# Set hardware acceleration, multiple models, and their priorities in label image

Source: [https://docs.qualcomm.com/doc/80-88500-3/topic/69_Set_hardware_acceleration__multiple_models__and_their_priorities_in_label_image.html](https://docs.qualcomm.com/doc/80-88500-3/topic/69_Set_hardware_acceleration__multiple_models__and_their_priorities_in_label_image.html)

1. To model, use `-m <list of model names comma separated m1,
  m2,...>`
2. To run multiple iterations, use `-c <# of iterations>`
3. For the Android neural networks API (NNAPI), use `-a 1` (model executes on the
        CPU).
4. To label the file, use `-l <label file, for multiple models, use it as
          separated comma l1,l2,... >`
5. To test image, use `-i <images filename or full path to image>`.
6. To use multithreaded CPU arch, use `-t <number of threads>`
7. For the Hexagon delegate, use `-j 1`
8. To set an inference rate limit, use `-J `&lt;inference rate for each model mapped
        one to one with models, that is, 30,15 means 30 inference per second for the model
          `m1` and 15 inference per second for `m2`&gt;.
9. To run a model at different delegate, use `-A`&lt;comma separated mapped one to
        one with model &gt;.
    - 0 → default CPU
    - 1 → NNAPI delegate
10. To check `Op` runs on a specific delegate or CPU, use profiling `-p
          1`.
11. To get full verbose information about the model, use `-v 1`.

**Parent Topic:** [Test TensorFlow Lite using label image](https://docs.qualcomm.com/doc/80-88500-3/topic/68_Test_TFLite_using_label_image.html)

Last Published: Sep 26, 2023

[Previous Topic
Test TensorFlow Lite using label image](https://docs.qualcomm.com/bundle/publicresource/80-88500-3/topics/68_Test_TFLite_using_label_image.md) [Next Topic
Test TensorFlow Lite using benchmark test application](https://docs.qualcomm.com/bundle/publicresource/80-88500-3/topics/70_Test_TFLite_using_benchmark_test_application_.md)