Distibuted ML Accelerator simulator
Plans to support
- Distributed execution of DNN layers
- Heterogenous cores
- Multitenant execution
Getting started is simple! Krittika is completely written in python and uses scalesim-v2 in backend.
You can clone the SCALE-Sim(v2) repository using the following command (ssh)
$ git clone git@github.com:scalesim-project/scale-sim-v2.git
Alternative, you can also clone using https
$ git clone https://github.com/scalesim-project/scale-sim-v2.git
If you are running for the first time and do not have all the dependencies installed, please install them first using the following command
$ pip3 install -r <path_to_scale_sim_repot>/requirements.txt
After cloning, install scalesim from path using the following command. This version will automatically reflect any code changes that you make.
$ pip3 install -e <path_to_scale_sim_repo_root>
Krittika can be run by using the krittika-sim.py script from the repository and providing the paths to the architecture configuration file (refer configs/krittika.cfg for example) and the topology descriptor csv file (refer scalesim repo for examples).
$ python3 krittika-sim.py -c <path_to_config_file> -t <path_to_topology_file>
Additional optional parameters
- -p <path_to_partition_file> (Refer partition_manager.py for more info)
- -o <path_to_the_log_dump_directory>
- --verbose <True/False> (Flag to change the verbosity of the run)
- --savetrace <True/False> (Flag to indicate if the traces should be saved)
The topology file is a CSV file which decribes the layers of the workload topology. The layers are typically described as convolution/GEMM/activation layer parameters as shown in the example below
Conv and GEMM layers follow scalesim topology structure link
The only difference is that there is no comma at the end of each layer.
Support for RELU activation is also added and can be used just as shown in the image.