Hugging Face's logo

qualcomm
/
ACT

Robotics
PyTorch
android
Model card Files
xet
Community
qaihm-bot commited on
Commit
213d473
·
verified ·
1 Parent(s): b8fff32

v0.47.0

Browse files

See https://github.com/quic/ai-hub-models/releases/v0.47.0 for changelog.

Files changed (2) hide show
  1. LICENSE +1 -0
  2. README.md +101 -0
LICENSE ADDED
@@ -0,0 +1 @@
 
 
1
+ The license of the original trained model can be found at https://github.com/tonyzhaozh/act/blob/main/LICENSE.
README.md ADDED
@@ -0,0 +1,101 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: pytorch
3
+ license: other
4
+ tags:
5
+ - android
6
+ pipeline_tag: robotics
7
+
8
+ ---
9
+
10
+ ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/act/web-assets/model_demo.png)
11
+
12
+ # ACT: Optimized for Qualcomm Devices
13
+
14
+ ACT (Action Chunking with Transformers) is a robotic policy model that is trained to predict the next chunk of actions that the robotic hand is expected to perform.
15
+
16
+ This is based on the implementation of ACT found [here](https://github.com/tonyzhaozh/act).
17
+ This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/act) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
18
+
19
+ 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.
20
+
21
+ ## Getting Started
22
+ There are two ways to deploy this model on your device:
23
+
24
+ ### Option 1: Download Pre-Exported Models
25
+
26
+ Below are pre-exported model assets ready for deployment.
27
+
28
+ | Runtime | Precision | Chipset | SDK Versions | Download |
29
+ |---|---|---|---|---|
30
+ | ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/act/releases/v0.47.0/act-onnx-float.zip)
31
+ | QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/act/releases/v0.47.0/act-qnn_dlc-float.zip)
32
+ | TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/act/releases/v0.47.0/act-tflite-float.zip)
33
+
34
+ For more device-specific assets and performance metrics, visit **[ACT on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/act)**.
35
+
36
+
37
+ ### Option 2: Export with Custom Configurations
38
+
39
+ Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/act) Python library to compile and export the model with your own:
40
+ - Custom weights (e.g., fine-tuned checkpoints)
41
+ - Custom input shapes
42
+ - Target device and runtime configurations
43
+
44
+ This option is ideal if you need to customize the model beyond the default configuration provided here.
45
+
46
+ See our repository for [ACT on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/act) for usage instructions.
47
+
48
+ ## Model Details
49
+
50
+ **Model Type:** Model_use_case.robotics
51
+
52
+ **Model Stats:**
53
+ - Model checkpoint: act
54
+ - Input resolution: 480x640
55
+ - Number of parameters: 83.92M
56
+ - Model size (float): 255M
57
+
58
+ ## Performance Summary
59
+ | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
60
+ |---|---|---|---|---|---|---
61
+ | ACT | ONNX | float | Snapdragon® X Elite | 11.898 ms | 62 - 62 MB | NPU
62
+ | ACT | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 8.113 ms | 1 - 376 MB | NPU
63
+ | ACT | ONNX | float | Qualcomm® QCS8550 (Proxy) | 11.237 ms | 0 - 81 MB | NPU
64
+ | ACT | ONNX | float | Qualcomm® QCS9075 | 19.025 ms | 4 - 10 MB | NPU
65
+ | ACT | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 6.563 ms | 1 - 329 MB | NPU
66
+ | ACT | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.513 ms | 4 - 333 MB | NPU
67
+ | ACT | ONNX | float | Snapdragon® X2 Elite | 6.147 ms | 63 - 63 MB | NPU
68
+ | ACT | QNN_DLC | float | Snapdragon® X Elite | 8.903 ms | 4 - 4 MB | NPU
69
+ | ACT | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 5.95 ms | 3 - 348 MB | NPU
70
+ | ACT | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 43.949 ms | 0 - 308 MB | NPU
71
+ | ACT | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 8.29 ms | 4 - 6 MB | NPU
72
+ | ACT | QNN_DLC | float | Qualcomm® SA8775P | 13.34 ms | 1 - 294 MB | NPU
73
+ | ACT | QNN_DLC | float | Qualcomm® QCS9075 | 16.057 ms | 4 - 9 MB | NPU
74
+ | ACT | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 16.303 ms | 4 - 264 MB | NPU
75
+ | ACT | QNN_DLC | float | Qualcomm® SA7255P | 43.949 ms | 0 - 308 MB | NPU
76
+ | ACT | QNN_DLC | float | Qualcomm® SA8295P | 15.529 ms | 0 - 231 MB | NPU
77
+ | ACT | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.798 ms | 0 - 295 MB | NPU
78
+ | ACT | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.741 ms | 4 - 321 MB | NPU
79
+ | ACT | QNN_DLC | float | Snapdragon® X2 Elite | 4.863 ms | 4 - 4 MB | NPU
80
+ | ACT | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 5.977 ms | 0 - 362 MB | NPU
81
+ | ACT | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 44.141 ms | 0 - 319 MB | NPU
82
+ | ACT | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 8.231 ms | 0 - 3 MB | NPU
83
+ | ACT | TFLITE | float | Qualcomm® SA8775P | 13.453 ms | 0 - 304 MB | NPU
84
+ | ACT | TFLITE | float | Qualcomm® QCS9075 | 16.21 ms | 0 - 71 MB | NPU
85
+ | ACT | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 16.408 ms | 0 - 274 MB | NPU
86
+ | ACT | TFLITE | float | Qualcomm® SA7255P | 44.141 ms | 0 - 319 MB | NPU
87
+ | ACT | TFLITE | float | Qualcomm® SA8295P | 15.698 ms | 0 - 235 MB | NPU
88
+ | ACT | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 4.838 ms | 0 - 307 MB | NPU
89
+ | ACT | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.8 ms | 0 - 327 MB | NPU
90
+
91
+ ## License
92
+ * The license for the original implementation of ACT can be found
93
+ [here](https://github.com/tonyzhaozh/act/blob/main/LICENSE).
94
+
95
+ ## References
96
+ * [Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware](https://arxiv.org/abs/2304.13705)
97
+ * [Source Model Implementation](https://github.com/tonyzhaozh/act)
98
+
99
+ ## Community
100
+ * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
101
+ * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).