mlperf.conf

# The format of this config file is 'key = value'.
# The key has the format 'model.scenario.key'. Value is mostly int64_t.
# Model maybe '*' as wildcard. In that case the value applies to all models.
# All times are in milli seconds

# Set performance_sample_count for each model.
# User can optionally set this to higher values in user.conf.
mobilenet.*.performance_sample_count_override = 1024
gnmt.*.performance_sample_count_override = 3903900
resnet50.*.performance_sample_count_override = 1024
ssd-mobilenet.*.performance_sample_count_override = 256
ssd-resnet34.*.performance_sample_count_override = 64
bert.*.performance_sample_count_override = 10833
dlrm.*.performance_sample_count_override = 204800
rnnt.*.performance_sample_count_override = 2513
3d-unet.*.performance_sample_count_override = 0

# Set seeds. The seeds will be distributed two weeks before the submission.
# 0x168ad48ada698a73
*.*.qsl_rng_seed = 1624344308455410291
# 0x07303fed113b8976
*.*.sample_index_rng_seed = 517984244576520566
# 0x8b7e1740dacb67f9
*.*.schedule_rng_seed = 10051496985653635065

*.SingleStream.target_latency_percentile = 90
*.SingleStream.min_duration = 600000
*.SingleStream.min_query_count = 1024

*.MultiStream.target_qps = 20
*.MultiStream.target_latency_percentile = 99
*.MultiStream.max_async_queries = 1
*.MultiStream.target_latency = 50
*.MultiStream.min_duration = 600000
*.MultiStream.min_query_count = 270336
ssd-resnet34.MultiStream.target_qps = 15
ssd-resnet34.MultiStream.target_latency = 66
gnmt.MultiStream.min_query_count = 90112
gnmt.MultiStream.target_latency = 100
gnmt.MultiStream.target_qps = 10
gnmt.MultiStream.target_latency_percentile = 97
3d-unet.Offline.sample_concatenate_permutation = 1

*.Server.target_latency = 10
*.Server.target_latency_percentile = 99
*.Server.target_duration = 0
*.Server.min_duration = 600000
*.Server.min_query_count = 270336
resnet50.Server.target_latency = 15
ssd-resnet34.Server.target_latency = 100
gnmt.Server.min_query_count = 90112
gnmt.Server.target_latency = 250
gnmt.Server.target_latency_percentile = 97
bert.Server.target_latency = 130
dlrm.Server.target_latency = 30
rnnt.Server.target_latency = 1000

*.Offline.target_latency_percentile = 90
*.Offline.min_duration = 600000
# In Offline scenario, we always have one query. But LoadGen maps this to
# min_sample_count internally in Offline scenario, so set this to 24576 since
# the rule requires that Offline scenario run for at least 24576 samples.
*.Offline.min_query_count = 24576

# These fields should be defined and overridden by user.conf.
*.SingleStream.target_latency = 10
*.Server.target_qps = 1.0
*.Offline.target_qps = 1.0
*.MultiStream.samples_per_query = 4