Hugging Face's logo

qualcomm
/
Pi0.5

Robotics
PyTorch
generative_ai
android
Model card Files
xet
 Community

Pi0.5: Optimized for Qualcomm Devices

Pi0.5 is a vision-language-action model that co-trains on diverse data sources (robot demos, web data, semantic subtasks) to enable open-world generalization for long-horizon robotic manipulation.

This is based on the implementation of Pi0.5 found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up 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
PRECOMPILED_QNN_ONNX mixed Snapdragon® X2 Elite QAIRT 2.45, ONNX Runtime 1.25.0 Download
PRECOMPILED_QNN_ONNX mixed Snapdragon® X Elite QAIRT 2.45, ONNX Runtime 1.25.0 Download
PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Gen 3 Mobile QAIRT 2.45, ONNX Runtime 1.25.0 Download
PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS8550 (Proxy) QAIRT 2.45, ONNX Runtime 1.25.0 Download
PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Elite Mobile QAIRT 2.45, ONNX Runtime 1.25.0 Download
PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Elite Gen 5 Mobile QAIRT 2.45, ONNX Runtime 1.25.0 Download
PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS9075 QAIRT 2.45, ONNX Runtime 1.25.0 Download
QNN_CONTEXT_BINARY mixed Snapdragon® X2 Elite QAIRT 2.45 Download
QNN_CONTEXT_BINARY mixed Snapdragon® X Elite QAIRT 2.45 Download
QNN_CONTEXT_BINARY mixed Snapdragon® 8 Gen 3 Mobile QAIRT 2.45 Download
QNN_CONTEXT_BINARY mixed Qualcomm® QCS8550 (Proxy) QAIRT 2.45 Download
QNN_CONTEXT_BINARY mixed Qualcomm® SA8775P QAIRT 2.45 Download
QNN_CONTEXT_BINARY mixed Snapdragon® 8 Elite Mobile QAIRT 2.45 Download
QNN_CONTEXT_BINARY mixed Snapdragon® 8 Elite Gen 5 Mobile QAIRT 2.45 Download
QNN_CONTEXT_BINARY mixed Qualcomm® SA7255P QAIRT 2.45 Download
QNN_CONTEXT_BINARY mixed Qualcomm® QCS9075 QAIRT 2.45 Download

For more device-specific assets and performance metrics, visit Pi0.5 on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models 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 Pi0.5 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.robotics

Model Stats:

  • Number of cameras: 3
  • Action chunk size: 50
  • Vision resolution: 224x224
  • Quantization: Mixed (w4a16 backbone, w8a16 vision encoder and action expert)

Performance Summary

Model Runtime Precision Chipset Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit
action_expert PRECOMPILED_QNN_ONNX mixed Snapdragon® X2 Elite 13.753 ms 35 - 35 MB NPU
action_expert PRECOMPILED_QNN_ONNX mixed Snapdragon® X Elite 28.304 ms 387 - 387 MB NPU
action_expert PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Gen 3 Mobile 22.039 ms 35 - 45 MB NPU
action_expert PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS8550 (Proxy) 29.558 ms 35 - 37 MB NPU
action_expert PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS9075 30.824 ms 34 - 71 MB NPU
action_expert PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Elite Gen 5 Mobile 14.132 ms 30 - 41 MB NPU
action_expert PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Elite Mobile 17.134 ms 15 - 26 MB NPU
action_expert PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS8750 17.134 ms 15 - 26 MB NPU
action_expert PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS7181 28.304 ms 387 - 387 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Snapdragon® X2 Elite 14.41 ms 34 - 34 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Snapdragon® X Elite 28.624 ms 34 - 34 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Snapdragon® 8 Gen 3 Mobile 22.167 ms 35 - 42 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Qualcomm® QCS8275 53.807 ms 7 - 14 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Qualcomm® QCS8550 (Proxy) 29.498 ms 35 - 36 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Qualcomm® QCS9075 31.193 ms 34 - 70 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Snapdragon® 8 Elite Gen 5 Mobile 14.093 ms 29 - 37 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Snapdragon® 8 Elite Mobile 17.337 ms 15 - 24 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Qualcomm® SA7255P 53.807 ms 7 - 14 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Qualcomm® QCS8750 17.337 ms 15 - 24 MB NPU
action_expert QNN_CONTEXT_BINARY mixed Qualcomm® QCS7181 28.624 ms 34 - 34 MB NPU
backbone PRECOMPILED_QNN_ONNX mixed Snapdragon® X2 Elite 253.689 ms 131 - 131 MB NPU
backbone PRECOMPILED_QNN_ONNX mixed Snapdragon® X Elite 444.201 ms 924 - 924 MB NPU
backbone PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Gen 3 Mobile 333.663 ms 85 - 97 MB NPU
backbone PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS8550 (Proxy) 450.247 ms 0 - 1013 MB NPU
backbone PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS9075 399.046 ms 12 - 27 MB NPU
backbone PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Elite Gen 5 Mobile 263.734 ms 74 - 86 MB NPU
backbone PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Elite Mobile 292.223 ms 76 - 87 MB NPU
backbone PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS8750 292.223 ms 76 - 87 MB NPU
backbone PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS7181 444.201 ms 924 - 924 MB NPU
backbone QNN_CONTEXT_BINARY mixed Snapdragon® X2 Elite 254.267 ms 12 - 12 MB NPU
backbone QNN_CONTEXT_BINARY mixed Snapdragon® X Elite 446.563 ms 12 - 12 MB NPU
backbone QNN_CONTEXT_BINARY mixed Snapdragon® 8 Gen 3 Mobile 331.429 ms 12 - 20 MB NPU
backbone QNN_CONTEXT_BINARY mixed Qualcomm® QCS8275 1093.594 ms 3 - 11 MB NPU
backbone QNN_CONTEXT_BINARY mixed Qualcomm® QCS8550 (Proxy) 455.733 ms 12 - 14 MB NPU
backbone QNN_CONTEXT_BINARY mixed Qualcomm® QCS9075 398.129 ms 12 - 60 MB NPU
backbone QNN_CONTEXT_BINARY mixed Snapdragon® 8 Elite Gen 5 Mobile 261.782 ms 4 - 12 MB NPU
backbone QNN_CONTEXT_BINARY mixed Snapdragon® 8 Elite Mobile 290.939 ms 1 - 14 MB NPU
backbone QNN_CONTEXT_BINARY mixed Qualcomm® SA7255P 1093.594 ms 3 - 11 MB NPU
backbone QNN_CONTEXT_BINARY mixed Qualcomm® QCS8750 290.939 ms 1 - 14 MB NPU
backbone QNN_CONTEXT_BINARY mixed Qualcomm® QCS7181 446.563 ms 12 - 12 MB NPU
token_emb PRECOMPILED_QNN_ONNX mixed Snapdragon® X2 Elite 2.097 ms 39 - 39 MB NPU
token_emb PRECOMPILED_QNN_ONNX mixed Snapdragon® X Elite 3.32 ms 1002 - 1002 MB NPU
token_emb PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Gen 3 Mobile 2.76 ms 0 - 12 MB NPU
token_emb PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS8550 (Proxy) 3.633 ms 0 - 1019 MB NPU
token_emb PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS9075 4.329 ms 6 - 15 MB NPU
token_emb PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Elite Gen 5 Mobile 2.211 ms 1 - 13 MB NPU
token_emb PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Elite Mobile 2.883 ms 1 - 12 MB NPU
token_emb PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS8750 2.883 ms 1 - 12 MB NPU
token_emb PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS7181 3.32 ms 1002 - 1002 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Snapdragon® X2 Elite 2.257 ms 6 - 6 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Snapdragon® X Elite 3.436 ms 6 - 6 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Snapdragon® 8 Gen 3 Mobile 2.757 ms 2 - 9 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Qualcomm® QCS8275 6.335 ms 0 - 8 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Qualcomm® QCS8550 (Proxy) 3.597 ms 2 - 16 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Qualcomm® QCS9075 4.219 ms 6 - 26 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Snapdragon® 8 Elite Gen 5 Mobile 2.192 ms 2 - 12 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Snapdragon® 8 Elite Mobile 2.787 ms 0 - 9 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Qualcomm® SA7255P 6.335 ms 0 - 8 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Qualcomm® QCS8750 2.787 ms 0 - 9 MB NPU
token_emb QNN_CONTEXT_BINARY mixed Qualcomm® QCS7181 3.436 ms 6 - 6 MB NPU
vision_encoder PRECOMPILED_QNN_ONNX mixed Snapdragon® X2 Elite 16.79 ms 5 - 5 MB NPU
vision_encoder PRECOMPILED_QNN_ONNX mixed Snapdragon® X Elite 42.973 ms 516 - 516 MB NPU
vision_encoder PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Gen 3 Mobile 27.834 ms 5 - 16 MB NPU
vision_encoder PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS8550 (Proxy) 41.673 ms 0 - 641 MB NPU
vision_encoder PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS9075 40.702 ms 4 - 7 MB NPU
vision_encoder PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Elite Gen 5 Mobile 17.225 ms 5 - 18 MB NPU
vision_encoder PRECOMPILED_QNN_ONNX mixed Snapdragon® 8 Elite Mobile 21.335 ms 5 - 16 MB NPU
vision_encoder PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS8750 21.335 ms 5 - 16 MB NPU
vision_encoder PRECOMPILED_QNN_ONNX mixed Qualcomm® QCS7181 42.973 ms 516 - 516 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Snapdragon® X2 Elite 17.244 ms 1 - 1 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Snapdragon® X Elite 43.083 ms 1 - 1 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Snapdragon® 8 Gen 3 Mobile 27.694 ms 1 - 8 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Qualcomm® QCS8275 84.555 ms 1 - 9 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Qualcomm® QCS8550 (Proxy) 41.778 ms 1 - 3 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Qualcomm® QCS9075 40.283 ms 1 - 5 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Snapdragon® 8 Elite Gen 5 Mobile 16.463 ms 1 - 9 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Snapdragon® 8 Elite Mobile 21.429 ms 1 - 10 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Qualcomm® SA7255P 84.555 ms 1 - 9 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Qualcomm® QCS8750 21.429 ms 1 - 10 MB NPU
vision_encoder QNN_CONTEXT_BINARY mixed Qualcomm® QCS7181 43.083 ms 1 - 1 MB NPU

License

  • The license for the original implementation of Pi0.5 can be found here.

References

Community

Usage and Limitations

This model may not be used for or in connection with any of the following applications:

  • Accessing essential private and public services and benefits;
  • Administration of justice and democratic processes;
  • Assessing or recognizing the emotional state of a person;
  • Biometric and biometrics-based systems, including categorization of persons based on sensitive characteristics;
  • Education and vocational training;
  • Employment and workers management;
  • Exploitation of the vulnerabilities of persons resulting in harmful behavior;
  • General purpose social scoring;
  • Law enforcement;
  • Management and operation of critical infrastructure;
  • Migration, asylum and border control management;
  • Predictive policing;
  • Real-time remote biometric identification in public spaces;
  • Recommender systems of social media platforms;
  • Scraping of facial images (from the internet or otherwise); and/or
  • Subliminal manipulation
Downloads last month

-

Downloads are not tracked for this model. How to track
Video Preview
Robotics
loading