def submit_synthetic_local(c, epochs=1):
"""Submit TensorFlow training job using synthetic imagenet data for local execution
Args:
epochs (int, optional): Number of epochs to run training for. Defaults to 1.
"""
from aml_compute import TFExperimentCLI
exp = TFExperimentCLI("synthetic_images_local")
run = exp.submit_local(
os.path.join(_BASE_PATH, "src"),
"resnet_main.py",
{"--epochs": epochs},
dependencies_file=os.path.join(_BASE_PATH, "environment_gpu.yml"),
wait_for_completion=True,
)
print(run)
def submit_local(c):
"""This command isn't implemented please modify to use.
The call below will work for submitting jobs to execute locally on a GPU.
"""
raise NotImplementedError(
"You need to modify this call before being able to use it")
from aml_compute import TFExperimentCLI
exp = TFExperimentCLI("<YOUR-EXPERIMENT-NAME>")
run = exp.submit_local(
os.path.join(_BASE_PATH, "src"),
"<YOUR-TRAINING-SCRIPT>",
{"YOUR": "ARGS"},
dependencies_file="TensorFlow/environment_gpu.yml",
wait_for_completion=True,
)
print(run)
def submit_benchmark_local(c):
"""Submit PyTorch training job using synthetic data for local execution
"""
from aml_compute import TFExperimentCLI
exp = TFExperimentCLI("synthetic_images_local")
run = exp.submit_local(
os.path.join(_BASE_PATH, "src"),
"pytorch_synthetic_benchmark.py",
{
"--model": "resnet50",
"--batch-size": 64
},
dependencies_file=os.path.join(_BASE_PATH, "environment_gpu.yml"),
wait_for_completion=True,
)
print(run)
def submit_tf_benchmark_local(c):
"""Submits TensorFlow benchmark job using synthetic data for local execution
Note:
Runs ResNet 50 model with batch size of 256 and mixed precision
"""
from aml_compute import TFExperimentCLI
exp = TFExperimentCLI("tf_benchmark")
run = exp.submit_local(
os.path.join(_BASE_PATH, "src"),
"tf_cnn_benchmarks.py",
{
"--model": "resnet50",
"--batch_size": 256,
"--variable_update": "horovod",
"--use_fp16": "",
},
dependencies_file=os.path.join(_BASE_PATH, "environment_gpu.yml"),
wait_for_completion=True,
)
print(run)
def submit_images_local(c, epochs=1):
"""Submit TensorFlow training job using real imagenet data for local execution
Args:
epochs (int, optional): Number of epochs to run training for. Defaults to 1.
"""
from aml_compute import TFExperimentCLI
exp = TFExperimentCLI("real_images_local")
run = exp.submit_local(
os.path.join(_BASE_PATH, "src"),
"resnet_main.py",
{
"--training_data_path": "/data/train",
"--validation_data_path": "/data/validation",
"--epochs": epochs,
"--data_type": "images",
"--data-format": "channels_first",
},
dependencies_file=os.path.join(_BASE_PATH, "environment_gpu.yml"),
docker_args=["-v", f"{env_values['DATA']}:/data"],
wait_for_completion=True,
)
print(run)
