# Header meta
spec_version = "0.1.0"

# These fields don't have any impact on the job to run, they contain
# merely informative data so the benchmark can be categorized when displayed
# in the dashboard.
[info]
description = """ TensorFlow training single-node {tf_version} CPU {py_version} resnet50 synthetic
"""
scheduling = 'single_run'

[info.labels]
# Labels and values must be 63 characters or less, beginning and ending with an alphanumeric character
# ([a-z0-9A-Z]) with dashes (-), underscores (_), dots (.), and alphanumerics between
task_name = " tf_train_single_node_{tf_version}_cpu_{py_version}_resnet50_synthetic"

# 1. Hardware
[hardware]
instance_type = "c5.18xlarge"
strategy = "single_node"
aws_zone_id="use1-az2"

# 2. Environment
[env]
docker_image = "tensorflow-{tf_version} {py_version} cpu docker image"

[env.vars]
OMP_NUM_THREADS = "36"
TENSORFLOW_INTER_OP_PARALLELISM='2'
TENSORFLOW_INTRA_OP_PARALLELISM='72'
KMP_AFFINITY='granularity=fine,verbose,compact,1,0'
TENSORFLOW_SESSION_PARALLELISM='9'
KMP_BLOCKTIME='1'
KMP_SETTINGS='0'

# 3. Machine learning related settings:
# dataset, benchmark code and parameters it takes
[ml]

benchmark_code = "python $(BAI_SCRIPTS_PATH)/benchmarks/scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py"

args = "--batch_size=256 --model=resnet50_v1.5 --train_dir=$HOME/test00 --device=cpu --data_format=NHWC --num_inter_threads=0 --num_intra_threads=36 --mkl=True --kmp_blocktime=0"

# 5. Output
[output]
# [Opt] Custom metrics descriptions
# List all required metrics descriptions below.
# Make an entry in same format as the one below.
[[output.metrics]]
# Name of the metric that will appear in the dashboards.
name = "throughput"
# Metric unit (required)
units = "img/sec"
# Pattern for log parsing for this metric.
pattern = 'images\/sec: (\d*.\d+|\d+)'