Profile Job Results
Jobs
j5wyr1x6g
Results Ready
Name
qwen3_vl_4b_instruct_vision_encoder_qwen3_vl_4b_instruct
Target Device
- Dragonwing IQ-9075 EVK
- Qualcomm Linux 1.9
- Qualcomm® Dragonwing™ IQ-9075 | QCS9075
Creator
ai-hub-support@qti.qualcomm.com
Input Specs
pixel_values: uint16[1024, 1536]position_ids_cos: uint16[1024, 32]position_ids_sin: uint16[1024, 32]full_attention_mask: uint16[1, 1024, 1024]window_attention_mask: float16[1, 1024, 1024]Completion Time
10/17/2026, 10:05:55 AM
Versions
- QAIRT: v2.50.0.260828221209
- QNN Backend API: 5.50.0
- QNN Core API: 2.39.0
- Qualcomm Linux: 1.9-ver.1.0
- AI Hub: aihub-2026.10.10.0
Estimated Inference Time
149 ms
Estimated Peak Memory Usage
7 ‑ 21 MB
Compute Units
NPU
1477
| Stage | Time | Memory |
|---|---|---|
First App Load | 463 ms | 2‑3 MB |
Subsequent App Load | 455 ms | 2‑3 MB |
Inference | 149 ms | 7‑21 MB |
| QNN | Value |
|---|---|
| context_options.htp_options.performance_mode | BURST |
| default_graph_options.htp_options.optimizations[0].type | FINALIZE_OPTIMIZATION_FLAG |
| default_graph_options.htp_options.optimizations[0].value | 3.0 |
| default_graph_options.htp_options.precision | FLOAT16 |
| default_graph_options.htp_options.vtcm_size | 0 |
Sign up to run this model on a hosted Qualcomm® device!
Run on device