| tag | 582993072a3d0d89655c2a2513511bd85f85d38d | |
|---|---|---|
| tagger | Tim Hall <tim.hall@arm.com> | Wed Apr 29 16:38:43 2020 +0100 |
| object | 79d07d2cbf1c5013ab40bb46a6ccd4c569966536 |
Beta release.
| commit | 79d07d2cbf1c5013ab40bb46a6ccd4c569966536 | [log] [tgz] |
|---|---|---|
| author | Tim Hall <tim.hall@arm.com> | Mon Apr 27 18:20:16 2020 +0100 |
| committer | Tim Hall <tim.hall@arm.com> | Wed Apr 29 13:00:51 2020 +0100 |
| tree | 410d17239b417be5593b3e6800001b797f8d3f98 | |
| parent | 47bca71566d4d10e48f5a4d66e1130b8bf60700d [diff] |
Add Vela codebase - Added modules ethosu.vela and ethosu.mlw_codec. - Added README and various configuration files. Change-Id: I3690f8c8f5966306ecddaeb2793c30ca9c6e2eee
This tool is used to compile a TensorFlow Lite for Microcontrollers neural network model into an optimised version that can run on an embedded system containing an Ethos-U55 NPU.
The optimised model will contain TensorFlow Lite Custom operators for those parts of the model that can be accelerated by the Ethos-U55. Parts of the model that cannot be accelerated are left unchanged and will instead run on the Cortex-M series CPU using an appropriate kernel (such as the Arm optimised CMSIS-NN kernels).
After compilation the optimised model can only be run on an Ethos-U55 NPU embedded system.
The tool will also generate performance estimates (EXPERIMENTAL) for the compiled model.
Vela runs on the Linux operating system.
The following should be installed prior to the installation of Vela:
Before running, the Vela package must be installed along with all its dependencies. To do this, first change to the directory that contains this README.md file. Then use the command:
pip3 install -U setuptools>=40.1.0 pip3 install .
Or, if you use the pipenv virtual environment tool:
pipenv install .
Vela is run with an input .tflite file passed on the command line. This file contains the neural network to be compiled. The tool then outputs an optimised version with a _vela.tflite file prefix, along with the performance estimate (EXPERIMENTAL) CSV files, all to the output directory.
If you use the pipenv virtual environment tool then first start by spawning a shell in the virtual environment.:
pipenv shell
After which running Vela is the same regardless of whether you are in a virtual environment or not.
Example usage:
my_model.tflite. The optimised version will be output to ./output/my_network_vela.tflite.vela my_model.tflite
/path/to/my_model.tflite and specify the output to go in the directory ./results_dir/.vela --output-dir ./results_dir /path/to/my_model.tflite
vela --help
MySysConfig settings that are described in the sys_cfg_vela.ini system configuration file. More details can be found in the next section.vela --config sys_cfg_vela.ini --system-config MySysConfig my_model.tflite
This is used to describe various properties of the embedded system that the network will run in.
Example of a Vela system configuration file.
; File: sys_cfg_vela.ini ; The file contains two parts; a system config part and a CPU operator ; performance part. ; System config ; Specifies properties such as the core clock speed, the size and speed of the ; four potential memory areas, and for various types of data which memory area ; is used to store them. The cpu property is used to link with the CPU operator ; performance. ; The four potential memory areas are: Sram, Dram, OnChipFlash, OffChipFlash. [SysConfig.MySysConfig] npu_freq=500e6 cpu=MyCpu Sram_clock_scale=1 Sram_port_width=64 Dram_clock_scale=1 Dram_port_width=64 OnChipFlash_clock_scale=1 OnChipFlash_port_width=64 OffChipFlash_clock_scale=0.25 OffChipFlash_port_width=32 permanent_storage_mem_area=OffChipFlash feature_map_storage_mem_area=Sram fast_storage_mem_area=Sram ; CPU operator performance ; Specifies properties that are used by a linear model to estimate the ; performance for any operations that will be run on the CPU (such as those not ; supported by the NPU). Setting the intercept and slope to 0 will result in ; the operator being excluded from the performance estimation. This is the same ; as not specifying the operator. If an explicit cpu is specified rather than ; using the default then the cpu name must match the cpu specified in the ; SysConfig.<system config name> section. [CpuPerformance.MyCpuOperator] default.intercept=0.0 default.slope=1.0 MyCpu.intercept=0.0 MyCpu.slope=1.0
Contributions are accepted under Apache License 2.0. Only submit contributions where you have authored all of the code.
Vela is licensed under Apache License 2.0