Profile Job Results
Jobs
jo5m90xwg
Results Ready
Name
whisper_tiny_en_WhisperDecoder
Target Device
- QCS8550 (Proxy)
- Android 12
Creator
ai-hub-support@qti.qualcomm.com
Target Model
Input Specs
x: int32[1, 1]index: int32[1, 1]k_cache_cross: float32[4, 6, 64, 1500]v_cache_cross: float32[4, 6, 1500, 64]k_cache_self: float32[4, 6, 64, 224]v_cache_self: float32[4, 6, 224, 64]Completion Time
8/11/2024, 5:37:50 AM
Versions
- TensorFlow Lite: 2.16.1
- QNN TfLite Delegate: v2.24.0.240626131148_96320
- Android: 14 (UP1A.231005.007)
- AI Hub: aihub-2024.08.01.0
Estimated Inference Time
10.3 ms
Estimated Peak Memory Usage
21 - 28 MB
Compute Units
NPU
552
CPU
4
GPU
1
| Stage | Time | Memory |
|---|---|---|
First App Load | 1.57 s | 344-345 MB |
Subsequent App Load | 376 ms | 100-159 MB |
Inference | 10.3 ms | 21-28 MB |
| TensorFlow Lite | Value |
|---|---|
| number_of_threads | 4 |
| QNN Delegate | Value |
|---|---|
| backend_type | kHtpBackend |
| log_level | kLogLevelWarn |
| htp_options.performance_mode | kHtpBurst |
| htp_options.precision | kHtpFp16 |
| htp_options.useConvHmx | true |
| GPUv2 Delegate | Value |
|---|---|
| inference_preference | TFLITE_GPU_INFERENCE_PREFERENCE_SUSTAINED_SPEED |
| inference_priority1 | TFLITE_GPU_INFERENCE_PRIORITY_MIN_LATENCY |
| inference_priority2 | TFLITE_GPU_INFERENCE_PRIORITY_MIN_MEMORY_USAGE |
| inference_priority3 | TFLITE_GPU_INFERENCE_PRIORITY_MAX_PRECISION |
| XNNPACK Delegate | Value |
|---|
Sign up to run this model on a hosted Qualcomm® device!
Run on device