RNN-T MLPerf Inference BKC

HW & SW requirements

  SPR 2 sockets
  GCC >= 11

Steps to run RNN-T with three options

Option 1: Run on bare metal

1. Install anaconda 3.0

  wget https://repo.continuum.io/archive/Anaconda3-5.0.0-Linux-x86_64.sh -O anaconda3.sh
  chmod +x anaconda3.sh
  ~/anaconda3.sh -b -p ~/anaconda3
  export PATH=~/anaconda3/bin:$PATH

2. End-to-end run inference

Execute run.sh. The end-to-end process including:

STAGE(default -2)	STEP
-2	Prepare conda environment
-1	Prepare environment
0	Download model
1	Download dataset
2	Pre-process dataset
3	Calibration
4	Build model
5	Run Offline/Server accuracy & benchmark

You can also use the following command to start with your custom conda-env/work-dir/step.

  [CONDA_ENV] [WORK_DIR] [STAGE] bash run.sh

Option 2: Build docker container

  cd docker
  bash build_rnnt-99_container.sh
  docker run --name intel_rnnt --privileged -itd -v /data/mlperf_data:/data/mlperf_data --net=host --ipc=host mlperf_inference_rnnt:3.0
  docker ps -a #get container "id"
  docker exec -it <id> bash
  cd /opt/workdir/code/rnnt/pytorch-cpu
  SKIP_BUILD=1 STAGE=0 bash run.sh

Option 3: Pull docker image

  <TBD: command to pull docker>
  docker run --name intel_rnnt --privileged -itd -v /data/mlperf_data:/data/mlperf_data --net=host --ipc=host mlperf_inference_rnnt:3.0
  docker ps -a #get container "id"
  docker exec -it <id> bash
  cd /opt/workdir/code/rnnt/pytorch-cpu
  SKIP_BUILD=1 STAGE=0 bash run.sh