Skip to content

Latest commit

 

History

History
68 lines (55 loc) · 1.85 KB

README.md

File metadata and controls

68 lines (55 loc) · 1.85 KB
Raw

Enviornment Installation

  • Install Anaconda3
  • For setting up the enviornment run :
    • for instantiating with localstack: bash scripts/create_experiment_env_linux.sh srtml-exp localstack_init
    • for instantiating without localstack: bash scripts/create_experiment_env_linux.sh srtml-exp
    • Finally 
           conda activate srtml-exp
     python -c "import srtml; srtml.init()"

Model Repository

Commands

  • cleanmr: Cleans the model repository
  • lsmr: lists the model repisotory
  • plsmr: specify s3 uri and dive deeper into model repository tree

Experiment1 - Prepoc

One end-to-end running example of image preproc

Commands

For any command run <cmd> --help to get inputs

  • prepoc_profile: profile the vertices given from a config file. Config files look like :

        [
       {
           "Model Name": "resnet50",
           "Accuracy": 75.8
       },
       {
           "Model Name": "resnet34",
           "Accuracy": 75.8
       }
    ]

  • prepoc_populate: puts the profiled models into model repository

  • prepoc_configure: configures the virtual abstract image classification model based on arrival curve config. Config looks like

        [
       {
           "mu (qps)": 100.0,
           "cv": 0,
           "# requests": 2000,
           "Latency Constraint (ms)": 100.0,
           "Planner": "SimulatedAnnealing"
       }
    ]

  • prepoc_provision: provisions the configured models to get latency, throughput information

Demo

prepoc_profile
prepoc_populate
prepoc_configure
prepoc_provision

ls image_preprocessing/two_vertex/accuracy_degradation/virtual/virtual_image_classification.xlsx
ls image_preprocessing/two_vertex/accuracy_degradation/physical/image_classification.xlsx