# Test TensorFlow Lite using benchmark test application 

Source: [https://docs.qualcomm.com/doc/80-88500-3/topic/70_Test_TFLite_using_benchmark_test_application_.html](https://docs.qualcomm.com/doc/80-88500-3/topic/70_Test_TFLite_using_benchmark_test_application_.html)

To test TensorFlow Lite using the benchmark test application, use the following command:

    benchmark_model --graph=mv1_1.0_224_quant.tflite --enable_op_profiling=true --use_nnapi=true --num_threads=4Copy to clipboard

| Parameters | Description |
| --- | --- |
| `Graph` | Model to test |
| `use_nnapi` | Enables NNAPI acceleration |
| `num_threads: int (default=1)` | The number of threads to use to run the TensorFlow Lite interpreter. |
| `warmup_runs: int (default=1)` | The number of warm-up runs to do before starting the benchmark. |
| `num_runs: int (default=50)` | The number of runs. Increase to reduce variance. |
| `run_delay: float (default=-1.0)` | The delay in seconds between subsequent benchmark runs. For nonpositive values,<br>              use no delay. |
| `enable_op_profiling: bool (default=false)` | Enables per-operator profiling measurement. |
| `enable_platform_tracing: bool (default=false)` | Enables platform-wide tracing. Must be combined with<br>                `enable_op_profiling`. The platform-wide tracing may not work if the<br>              tool is run as a command line native binary. For example, on Android, the ATrace-based<br>              tracing works only when the tool is launched as an APK. |
| `profiling_output_csv_file: str (default="")` | File path to export profile data to as CSV. The results are printed to stdout if<br>              the option is not set. This parameter requires `enable_op_profiling` to<br>              be true and the path to include the name of the output CSV, otherwise the results are<br>              printed to stdout. |
| max_delegated_partitions: int<br>                    (default=0)Copy to clipboard<br><br><br><br>    min_nodes_per_partition:int (default=0Copy to clipboard | Common parameters |

Table : NNAPI delegate

| Parameter | Description |
| --- | --- |
| `use_nnapi: bool (default=false)` | This parameter enables NNAPI acceleration in the benchmark application. |
| `nnapi_accelerator_name: str (default="")` | Specify only limited devices for running a model. |
| `disable_nnapi_cpu: bool (default=false)` | Disable NNAPI CPU, so ops falls back to TensorFlow Lite CPU |
| `nnapi_allow_fp16: bool (default=false)` | Allow FP16 based models on NNAPI runtime |

**Parent Topic:** [Configure AI/ML modules](https://docs.qualcomm.com/doc/80-88500-3/topic/66_Configure_AI_ML_modules.html)

Last Published: Sep 26, 2023

[Previous Topic
Set hardware acceleration, multiple models, and their priorities in label image](https://docs.qualcomm.com/bundle/publicresource/80-88500-3/topics/69_Set_hardware_acceleration__multiple_models__and_their_priorities_in_label_image.md) [Next Topic
Install GStreamer plug-in](https://docs.qualcomm.com/bundle/publicresource/80-88500-3/topics/72_Install_qtimlsnpe_GStreamer_plug_in_.md)