Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

HPC 3.0.0 missing files #331

Merged
merged 2 commits into from
Sep 18, 2023
Merged

HPC 3.0.0 missing files #331

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7393de9
Fix power weak scaling concatenation
pgmpablo157321 Sep 7, 2023
7099780
Add HPC 3.0 missing files
pgmpablo157321 Sep 7, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
82 changes: 82 additions & 0 deletions mlperf_logging/rcp_checker/hpc_3.0.0/rcps_cosmoflow.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,82 @@
{
"cosmoflow_ref_32": {
"Benchmark": "cosmoflow",
"BS": 32,
"Epochs to converge": [ 8, 8, 9, 8, 6, 8, 6, 6, 7, 8, 8, 8, 9, 8, 7, 8, 6, 9, 8, 8, 8, 8, 9, 8, 9, 8, 8, 8, 8, 8, 8, 8, 8 ],
"Hyperparams": {
"global_batch_size": 32,
"opt_name": "SGD",
"opt_base_learning_rate": 0.001,
"opt_learning_rate_warmup_epochs": 2,
"opt_learning_rate_warmup_factor": 4.0,
"opt_learning_rate_decay_boundary_epochs": [ 4, 6 ],
"opt_learning_rate_decay_factor": [0.25, 0.0625],
"dropout": 0.5,
"opt_weight_decay": 0.0
}
},
"cosmoflow_ref_64": {
"Benchmark": "cosmoflow",
"BS": 64,
"Epochs to converge": [ 19, 18, 20, 17, 18, 19, 18, 18, 18, 18, 18, 18, 18, 19, 18, 18, 18, 18, 18, 20 ],
"Hyperparams": {
"global_batch_size": 64,
"opt_name": "SGD",
"opt_base_learning_rate": 0.001,
"opt_learning_rate_warmup_epochs": 4,
"opt_learning_rate_warmup_factor": 1.0,
"opt_learning_rate_decay_boundary_epochs": [ 16, 32 ],
"opt_learning_rate_decay_factor": 0.25,
"dropout": 0.0,
"opt_weight_decay": 0.01
}
},
"cosmoflow_ref_128": {
"Benchmark": "cosmoflow",
"BS": 128,
"Epochs to converge": [ 16, 18, 18, 18, 18, 17, 18, 18, 18, 18, 18, 18, 17, 18, 18, 18, 19, 18, 18, 18 ],
"Hyperparams": {
"global_batch_size": 128,
"opt_name": "SGD",
"opt_base_learning_rate": 0.004,
"opt_learning_rate_warmup_epochs": 4,
"opt_learning_rate_warmup_factor": 4.0,
"opt_learning_rate_decay_boundary_epochs": [ 16, 32 ],
"opt_learning_rate_decay_factor": 0.25,
"dropout": 0.5,
"opt_weight_decay": 0.0
}
},
"cosmoflow_ref_512": {
"Benchmark": "cosmoflow",
"BS": 512,
"Epochs to converge": [ 22, 23, 23, 24, 24, 21, 23, 23, 23, 23, 24, 23, 23, 21, 23, 23, 21, 23, 23, 23 ],
"Hyperparams": {
"global_batch_size": 512,
"opt_name": "SGD",
"opt_base_learning_rate": 0.006,
"opt_learning_rate_warmup_epochs": 6,
"opt_learning_rate_warmup_factor": 4.0,
"opt_learning_rate_decay_boundary_epochs": [ 19, 21, 22, 23 ],
"opt_learning_rate_decay_factor": [0.5, 0.25, 0.125, 0.0625],
"dropout": 0.0,
"opt_weight_decay": 0.01
}
},
"cosmoflow_ref_1024": {
"Benchmark": "cosmoflow",
"BS": 1024,
"Epochs to converge": [ 42, 38, 42, 40, 40, 39, 43, 37, 39, 37, 43, 39, 38, 42, 40, 42, 42, 38, 36, 43 ],
"Hyperparams": {
"global_batch_size": 1024,
"opt_name": "SGD",
"opt_base_learning_rate": 0.012,
"opt_learning_rate_warmup_epochs": 0,
"opt_learning_rate_warmup_factor": 1.0,
"opt_learning_rate_decay_boundary_epochs": [ 32, 64 ],
"opt_learning_rate_decay_factor": 0.25,
"dropout": 0.5,
"opt_weight_decay": 0.0
}
}
}
119 changes: 119 additions & 0 deletions mlperf_logging/rcp_checker/hpc_3.0.0/rcps_deepcam.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,119 @@
{
"deepcam_ref_128": {
"Benchmark": "deepcam",
"BS": 128,
"Epochs to converge": [ 7, 7, 7, 7, 7, 7, 7, 8, 8, 6, 7, 8, 7, 6, 6, 7, 7, 6, 7, 7 ],
"Hyperparams": {
"batchnorm_groupsize": 1,
"global_batch_size": 128,
"gradient_accumulation_frequency": 1,
"num_workers": 32,
"opt_betas": [ 0.9, 0.999 ],
"opt_bias_correction": true,
"opt_eps": 1e-06,
"opt_grad_averaging": true,
"opt_lr": 0.00155,
"opt_max_grad_norm": 1.0,
"opt_weight_decay": 0.01,
"opt_name": "LAMB",
"scheduler_decay_rate": "0.1",
"scheduler_lr_warmup_factor": 1.0,
"scheduler_lr_warmup_steps": 0,
"scheduler_t_max": "9000",
"scheduler_eta_min": "0.0",
"scheduler_type": "cosine_annealing"
}
},
"deepcam_ref_256": {
"Benchmark": "deepcam",
"BS": 256,
"Epochs to converge": [ 9, 9, 9, 9, 9, 9, 9, 9, 9, 9 ],
"Hyperparams": {
"batchnorm_groupsize": 1,
"global_batch_size": 256,
"gradient_accumulation_frequency": 1,
"opt_betas": [ 0.9, 0.999 ],
"opt_bias_correction": true,
"opt_eps": 1e-06,
"opt_grad_averaging": true,
"opt_lr": 0.002,
"opt_max_grad_norm": 1.0,
"opt_weight_decay": 0.01,
"opt_name": "LAMB",
"scheduler_decay_rate": 0.1,
"scheduler_lr_warmup_factor": 1.0,
"scheduler_lr_warmup_steps": 0,
"scheduler_milestones": [ 4096, 8192 ],
"scheduler_type": "multistep"
}
},
"deepcam_ref_512": {
"Benchmark": "deepcam",
"BS": 512,
"Epochs to converge": [ 11, 11, 11, 11, 11, 10, 16, 12, 11, 11, 11, 10, 11, 11, 11, 10, 11, 11, 11, 11, 11, 11, 11, 12, 11, 11, 11, 11, 13, 10, 11, 13, 11, 11, 11, 10, 14, 10, 10, 10 ],
"Hyperparams": {
"batchnorm_groupsize": 1,
"global_batch_size": 512,
"gradient_accumulation_frequency": 1,
"opt_betas": [ 0.9, 0.999 ],
"opt_bias_correction": true,
"opt_eps": 1e-06,
"opt_grad_averaging": true,
"opt_lr": 0.004,
"opt_max_grad_norm": 1.0,
"opt_weight_decay": 0.01,
"opt_name": "LAMB",
"scheduler_decay_rate": 0.1,
"scheduler_lr_warmup_factor": 1.0,
"scheduler_lr_warmup_steps": 100,
"scheduler_milestones": [ 2048, 4096 ],
"scheduler_type": "multistep"
}
},
"deepcam_ref_1024": {
"Benchmark": "deepcam",
"BS": 1024,
"Epochs to converge": [ 13, 13, 12, 13, 13, 13, 14, 13, 13, 13 ],
"Hyperparams": {
"batchnorm_groupsize": 1,
"global_batch_size": 1024,
"gradient_accumulation_frequency": 1,
"opt_betas": [ 0.9, 0.999 ],
"opt_bias_correction": true,
"opt_eps": 1e-06,
"opt_grad_averaging": true,
"opt_lr": 0.004,
"opt_max_grad_norm": 1.0,
"opt_weight_decay": 0.01,
"opt_name": "LAMB",
"scheduler_decay_rate": 0.1,
"scheduler_lr_warmup_factor": 1.0,
"scheduler_lr_warmup_steps": 200,
"scheduler_milestones": [ 1100, 4096 ],
"scheduler_type": "multistep"
}
},
"deepcam_ref_2048": {
"Benchmark": "deepcam",
"BS": 2048,
"Epochs to converge": [ 23, 23, 22, 23, 22, 26, 22, 22, 24, 22 ],
"Hyperparams": {
"batchnorm_groupsize": 1,
"global_batch_size": 2048,
"gradient_accumulation_frequency": 1,
"opt_betas": [ 0.9, 0.999 ],
"opt_bias_correction": true,
"opt_eps": 1e-06,
"opt_grad_averaging": true,
"opt_lr": 0.0055,
"opt_max_grad_norm": 1.0,
"opt_weight_decay": 0.01,
"opt_name": "LAMB",
"scheduler_decay_rate": "0.1",
"scheduler_lr_warmup_factor": 1.0,
"scheduler_lr_warmup_steps": 400,
"scheduler_milestones": [ 800 ],
"scheduler_type": "multistep"
}
}
}
41 changes: 41 additions & 0 deletions mlperf_logging/rcp_checker/hpc_3.0.0/rcps_oc20.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,41 @@
{
"oc20_ref_256": {
"Benchmark": "oc20",
"BS": 256,
"Epochs to converge": [ 20, 18, 20, 20, 22, 18, 19, 19, 21, 21 ],
"Hyperparams": {
"global_batch_size": 256,
"opt_base_learning_rate": 0.0004,
"opt_learning_rate_warmup_steps": 31252,
"opt_learning_rate_warmup_factor": 0.2,
"opt_learning_rate_decay_boundary_steps": [ 125008, 187512, 250016 ],
"opt_learning_rate_decay_factor": 0.1
}
},
"oc20_ref_1024": {
"Benchmark": "oc20",
"BS": 1024,
"Epochs to converge": [ 23, 25, 22, 25, 25, 25, 25, 24, 23, 25 ],
"Hyperparams": {
"global_batch_size": 1024,
"opt_base_learning_rate": 0.0012,
"opt_learning_rate_warmup_steps": 7816,
"opt_learning_rate_warmup_factor": 0.2,
"opt_learning_rate_decay_boundary_steps": [ 31264, 46896 ],
"opt_learning_rate_decay_factor": 0.1
}
},
"oc20_ref_2048": {
"Benchmark": "oc20",
"BS": 2048,
"Epochs to converge": [ 33, 32, 33, 33, 33, 34, 33, 33, 30, 33 ],
"Hyperparams": {
"global_batch_size": 2048,
"opt_base_learning_rate": 0.0016,
"opt_learning_rate_warmup_steps": 3908,
"opt_learning_rate_warmup_factor": 0.2,
"opt_learning_rate_decay_boundary_steps": [ 23448, 31264 ],
"opt_learning_rate_decay_factor": 0.1
}
}
}
21 changes: 16 additions & 5 deletions mlperf_logging/result_summarizer/result_summarizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -322,7 +322,7 @@ def _compute_strong_scaling_scores(desc, system_folder, usage, ruleset):
benchmark_folder_parent = os.path.join(
system_folder, 'strong') if usage == 'hpc' else system_folder
if not os.path.isdir(benchmark_folder_parent):
return benchmark_scores
return benchmark_scores, benchmark_power_scores
for benchmark_folder in _get_sub_folders(benchmark_folder_parent):
folder_parts = benchmark_folder.split('/')
# Check if this benchmark has power results
Expand Down Expand Up @@ -404,7 +404,7 @@ def _compute_weak_scaling_scores(desc, system_folder, usage, ruleset):
benchmark_power_scores = {}
has_power = None
if not os.path.isdir(benchmark_folder_parent):
return benchmark_scores
return benchmark_scores, benchmark_power_scores
for benchmark_folder in _get_sub_folders(benchmark_folder_parent):
folder_parts = benchmark_folder.split('/')
benchmark = _benchmark_alias(folder_parts[-1])
Expand Down Expand Up @@ -457,13 +457,24 @@ def _compute_weak_scaling_scores(desc, system_folder, usage, ruleset):
olympic_avg = _compute_olympic_average(
power_scores, 1, 1)
if olympic_avg is not None:
benchmark_power_scores[benchmark] = olympic_avg
benchmark_power_scores['{}:{}'.format(
benchmark,
'time_to_train_all',
)] = olympic_avg
benchmark_power_scores['{}:{}'.format(
benchmark,
'number_of_models',
)] = olympic_avg
benchmark_power_scores['{}:{}'.format(
benchmark,
'instance_scale',
)] = olympic_avg

_fill_empty_benchmark_scores(benchmark_scores,
usage,
ruleset,
weak_scaling=True)
_fill_empty_benchmark_scores(benchmark_power_scores, usage, ruleset)
_fill_empty_benchmark_scores(benchmark_power_scores, usage, ruleset, weak_scaling=True)
return benchmark_scores, benchmark_power_scores


Expand Down Expand Up @@ -565,7 +576,7 @@ def summarize_results(folder, usage, ruleset, csv_file=None):
ruleset,
weak_scaling=True)
power_summary = _get_empty_summary(usage, ruleset)
power_weak_scaling_summary = _get_empty_summary(usage, ruleset)
power_weak_scaling_summary = _get_empty_summary(usage, ruleset, weak_scaling=True)

for system_folder in _get_sub_folders(results_folder):
folder_parts = system_folder.split('/')
Expand Down
18 changes: 18 additions & 0 deletions scripts/verify_for_v3.0_hpc.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
#!/bin/bash

set -e

# rcp_bypass and rcp_bert_train_samples packahe checker params
# need to be retrieved at package_checker_params file at top-level submission dir.
PACKAGE_CHECKER_PARAMS=""
PACKAGE_CHECKER_PARAMS_FILE="$1/package_checker_params"
if test -f "$PACKAGE_CHECKER_PARAMS_FILE"; then
while IFS= read -r line
do
PACKAGE_CHECKER_PARAMS="$PACKAGE_CHECKER_PARAMS --$line"
done < "$PACKAGE_CHECKER_PARAMS_FILE"
fi

python3 -m mlperf_logging.package_checker $1 hpc 3.0.0 $PACKAGE_CHECKER_PARAMS
python3 -m mlperf_logging.result_summarizer $1 hpc 3.0.0
python3 -m mlperf_logging.repo_checker $1 hpc 3.0.0