Other
PyTorch
android
qaihm-bot commited on
Commit
cdd70b5
·
verified ·
1 Parent(s): 418d486

See https://github.com/qualcomm/ai-hub-models/releases/v0.48.0 for changelog.

Files changed (1) hide show
  1. README.md +77 -79
README.md CHANGED
@@ -14,7 +14,7 @@ pipeline_tag: other
14
  Geometry-guided Kernel Transformer is a machine learning model for generating a birds eye view represenation from the sensors(cameras) mounted on a vehicle.
15
 
16
  This is based on the implementation of GKT found [here](https://github.com/hustvl/GKT/ https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf).
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/gkt) 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
 
@@ -27,58 +27,56 @@ Below are pre-exported model assets ready for deployment.
27
 
28
  | Runtime | Precision | Chipset | SDK Versions | Download |
29
  |---|---|---|---|---|
30
- | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-float-qualcomm_snapdragon_x_elite.zip)
31
- | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-float-qualcomm_snapdragon_8gen3.zip)
32
- | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-float-qualcomm_qcs8550_proxy.zip)
33
- | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-float-qualcomm_snapdragon_8_elite_for_galaxy.zip)
34
- | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-float-qualcomm_snapdragon_8_elite_gen5.zip)
35
- | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-float-qualcomm_snapdragon_x2_elite.zip)
36
- | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-float-qualcomm_qcs9075.zip)
37
- | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_snapdragon_x_elite.zip)
38
- | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_snapdragon_8gen3.zip)
39
- | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_qcs8550_proxy.zip)
40
- | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_snapdragon_8_elite_for_galaxy.zip)
41
- | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_snapdragon_8_elite_gen5.zip)
42
- | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X2 Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_snapdragon_x2_elite.zip)
43
- | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® QCS9075 | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_qcs9075.zip)
44
- | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_snapdragon_x_elite.zip)
45
- | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_snapdragon_8gen3.zip)
46
- | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_qcs8275_proxy.zip)
47
- | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_qcs8550_proxy.zip)
48
- | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_sa8775p.zip)
49
- | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_snapdragon_8_elite_for_galaxy.zip)
50
- | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_snapdragon_8_elite_gen5.zip)
51
- | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_snapdragon_x2_elite.zip)
52
- | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_sa7255p.zip)
53
- | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_sa8295p.zip)
54
- | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_qcs9075.zip)
55
- | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-float-qualcomm_qcs8450_proxy.zip)
56
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_snapdragon_x_elite.zip)
57
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_snapdragon_8gen3.zip)
58
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8275 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_qcs8275_proxy.zip)
59
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_qcs8550_proxy.zip)
60
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8775P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_sa8775p.zip)
61
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_snapdragon_8_elite_for_galaxy.zip)
62
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_snapdragon_8_elite_gen5.zip)
63
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X2 Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_snapdragon_x2_elite.zip)
64
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA7255P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_sa7255p.zip)
65
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8295P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_sa8295p.zip)
66
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS9075 | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_qcs9075.zip)
67
- | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8450 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.47.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_qcs8450_proxy.zip)
68
 
69
  For more device-specific assets and performance metrics, visit **[GKT on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/gkt)**.
70
 
71
 
72
  ### Option 2: Export with Custom Configurations
73
 
74
- Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/gkt) Python library to compile and export the model with your own:
75
  - Custom weights (e.g., fine-tuned checkpoints)
76
  - Custom input shapes
77
  - Target device and runtime configurations
78
 
79
  This option is ideal if you need to customize the model beyond the default configuration provided here.
80
 
81
- See our repository for [GKT on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/gkt) for usage instructions.
82
 
83
  ## Model Details
84
 
@@ -93,44 +91,44 @@ See our repository for [GKT on GitHub](https://github.com/quic/ai-hub-models/blo
93
  ## Performance Summary
94
  | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
95
  |---|---|---|---|---|---|---
96
- | GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 106.4 ms | 7 - 7 MB | NPU
97
- | GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 79.454 ms | 8 - 15 MB | NPU
98
- | GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 110.344 ms | 8 - 11 MB | NPU
99
- | GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 109.915 ms | 7 - 10 MB | NPU
100
- | GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 71.538 ms | 1 - 8 MB | NPU
101
- | GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 56.714 ms | 8 - 18 MB | NPU
102
- | GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 61.693 ms | 6 - 6 MB | NPU
103
- | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X Elite | 128.247 ms | 6 - 6 MB | NPU
104
- | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | 95.849 ms | 4 - 12 MB | NPU
105
- | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | 129.961 ms | 1 - 10 MB | NPU
106
- | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® QCS9075 | 130.31 ms | 4 - 6 MB | NPU
107
- | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | 82.212 ms | 1 - 8 MB | NPU
108
- | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | 70.214 ms | 4 - 14 MB | NPU
109
- | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X2 Elite | 79.574 ms | 7 - 7 MB | NPU
110
- | GKT | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 104.991 ms | 7 - 7 MB | NPU
111
- | GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 77.534 ms | 8 - 15 MB | NPU
112
- | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 183.072 ms | 1 - 9 MB | NPU
113
- | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 109.361 ms | 8 - 9 MB | NPU
114
- | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 110.683 ms | 0 - 10 MB | NPU
115
- | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 110.066 ms | 7 - 17 MB | NPU
116
- | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 208.536 ms | 8 - 17 MB | NPU
117
- | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 183.072 ms | 1 - 9 MB | NPU
118
- | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 140.472 ms | 0 - 6 MB | NPU
119
- | GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 71.856 ms | 0 - 13 MB | NPU
120
- | GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 57.236 ms | 7 - 17 MB | NPU
121
- | GKT | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 61.925 ms | 7 - 7 MB | NPU
122
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X Elite | 128.364 ms | 4 - 4 MB | NPU
123
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | 95.11 ms | 6 - 14 MB | NPU
124
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8275 (Proxy) | 191.142 ms | 0 - 9 MB | NPU
125
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | 129.628 ms | 4 - 6 MB | NPU
126
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8775P | 129.449 ms | 0 - 10 MB | NPU
127
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS9075 | 129.235 ms | 3 - 8 MB | NPU
128
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8450 (Proxy) | 143.045 ms | 4 - 13 MB | NPU
129
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA7255P | 191.142 ms | 0 - 9 MB | NPU
130
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8295P | 152.623 ms | 0 - 6 MB | NPU
131
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | 82.501 ms | 4 - 17 MB | NPU
132
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | 69.781 ms | 4 - 13 MB | NPU
133
- | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X2 Elite | 79.554 ms | 4 - 4 MB | NPU
134
 
135
  ## License
136
  * The license for the original implementation of GKT can be found
 
14
  Geometry-guided Kernel Transformer is a machine learning model for generating a birds eye view represenation from the sensors(cameras) mounted on a vehicle.
15
 
16
  This is based on the implementation of GKT found [here](https://github.com/hustvl/GKT/ https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf).
17
+ 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/main/qai_hub_models/models/gkt) 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
 
 
27
 
28
  | Runtime | Precision | Chipset | SDK Versions | Download |
29
  |---|---|---|---|---|
30
+ | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-float-qualcomm_snapdragon_x2_elite.zip)
31
+ | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-float-qualcomm_snapdragon_x_elite.zip)
32
+ | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-float-qualcomm_snapdragon_8gen3.zip)
33
+ | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-float-qualcomm_qcs8550_proxy.zip)
34
+ | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-float-qualcomm_snapdragon_8_elite_for_galaxy.zip)
35
+ | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-float-qualcomm_snapdragon_8_elite_gen5.zip)
36
+ | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-float-qualcomm_qcs9075.zip)
37
+ | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X2 Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_snapdragon_x2_elite.zip)
38
+ | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X Elite | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_snapdragon_x_elite.zip)
39
+ | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_snapdragon_8gen3.zip)
40
+ | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_qcs8550_proxy.zip)
41
+ | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_snapdragon_8_elite_for_galaxy.zip)
42
+ | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_snapdragon_8_elite_gen5.zip)
43
+ | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® QCS9075 | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-precompiled_qnn_onnx-w8a16_mixed_fp16-qualcomm_qcs9075.zip)
44
+ | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_snapdragon_x2_elite.zip)
45
+ | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_snapdragon_x_elite.zip)
46
+ | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_snapdragon_8gen3.zip)
47
+ | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_qcs8550_proxy.zip)
48
+ | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_sa8775p.zip)
49
+ | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_snapdragon_8_elite_for_galaxy.zip)
50
+ | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_snapdragon_8_elite_gen5.zip)
51
+ | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_sa7255p.zip)
52
+ | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_sa8295p.zip)
53
+ | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_qcs9075.zip)
54
+ | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-float-qualcomm_qcs8450_proxy.zip)
55
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X2 Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_snapdragon_x2_elite.zip)
56
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X Elite | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_snapdragon_x_elite.zip)
57
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_snapdragon_8gen3.zip)
58
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_qcs8550_proxy.zip)
59
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8775P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_sa8775p.zip)
60
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_snapdragon_8_elite_for_galaxy.zip)
61
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_snapdragon_8_elite_gen5.zip)
62
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA7255P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_sa7255p.zip)
63
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8295P | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_sa8295p.zip)
64
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS9075 | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_qcs9075.zip)
65
+ | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8450 (Proxy) | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/gkt/releases/v0.48.0/gkt-qnn_context_binary-w8a16_mixed_fp16-qualcomm_qcs8450_proxy.zip)
 
 
66
 
67
  For more device-specific assets and performance metrics, visit **[GKT on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/gkt)**.
68
 
69
 
70
  ### Option 2: Export with Custom Configurations
71
 
72
+ Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/gkt) Python library to compile and export the model with your own:
73
  - Custom weights (e.g., fine-tuned checkpoints)
74
  - Custom input shapes
75
  - Target device and runtime configurations
76
 
77
  This option is ideal if you need to customize the model beyond the default configuration provided here.
78
 
79
+ See our repository for [GKT on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/gkt) for usage instructions.
80
 
81
  ## Model Details
82
 
 
91
  ## Performance Summary
92
  | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
93
  |---|---|---|---|---|---|---
94
+ | GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 61.461 ms | 6 - 6 MB | NPU
95
+ | GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 106.483 ms | 7 - 7 MB | NPU
96
+ | GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 79.58 ms | 8 - 15 MB | NPU
97
+ | GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 110.077 ms | 8 - 11 MB | NPU
98
+ | GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 110.099 ms | 7 - 10 MB | NPU
99
+ | GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 71.576 ms | 1 - 11 MB | NPU
100
+ | GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 57.391 ms | 8 - 18 MB | NPU
101
+ | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X2 Elite | 84.266 ms | 7 - 7 MB | NPU
102
+ | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X Elite | 136.395 ms | 6 - 6 MB | NPU
103
+ | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | 101.083 ms | 4 - 11 MB | NPU
104
+ | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | 133.736 ms | 0 - 10 MB | NPU
105
+ | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® QCS9075 | 137.16 ms | 4 - 6 MB | NPU
106
+ | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | 86.587 ms | 0 - 11 MB | NPU
107
+ | GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | 74.796 ms | 4 - 14 MB | NPU
108
+ | GKT | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 61.879 ms | 7 - 7 MB | NPU
109
+ | GKT | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 105.328 ms | 7 - 7 MB | NPU
110
+ | GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 78.845 ms | 8 - 16 MB | NPU
111
+ | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 183.379 ms | 0 - 9 MB | NPU
112
+ | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 108.194 ms | 8 - 9 MB | NPU
113
+ | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 110.72 ms | 2 - 7 MB | NPU
114
+ | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 109.717 ms | 7 - 17 MB | NPU
115
+ | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 205.284 ms | 8 - 18 MB | NPU
116
+ | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 183.379 ms | 0 - 9 MB | NPU
117
+ | GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 140.31 ms | 0 - 6 MB | NPU
118
+ | GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 71.692 ms | 0 - 13 MB | NPU
119
+ | GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 57.242 ms | 7 - 17 MB | NPU
120
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X2 Elite | 84.347 ms | 4 - 4 MB | NPU
121
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X Elite | 136.085 ms | 4 - 4 MB | NPU
122
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | 99.675 ms | 4 - 11 MB | NPU
123
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8275 (Proxy) | 209.065 ms | 0 - 8 MB | NPU
124
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | 135.674 ms | 5 - 6 MB | NPU
125
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8775P | 137.895 ms | 1 - 10 MB | NPU
126
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS9075 | 137.624 ms | 0 - 6 MB | NPU
127
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8450 (Proxy) | 149.718 ms | 4 - 14 MB | NPU
128
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA7255P | 209.065 ms | 0 - 8 MB | NPU
129
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8295P | 164.207 ms | 0 - 6 MB | NPU
130
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | 86.837 ms | 4 - 17 MB | NPU
131
+ | GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | 74.981 ms | 4 - 13 MB | NPU
132
 
133
  ## License
134
  * The license for the original implementation of GKT can be found