung Galaxy S23") # Trace model text_encoder_input_shape = text_encoder_model.get_input_spec() text_encoder_sample_inputs = text_encoder_model.sample_inputs() traced_text_encoder_model = torch.jit.trace(text_encoder_model, [torch.tensor(data[0]) for _, data in text_encoder_sample_inputs.items()]) # Compile model on a specific device text_encoder_compile_job = hub.submit_compile_job( model=traced_text_encoder_model , device=device, input_specs=text_encoder_model.get_input_spec(), ) # Get target model to run on-device text_encoder_target_model = text_encoder_compile_job.get_target_model() # Trace model image_encoder_input_shape = image_encoder_model.get_input_spec() image_encoder_sample_inputs = image_encoder_model.sample_inputs() traced_image_encoder_model = torch.jit.trace(image_encoder_model, [torch.tensor(data[0]) for _, data in image_encoder_sample_inputs.items()]) # Compile model on a specific device image_encoder_compile_job = hub.submit_compile_job( model=traced_image_encoder_model , device=device, input_specs=image_encoder_model.get_input_spec(), ) # Get target model to run on-device image_encoder_target_model = image_encoder_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 text_encoder_profile_job = hub.submit_profile_job( model=text_encoder_target_model, device=device, ) image_encoder_profile_job = hub.submit_profile_job( model=image_encoder_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 text_encoder_input_data = text_encoder_model.sample_inputs() text_encoder_inference_job = hub.submit_inference_job( model=text_encoder_target_model, device=device, inputs=text_encoder_input_data, ) text_encoder_inference_job.download_output_data() image_encoder_input_data = image_encoder_model.sample_inputs() image_encoder_inference_job = hub.submit_inference_job( model=image_encoder_target_model, device=device, inputs=image_encoder_input_data, ) image_encoder_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). ## 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 OpenAI-Clip's performance across various devices [here](https://aihub.qualcomm.com/models/openai_clip). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of OpenAI-Clip can be found [here](https://github.com/openai/CLIP/blob/main/LICENSE). * 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 * [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) * [Source Model Implementation](https://github.com/openai/CLIP/) ## 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).