LLM Inference on Phi4 Mini Google Pixel 9 Pro XL 3.8B (test)
119.64Prefill Min Throughput (tokens/sec)ET Vulkan
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ET VulkanHardware=GPU, GS=1282026.05 | 119.64 | 120.07 | 9.58 | 9.64 | 2,088 | |
| llama.cppHardware=CPU, GS=322026.05 | 57.2 | 58.6 | 10 | 10.3 | 2,216 | |
| ET XNNPACKHardware=CPU, GS=1282026.05 | 55.65 | 63.27 | 12 | 12.51 | 2,201 | |
| ET VulkanHardware=GPU, GS=322026.05 | 49.67 | 50.22 | 8.67 | 8.86 | 2,829 | |
| ET XNNPACKHardware=CPU, GS=322026.05 | 39.32 | 51.94 | 7.93 | 11.25 | 2,428 | |
| ONNXHardware=CPU, GS=1282026.05 | 38.28 | 54.64 | 6.87 | 7.76 | 1,994 | |
| ONNXHardware=CPU, GS=322026.05 | 32.75 | 34.89 | 6.06 | 6.48 | 2,337 | |
| llama.cppHardware=GPU, GS=322026.05 | 11.5 | 11.7 | 6.7 | 6.8 | 2,216 |