def run_pos_tagging(client, gpus):
task = TaskType.POS_TAGGING
# Randomly generate app & model names to avoid naming conflicts
app_id = gen_id()
app = 'pos_tagging_app_{}'.format(app_id)
bihmm_model_name = 'BigramHmm_{}'.format(app_id)
py_model_name = 'PyBiLstm_{}'.format(app_id)
print('Adding models "{}" and "{}" to Rafiki...'.format(
bihmm_model_name, py_model_name))
bihmm_model = client.create_model(bihmm_model_name, task, 'examples/models/pos_tagging/BigramHmm.py', \
'BigramHmm', dependencies={})
py_model = client.create_model(py_model_name, task, 'examples/models/pos_tagging/PyBiLstm.py', \
'PyBiLstm', dependencies={ ModelDependency.PYTORCH: '0.4.1' })
model_ids = [bihmm_model['id'], py_model['id']]
print('Creating train job for app "{}" on Rafiki...'.format(app))
budget = {BudgetType.MODEL_TRIAL_COUNT: 5, BudgetType.GPU_COUNT: gpus}
train_dataset_uri = 'https://github.com/nginyc/rafiki-datasets/blob/master/pos_tagging/ptb_for_pos_tagging_train.zip?raw=true'
test_dataset_uri = 'https://github.com/nginyc/rafiki-datasets/blob/master/pos_tagging/ptb_for_pos_tagging_test.zip?raw=true'
train_job = client.create_train_job(app,
task,
train_dataset_uri,
test_dataset_uri,
budget,
models=model_ids)
pprint.pprint(train_job)
print('Waiting for train job to complete...')
print('This might take a few minutes')
wait_until_train_job_has_stopped(client, app)
print('Train job has been stopped')
print(
'Listing best trials of latest train job for app "{}"...'.format(app))
pprint.pprint(client.get_best_trials_of_train_job(app))
print('Creating inference job for app "{}" on Rafiki...'.format(app))
pprint.pprint(client.create_inference_job(app))
predictor_host = get_predictor_host(client, app)
if not predictor_host:
raise Exception('Inference job has errored or stopped')
print('Inference job is running!')
print('Making predictions for queries:')
queries = [['Ms.', 'Haag', 'plays', 'Elianti', '18', '.'],
[
'The', 'luxury', 'auto', 'maker', 'last', 'year', 'sold',
'1,214', 'cars', 'in', 'the', 'U.S.'
]]
print(queries)
predictions = make_predictions(client, predictor_host, queries)
print('Predictions are:')
print(predictions)
print('Stopping inference job...')
pprint.pprint(client.stop_inference_job(app))
def run_speech_recognition(client, train_dataset_path, val_dataset_path, gpus,
hours):
'''
Conducts training with the `TfDeepSpeech` model for the task ``SPEECH_RECOGNITION`.
'''
task = 'SPEECH_RECOGNITION'
# Randomly generate app & model names to avoid naming conflicts
app_id = gen_id()
app = 'speech_recognition_app_{}'.format(app_id)
tf_model_name = 'TfDeepSpeech_{}'.format(app_id)
print('Creating & uploading datasets onto SINGA-Auto...')
train_dataset = client.create_dataset('{}_train'.format(app), task,
train_dataset_path)
pprint(train_dataset)
val_dataset = client.create_dataset('{}_val'.format(app), task,
val_dataset_path)
pprint(val_dataset)
print('Adding models "{}" to SINGA-Auto...'.format(tf_model_name))
tf_model = client.create_model(
tf_model_name,
task,
'examples/models/speech_recognition/TfDeepSpeech.py',
'TfDeepSpeech',
docker_image=IMAGE_TFDEEPSPEECH,
dependencies={
ModelDependency.TENSORFLOW: '1.12.0',
ModelDependency.DS_CTCDECODER: '0.6.0-alpha.4'
})
pprint(tf_model)
print('Creating train job for app "{}" on SINGA-Auto...'.format(app))
budget = {BudgetOption.TIME_HOURS: hours, BudgetOption.GPU_COUNT: gpus}
train_job = client.create_train_job(app,
task,
train_dataset['id'],
val_dataset['id'],
budget,
models=[tf_model['id']])
pprint(train_job)
print('Monitor the train job on SINGA-Auto Web Admin')
def run_pos_tagging(client, train_dataset_path, val_dataset_path, gpus, hours):
'''
Conducts a full train-inference flow on the Penn Treebank sample dataset with
models `BigramHmm` and `PyBiLstm` for the task `POS_TAGGING`.
'''
task = 'POS_TAGGING'
# Randomly generate app & model names to avoid naming conflicts
app_id = gen_id()
app = 'pos_tagging_app_{}'.format(app_id)
bihmm_model_name = 'BigramHmm_{}'.format(app_id)
py_model_name = 'PyBiLstm_{}'.format(app_id)
print('Preprocessing datasets...')
load_sample_ptb(train_dataset_path, val_dataset_path)
print('Creating & uploading datasets onto SINGA-Auto...')
train_dataset = client.create_dataset('{}_train'.format(app), task,
train_dataset_path)
pprint(train_dataset)
val_dataset = client.create_dataset('{}_val'.format(app), task,
val_dataset_path)
pprint(val_dataset)
print('Preprocessing datasets...')
load_sample_ptb(train_dataset_path, val_dataset_path)
print('Adding models "{}" and "{}" to SINGA-Auto...'.format(
bihmm_model_name, py_model_name))
bihmm_model = client.create_model(bihmm_model_name, task, 'examples/models/pos_tagging/BigramHmm.py', \
'BigramHmm', dependencies={})
pprint(bihmm_model)
py_model = client.create_model(py_model_name, task, 'examples/models/pos_tagging/PyBiLstm.py', \
'PyBiLstm', dependencies={ ModelDependency.TORCH: '0.4.1' })
pprint(py_model)
model_ids = [bihmm_model['id'], py_model['id']]
print('Creating train job for app "{}" on SINGA-Auto...'.format(app))
budget = {BudgetOption.TIME_HOURS: hours, BudgetOption.GPU_COUNT: gpus}
train_job = client.create_train_job(app,
task,
train_dataset['id'],
val_dataset['id'],
budget,
models=model_ids)
pprint(train_job)
print('Waiting for train job to complete...')
print('This might take a few minutes')
wait_until_train_job_has_stopped(client, app)
print('Train job has been stopped')
print('Listing best trials of latest train job for app "{}"...'.format(app))
pprint(client.get_best_trials_of_train_job(app))
print('Creating inference job for app "{}" on SINGA-Auto...'.format(app))
pprint(client.create_inference_job(app))
predictor_host = get_predictor_host(client, app)
if not predictor_host:
raise Exception('Inference job has errored or stopped')
print('Inference job is running!')
print('Making predictions for queries:')
queries = [['Ms.', 'Haag', 'plays', 'Elianti', '18', '.'],
[
'The', 'luxury', 'auto', 'maker', 'last', 'year', 'sold',
'1,214', 'cars', 'in', 'the', 'U.S.'
]]
print(queries)
predictions = make_predictions(client, predictor_host, queries)
print('Predictions are:')
print(predictions)
print('Stopping inference job...')
pprint(client.stop_inference_job(app))
def run_tabular_regression(client,
csv_file_url,
gpus,
hours,
features=None,
target=None,
queries=None):
'''
Runs a sample full train-inference flow for the task ``TABULAR_REGRESSION``.
'''
task = 'TABULAR_REGRESSION'
# Randomly generate app & model names to avoid naming conflicts
app_id = gen_id()
app = 'tabular_regression_app_{}'.format(app_id)
xgb_model_name = 'XgbReg_{}'.format(app_id)
train_dataset_path = 'data/{}_train.csv'.format(app)
val_dataset_path = 'data/{}_val.csv'.format(app)
print('Preprocessing dataset...')
load(csv_file_url, train_dataset_path, val_dataset_path)
print('Creating & uploading datasets onto SINGA-Auto...')
train_dataset = client.create_dataset('{}_train'.format(app), task,
train_dataset_path)
pprint(train_dataset)
val_dataset = client.create_dataset('{}_val'.format(app), task,
val_dataset_path)
pprint(val_dataset)
print('Adding models "{}" to SINGA-Auto...'.format(xgb_model_name))
xgb_model = client.create_model(xgb_model_name, task, 'examples/models/tabular_regression/XgbReg.py', \
'XgbReg', dependencies={ ModelDependency.XGBOOST: '0.90' })
pprint(xgb_model)
print('Creating train job for app "{}" on SINGA-Auto...'.format(app))
budget = {BudgetOption.TIME_HOURS: hours, BudgetOption.GPU_COUNT: gpus}
train_job = client.create_train_job(app,
task,
train_dataset['id'],
val_dataset['id'],
budget,
models=[xgb_model['id']],
train_args={
'features': features,
'target': target
})
pprint(train_job)
print('Waiting for train job to complete...')
print('This might take a few minutes')
wait_until_train_job_has_stopped(client, app)
print('Train job has been stopped')
print(
'Listing best trials of latest train job for app "{}"...'.format(app))
pprint(client.get_best_trials_of_train_job(app))
print('Creating inference job for app "{}" on SINGA-Auto...'.format(app))
pprint(client.create_inference_job(app))
predictor_host = get_predictor_host(client, app)
if not predictor_host:
raise Exception('Inference job has errored or stopped')
print('Inference job is running!')
if queries is not None:
print('Making predictions for queries:')
print(queries)
predictions = make_predictions(client, predictor_host, queries)
print('Predictions are:')
print(predictions)
print('Stopping inference job...')
pprint(client.stop_inference_job(app))
