diff options
Diffstat (limited to 'docs')
-rw-r--r-- | docs/documentation.md | 2 | ||||
-rw-r--r-- | docs/quick_start.md | 32 | ||||
-rw-r--r-- | docs/use_cases/kws.md | 71 | ||||
-rw-r--r-- | docs/use_cases/kws_asr.md | 106 |
4 files changed, 102 insertions, 109 deletions
diff --git a/docs/documentation.md b/docs/documentation.md index f1fab8c..2ec2028 100644 --- a/docs/documentation.md +++ b/docs/documentation.md @@ -211,7 +211,7 @@ What these folders contain: The models used in the use-cases implemented in this project can be downloaded from: [Arm ML-Zoo](https://github.com/ARM-software/ML-zoo). - [Mobilenet V2](https://github.com/ARM-software/ML-zoo/tree/e0aa361b03c738047b9147d1a50e3f2dcb13dbcb/models/image_classification/mobilenet_v2_1.0_224/tflite_int8) -- [DS-CNN](https://github.com/ARM-software/ML-zoo/tree/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b//models/keyword_spotting/ds_cnn_large/tflite_clustered_int8) +- [MicroNet for Keyword Spotting](https://github.com/ARM-software/ML-zoo/tree/9f506fe52b39df545f0e6c5ff9223f671bc5ae00/models/keyword_spotting/micronet_medium/tflite_int8) - [Wav2Letter](https://github.com/ARM-software/ML-zoo/tree/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8) - [MicroNet for Anomaly Detection](https://github.com/ARM-software/ML-zoo/tree/7c32b097f7d94aae2cd0b98a8ed5a3ba81e66b18/models/anomaly_detection/micronet_medium/tflite_int8) - [MicroNet for Visual Wake Word](https://github.com/ARM-software/ML-zoo/raw/7dd3b16bb84007daf88be8648983c07f3eb21140/models/visual_wake_words/micronet_vww4/tflite_int8/vww4_128_128_INT8.tflite) diff --git a/docs/quick_start.md b/docs/quick_start.md index 6522bdd..252b084 100644 --- a/docs/quick_start.md +++ b/docs/quick_start.md @@ -84,11 +84,11 @@ curl -L https://github.com/ARM-software/ML-zoo/raw/e0aa361b03c738047b9147d1a50e3 --output ./resources_downloaded/img_class/ifm0.npy curl -L https://github.com/ARM-software/ML-zoo/raw/e0aa361b03c738047b9147d1a50e3f2dcb13dbcb/models/image_classification/mobilenet_v2_1.0_224/tflite_int8/testing_output/MobilenetV2/Predictions/Reshape_11/0.npy \ --output ./resources_downloaded/img_class/ofm0.npy -curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/keyword_spotting/ds_cnn_large/tflite_clustered_int8/ds_cnn_clustered_int8.tflite \ - --output ./resources_downloaded/kws/ds_cnn_clustered_int8.tflite -curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/keyword_spotting/ds_cnn_large/tflite_clustered_int8/testing_input/input_2/0.npy \ +curl -L https://github.com/ARM-software/ML-zoo/raw/9f506fe52b39df545f0e6c5ff9223f671bc5ae00/models/keyword_spotting/micronet_medium/tflite_int8/kws_micronet_m.tflite \ + --output ./resources_downloaded/kws/kws_micronet_m.tflite +curl -L https://github.com/ARM-software/ML-zoo/raw/9f506fe52b39df545f0e6c5ff9223f671bc5ae00/models/keyword_spotting/micronet_medium/tflite_int8/testing_input/input/0.npy \ --output ./resources_downloaded/kws/ifm0.npy -curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/keyword_spotting/ds_cnn_large/tflite_clustered_int8/testing_output/Identity/0.npy \ +curl -L https://github.com/ARM-software/ML-zoo/raw/9f506fe52b39df545f0e6c5ff9223f671bc5ae00/models/keyword_spotting/micronet_medium/tflite_int8/testing_output/Identity/0.npy \ --output ./resources_downloaded/kws/ofm0.npy curl -L https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/wav2letter_pruned_int8.tflite \ --output ./resources_downloaded/kws_asr/wav2letter_pruned_int8.tflite @@ -96,11 +96,11 @@ curl -L https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59d --output ./resources_downloaded/kws_asr/asr/ifm0.npy curl -L https://github.com/ARM-software/ML-zoo/raw/1a92aa08c0de49a7304e0a7f3f59df6f4fd33ac8/models/speech_recognition/wav2letter/tflite_pruned_int8/testing_output/Identity_int8/0.npy \ --output ./resources_downloaded/kws_asr/asr/ofm0.npy -curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/keyword_spotting/ds_cnn_large/tflite_clustered_int8/ds_cnn_clustered_int8.tflite \ - --output ./resources_downloaded/kws_asr/ds_cnn_clustered_int8.tflite -curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/keyword_spotting/ds_cnn_large/tflite_clustered_int8/testing_input/input_2/0.npy \ +curl -L https://github.com/ARM-software/ML-zoo/raw/9f506fe52b39df545f0e6c5ff9223f671bc5ae00/models/keyword_spotting/micronet_medium/tflite_int8/kws_micronet_m.tflite \ + --output ./resources_downloaded/kws_asr/kws_micronet_m.tflite +curl -L https://github.com/ARM-software/ML-zoo/raw/9f506fe52b39df545f0e6c5ff9223f671bc5ae00/models/keyword_spotting/micronet_medium/tflite_int8/testing_input/input/0.npy \ --output ./resources_downloaded/kws_asr/kws/ifm0.npy -curl -L https://github.com/ARM-software/ML-zoo/raw/68b5fbc77ed28e67b2efc915997ea4477c1d9d5b/models/keyword_spotting/ds_cnn_large/tflite_clustered_int8/testing_output/Identity/0.npy \ +curl -L https://github.com/ARM-software/ML-zoo/raw/9f506fe52b39df545f0e6c5ff9223f671bc5ae00/models/keyword_spotting/micronet_medium/tflite_int8/testing_output/Identity/0.npy \ --output ./resources_downloaded/kws_asr/kws/ofm0.npy curl -L https://github.com/ARM-software/ML-zoo/raw/a061600058097a2785d6f1f7785e5a2d2a142955/models/noise_suppression/RNNoise/tflite_int8/rnnoise_INT8.tflite \ --output ./resources_downloaded/noise_reduction/rnnoise_INT8.tflite @@ -131,22 +131,22 @@ curl -L https://github.com/ARM-software/ML-zoo/raw/7dd3b16bb84007daf88be8648983c curl -L https://github.com/ARM-software/ML-zoo/raw/7dd3b16bb84007daf88be8648983c07f3eb21140/models/visual_wake_words/micronet_vww4/tflite_int8/testing_output/Identity/0.npy \ --output ./resources_downloaded/vww/ofm0.npy -. resources_downloaded/env/bin/activate && vela resources_downloaded/kws/ds_cnn_clustered_int8.tflite \ +. resources_downloaded/env/bin/activate && vela resources_downloaded/kws/kws_micronet_m.tflite \ --accelerator-config=ethos-u55-128 \ --optimise Performance --config scripts/vela/default_vela.ini \ --memory-mode=Shared_Sram \ --system-config=Ethos_U55_High_End_Embedded \ --output-dir=resources_downloaded/kws \ --arena-cache-size=2097152 -mv resources_downloaded/kws/ds_cnn_clustered_int8_vela.tflite resources_downloaded/kws/ds_cnn_clustered_int8_vela_H128.tflite +mv resources_downloaded/kws/kws_micronet_m.tflite resources_downloaded/kws/kws_micronet_m_vela_H128.tflite -. resources_downloaded/env/bin/activate && vela resources_downloaded/kws/ds_cnn_clustered_int8.tflite \ +. resources_downloaded/env/bin/activate && vela resources_downloaded/kws/kws_micronet_m.tflite \ --accelerator-config=ethos-u65-256 \ --optimise Performance --config scripts/vela/default_vela.ini \ --memory-mode=Dedicated_Sram \ --system-config=Ethos_U65_High_End \ --output-dir=resources_downloaded/kws -mv resources_downloaded/kws/ds_cnn_clustered_int8_vela.tflite resources_downloaded/kws/ds_cnn_clustered_int8_vela_Y256.tflite +mv resources_downloaded/kws/kws_micronet_vela.tflite resources_downloaded/kws/kws_micronet_m_vela_Y256.tflite . resources_downloaded/env/bin/activate && vela resources_downloaded/kws_asr/wav2letter_int8.tflite \ --accelerator-config=ethos-u55-128 \ @@ -165,22 +165,22 @@ mv resources_downloaded/kws_asr/wav2letter_int8_vela.tflite resources_downloaded --output-dir=resources_downloaded/kws_asr mv resources_downloaded/kws_asr/wav2letter_int8_vela.tflite resources_downloaded/kws_asr/wav2letter_int8_vela_Y256.tflite -. resources_downloaded/env/bin/activate && vela resources_downloaded/kws_asr/ds_cnn_clustered_int8.tflite \ +. resources_downloaded/env/bin/activate && vela resources_downloaded/kws_asr/kws_micronet_m.tflite \ --accelerator-config=ethos-u55-128 \ --optimise Performance --config scripts/vela/default_vela.ini \ --memory-mode=Shared_Sram \ --system-config=Ethos_U55_High_End_Embedded \ --output-dir=resources_downloaded/kws_asr \ --arena-cache-size=2097152 -mv resources_downloaded/kws_asr/ds_cnn_clustered_int8_vela.tflite resources_downloaded/kws_asr/ds_cnn_clustered_int8_vela_H128.tflite +mv resources_downloaded/kws_asr/kws_micronet_m.tflite_vela.tflite resources_downloaded/kws_asr/kws_micronet_m.tflite_vela_H128.tflite -. resources_downloaded/env/bin/activate && vela resources_downloaded/kws_asr/ds_cnn_clustered_int8.tflite \ +. resources_downloaded/env/bin/activate && vela resources_downloaded/kws_asr/kws_micronet_m.tflite \ --accelerator-config=ethos-u65-256 \ --optimise Performance --config scripts/vela/default_vela.ini \ --memory-mode=Dedicated_Sram \ --system-config=Ethos_U65_High_End \ --output-dir=resources_downloaded/kws_asr -mv resources_downloaded/kws_asr/ds_cnn_clustered_int8_vela.tflite resources_downloaded/kws_asr/ds_cnn_clustered_int8_vela_Y256.tflite +mv resources_downloaded/kws_asr/kws_micronet_m.tflite_vela.tflite resources_downloaded/kws_asr/kws_micronet_m.tflite_vela_Y256.tflite . resources_downloaded/env/bin/activate && vela resources_downloaded/inference_runner/dnn_s_quantized.tflite \ --accelerator-config=ethos-u55-128 \ diff --git a/docs/use_cases/kws.md b/docs/use_cases/kws.md index 84eddd6..0c50fe5 100644 --- a/docs/use_cases/kws.md +++ b/docs/use_cases/kws.md @@ -23,7 +23,7 @@ Use-case code could be found in the following directory: [source/use_case/kws](. ### Preprocessing and feature extraction -The `DS-CNN` keyword spotting model that is used with the Code Samples expects audio data to be preprocessed in a +The `MicroNet` keyword spotting model that is used with the Code Samples expects audio data to be preprocessed in a specific way before performing an inference. Therefore, this section aims to provide an overview of the feature extraction process used. @@ -62,7 +62,7 @@ used. ### Postprocessing After an inference is complete, the word with the highest detected probability is output to console. Providing that the -probability is larger than a threshold value. The default is set to `0.9`. +probability is larger than a threshold value. The default is set to `0.7`. If multiple inferences are performed for an audio clip, then multiple results are output. @@ -107,7 +107,7 @@ In addition to the already specified build option in the main documentation, the this number, then it is padded with zeros. The default value is `16000`. - `kws_MODEL_SCORE_THRESHOLD`: Threshold value that must be applied to the inference results for a label to be deemed - valid. Goes from 0.00 to 1.0. The default is `0.9`. + valid. Goes from 0.00 to 1.0. The default is `0.7`. - `kws_ACTIVATION_BUF_SZ`: The intermediate, or activation, buffer size reserved for the NN model. By default, it is set to 2MiB and is enough for most models @@ -247,7 +247,7 @@ For further information: [Optimize model with Vela compiler](../sections/buildin To run the application with a custom model, you must provide a `labels_<model_name>.txt` file of labels that are associated with the model. Each line of the file must correspond to one of the outputs in your model. Refer to the -provided `ds_cnn_labels.txt` file for an example. +provided `micronet_kws_labels.txt` file for an example. Then, you must set `kws_MODEL_TFLITE_PATH` to the location of the Vela processed model file and `kws_LABELS_TXT_FILE`to the location of the associated labels file. @@ -369,24 +369,24 @@ What the preceding choices do: INFO - Model INPUT tensors: INFO - tensor type is INT8 INFO - tensor occupies 490 bytes with dimensions - INFO - 0: 1 - INFO - 1: 1 - INFO - 2: 49 - INFO - 3: 10 + INFO - 0: 1 + INFO - 1: 49 + INFO - 2: 10 + INFO - 3: 1 INFO - Quant dimension: 0 - INFO - Scale[0] = 1.107164 - INFO - ZeroPoint[0] = 95 + INFO - Scale[0] = 0.201095 + INFO - ZeroPoint[0] = -5 INFO - Model OUTPUT tensors: - INFO - tensor type is INT8 - INFO - tensor occupies 12 bytes with dimensions - INFO - 0: 1 - INFO - 1: 12 + INFO - tensor type is INT8 + INFO - tensor occupies 12 bytes with dimensions + INFO - 0: 1 + INFO - 1: 12 INFO - Quant dimension: 0 - INFO - Scale[0] = 0.003906 - INFO - ZeroPoint[0] = -128 - INFO - Activation buffer (a.k.a tensor arena) size used: 72848 - INFO - Number of operators: 1 - INFO - Operator 0: ethos-u + INFO - Scale[0] = 0.056054 + INFO - ZeroPoint[0] = -54 + INFO - Activation buffer (a.k.a tensor arena) size used: 127068 + INFO - Number of operators: 0 + INFO - Operator 0: ethos-u ``` 5. List audio clips: Prints a list of pair ... indexes. The original filenames are embedded in the application, like so: @@ -405,18 +405,21 @@ Please select the first menu option to execute inference on the first file. The following example illustrates the output for classification: -```logINFO - Running inference on audio clip 0 => down.wav +```log + +INFO - Running inference on audio clip 0 => down.wav INFO - Inference 1/1 INFO - Final results: INFO - Total number of inferences: 1 -INFO - For timestamp: 0.000000 (inference #: 0); label: down, score: 0.996094; threshold: 0.900000 +INFO - For timestamp: 0.000000 (inference #: 0); label: down, score: 0.986182; threshold: 0.700000 INFO - Profile for Inference: -INFO - NPU AXI0_RD_DATA_BEAT_RECEIVED beats: 217385 -INFO - NPU AXI0_WR_DATA_BEAT_WRITTEN beats: 82607 -INFO - NPU AXI1_RD_DATA_BEAT_RECEIVED beats: 59608 -INFO - NPU ACTIVE cycles: 680611 -INFO - NPU IDLE cycles: 561 -INFO - NPU TOTAL cycles: 681172 +INFO - NPU AXI0_RD_DATA_BEAT_RECEIVED beats: 132130 +INFO - NPU AXI0_WR_DATA_BEAT_WRITTEN beats: 48252 +INFO - NPU AXI1_RD_DATA_BEAT_RECEIVED beats: 17544 +INFO - NPU ACTIVE cycles: 413814 +INFO - NPU IDLE cycles: 358 +INFO - NPU TOTAL cycles: 414172 + ``` On most systems running Fast Model, each inference takes under 30 seconds. @@ -425,22 +428,22 @@ The profiling section of the log shows that for this inference: - *Ethos-U* PMU report: - - 681,172 total cycle: The number of NPU cycles. + - 414,172 total cycle: The number of NPU cycles. - - 680,611 active cycles: The number of NPU cycles that were used for computation. + - 413,814 active cycles: The number of NPU cycles that were used for computation. - - 561 idle cycles: The number of cycles for which the NPU was idle. + - 358 idle cycles: The number of cycles for which the NPU was idle. - - 217,385 AXI0 read beats: The number of AXI beats with read transactions from the AXI0 bus. AXI0 is the bus where the + - 132,130 AXI0 read beats: The number of AXI beats with read transactions from the AXI0 bus. AXI0 is the bus where the *Ethos-U* NPU reads and writes to the computation buffers, activation buf, or tensor arenas. - - 82,607 write cycles: The number of AXI beats with write transactions to AXI0 bus. + - 48,252 write cycles: The number of AXI beats with write transactions to AXI0 bus. - - 59,608 AXI1 read beats: The number of AXI beats with read transactions from the AXI1 bus. AXI1 is the bus where the + - 17,544 AXI1 read beats: The number of AXI beats with read transactions from the AXI1 bus. AXI1 is the bus where the *Ethos-U* NPU reads the model. So, read-only. - For FPGA platforms, a CPU cycle count can also be enabled. However, do not use cycle counters for FVP, as the CPU model is not cycle-approximate or cycle-accurate. -> **Note:** The application prints the highest confidence score and the associated label from the `ds_cnn_labels.txt` +> **Note:** The application prints the highest confidence score and the associated label from the `micronet_kws_labels.txt` > file. diff --git a/docs/use_cases/kws_asr.md b/docs/use_cases/kws_asr.md index 42c9d3a..22f1e9d 100644 --- a/docs/use_cases/kws_asr.md +++ b/docs/use_cases/kws_asr.md @@ -44,7 +44,7 @@ By default, the KWS model is run purely on the CPU and **not** on the *Ethos-U55 #### Keyword Spotting Preprocessing -The `DS-CNN` keyword spotting model that is used with the Code Samples expects audio data to be preprocessed in a +The `MicroNet` keyword spotting model that is used with the Code Samples expects audio data to be preprocessed in a specific way before performing an inference. Therefore, this section aims to provide an overview of the feature extraction process used. @@ -455,43 +455,30 @@ What the preceding choices do: 4. Show NN model info: Prints information about the model data type, input, and output, tensor sizes: - ```log + ```log + INFO - Model info: INFO - Model INPUT tensors: INFO - tensor type is INT8 INFO - tensor occupies 490 bytes with dimensions - INFO - 0: 1 - INFO - 1: 1 - INFO - 2: 49 - INFO - 3: 10 + INFO - 0: 1 + INFO - 1: 49 + INFO - 2: 10 + INFO - 3: 1 INFO - Quant dimension: 0 - INFO - Scale[0] = 1.107164 - INFO - ZeroPoint[0] = 95 + INFO - Scale[0] = 0.201095 + INFO - ZeroPoint[0] = -5 INFO - Model OUTPUT tensors: - INFO - tensor type is INT8 - INFO - tensor occupies 12 bytes with dimensions - INFO - 0: 1 - INFO - 1: 12 + INFO - tensor type is INT8 + INFO - tensor occupies 12 bytes with dimensions + INFO - 0: 1 + INFO - 1: 12 INFO - Quant dimension: 0 - INFO - Scale[0] = 0.003906 - INFO - ZeroPoint[0] = -128 - INFO - Activation buffer (a.k.a tensor arena) size used: 123616 - INFO - Number of operators: 16 - INFO - Operator 0: RESHAPE - INFO - Operator 1: CONV_2D - INFO - Operator 2: DEPTHWISE_CONV_2D - INFO - Operator 3: CONV_2D - INFO - Operator 4: DEPTHWISE_CONV_2D - INFO - Operator 5: CONV_2D - INFO - Operator 6: DEPTHWISE_CONV_2D - INFO - Operator 7: CONV_2D - INFO - Operator 8: DEPTHWISE_CONV_2D - INFO - Operator 9: CONV_2D - INFO - Operator 10: DEPTHWISE_CONV_2D - INFO - Operator 11: CONV_2D - INFO - Operator 12: AVERAGE_POOL_2D - INFO - Operator 13: RESHAPE - INFO - Operator 14: FULLY_CONNECTED - INFO - Operator 15: SOFTMAX + INFO - Scale[0] = 0.056054 + INFO - ZeroPoint[0] = -54 + INFO - Activation buffer (a.k.a tensor arena) size used: 127068 + INFO - Number of operators: 1 + INFO - Operator 0: ethos-u + INFO - Model INPUT tensors: INFO - tensor type is INT8 INFO - tensor occupies 11544 bytes with dimensions @@ -511,9 +498,9 @@ What the preceding choices do: INFO - Quant dimension: 0 INFO - Scale[0] = 0.003906 INFO - ZeroPoint[0] = -128 - INFO - Activation buffer (a.k.a tensor arena) size used: 809808 + INFO - Activation buffer (a.k.a tensor arena) size used: 4184332 INFO - Number of operators: 1 - INFO - Operator 0: ethos-u + INFO - Operator 0: ethos-u ``` 5. List audio clips: Prints a list of pair ... indexes. The original filenames are embedded in the application, like so: @@ -534,27 +521,31 @@ INFO - KWS audio data window size 16000 INFO - Running KWS inference on audio clip 0 => yes_no_go_stop.wav INFO - Inference 1/7 INFO - For timestamp: 0.000000 (inference #: 0); threshold: 0.900000 -INFO - label @ 0: yes, score: 0.996094 +INFO - label @ 0: yes, score: 0.997407 INFO - Profile for Inference: -INFO - NPU AXI0_RD_DATA_BEAT_RECEIVED beats: 217385 -INFO - NPU AXI0_WR_DATA_BEAT_WRITTEN beats: 82607 -INFO - NPU AXI1_RD_DATA_BEAT_RECEIVED beats: 59608 -INFO - NPU ACTIVE cycles: 680611 -INFO - NPU IDLE cycles: 561 -INFO - NPU TOTAL cycles: 681172 +INFO - NPU AXI0_RD_DATA_BEAT_RECEIVED beats: 132130 +INFO - NPU AXI0_WR_DATA_BEAT_WRITTEN beats: 48252 +INFO - NPU AXI1_RD_DATA_BEAT_RECEIVED beats: 17544 +INFO - NPU ACTIVE cycles: 413814 +INFO - NPU IDLE cycles: 358 +INFO - NPU TOTAL cycles: 414172 INFO - Keyword spotted INFO - Inference 1/2 INFO - Inference 2/2 -INFO - Result for inf 0: no gow -INFO - Result for inf 1: stoppe -INFO - Final result: no gow stoppe +INFO - Result for inf 0: no go +INFO - Result for inf 1: stop +INFO - Final result: no go stop INFO - Profile for Inference: -INFO - NPU AXI0_RD_DATA_BEAT_RECEIVED beats: 13520864 -INFO - NPU AXI0_WR_DATA_BEAT_WRITTEN beats: 2841970 -INFO - NPU AXI1_RD_DATA_BEAT_RECEIVED beats: 2717670 -INFO - NPU ACTIVE cycles: 28909309 -INFO - NPU IDLE cycles: 863 -INFO - NPU TOTAL cycles: 28910172 +INFO - NPU AXI0_RD_DATA_BEAT_RECEIVED beats: 8895431 +INFO - NPU AXI0_WR_DATA_BEAT_WRITTEN beats: 1890168 +INFO - NPU AXI1_RD_DATA_BEAT_RECEIVED beats: 1740069 +INFO - NPU ACTIVE cycles: 30164330 +INFO - NPU IDLE cycles: 342 +INFO - NPU TOTAL cycles: 30164672 +INFO - Main loop terminated. +INFO - program terminating... +INFO - releasing platform Arm Corstone-300 (SSE-300) + ``` It can take several minutes to complete one inference run. The average time is around 2-3 minutes. @@ -567,22 +558,21 @@ The profiling section of the log shows that for the ASR inference: - *Ethos-U* PMU report: - - 28,910,172 total cycle: The number of NPU cycles. + - 30,164,672 total cycle: The number of NPU cycles. - - 28,909,309 active cycles: The number of NPU cycles that were used for computation. + - 30,164,330 active cycles: The number of NPU cycles that were used for computation. - - 863 idle cycles: The number of cycles for which the NPU was idle. + - 342 idle cycles: The number of cycles for which the NPU was idle. - - 13,520,864 AXI0 read beats: The number of AXI beats with read transactions from the AXI0 bus. AXI0 is the bus where + - 8,895,431 AXI0 read beats: The number of AXI beats with read transactions from the AXI0 bus. AXI0 is the bus where the *Ethos-U* NPU reads and writes to the computation buffers, activation buf, or tensor arenas. - - 2,841,970 AXI0 write beats: The number of AXI beats with write transactions to AXI0 bus. + - 1,890,168 AXI0 write beats: The number of AXI beats with write transactions to AXI0 bus. - - 2,717,670 AXI1 read beats: The number of AXI beats with read transactions from the AXI1 bus. AXI1 is the bus where + - 1,740,069 AXI1 read beats: The number of AXI beats with read transactions from the AXI1 bus. AXI1 is the bus where the *Ethos-U55* NPU reads the model. So, read-only. - For FPGA platforms, a CPU cycle count can also be enabled. However, do not use cycle counters for FVP, as the CPU model is not cycle-approximate or cycle-accurate. -> **Note:** In this example, the KWS inference does *not* use the *Ethos-U55* and only runs on the CPU. Therefore, `0` -> Active NPU cycles are shown. + |