def get_memory_usage_graphics():
"""
Get graphic representation of memory usage.
Returns:
Response, the graphic representation of memory usage.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/memory-graphics
"""
summary_dir = request.args.get("dir")
profiler_dir_abs = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
check_train_job_and_profiler_dir(profiler_dir_abs)
device_id = request.args.get("device_id", default='0')
to_int(device_id, 'device_id')
device_type = request.args.get("device_type", default='ascend')
if device_type not in ['ascend']:
logger.info(
"Invalid device_type, Memory Usage only supports Ascend for now.")
raise ParamValueError("Invalid device_type.")
analyser = AnalyserFactory.instance().get_analyser('memory_usage',
profiler_dir_abs,
device_id)
graphics = analyser.get_memory_usage_graphics(device_type)
return graphics
def get_memory_usage_breakdowns():
"""
Get memory breakdowns of each node.
Returns:
Response, the memory breakdowns for each node.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/memory-breakdowns
"""
summary_dir = request.args.get("dir")
profiler_dir_abs = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
check_train_job_and_profiler_dir(profiler_dir_abs)
device_id = request.args.get("device_id", default='0')
to_int(device_id, 'device_id')
device_type = request.args.get("device_type", default='ascend')
graph_id = request.args.get("graph_id", default='0')
node_id = request.args.get("node_id", default='0')
node_id = to_int(node_id, 'node_id')
if device_type not in ['ascend']:
logger.error(
"Invalid device_type, Memory Usage only supports Ascend for now.")
raise ParamValueError("Invalid device_type.")
analyser = AnalyserFactory.instance().get_analyser('memory_usage',
profiler_dir_abs,
device_id)
breakdowns = analyser.get_memory_usage_breakdowns(device_type, graph_id,
node_id)
return breakdowns
def get_timeline_summary():
"""
Get timeline summary info.
Returns:
Response, the timeline summary info.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/timeline-summary
"""
summary_dir = request.args.get("dir")
profiler_dir_abs = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
check_train_job_and_profiler_dir(profiler_dir_abs)
device_id = request.args.get("device_id", default='0')
to_int(device_id, 'device_id')
device_type = request.args.get("device_type", default='ascend')
if device_type not in ['gpu', 'ascend']:
logger.info(
"Invalid device_type, device_type should be gpu or ascend.")
raise ParamValueError("Invalid device_type.")
analyser = AnalyserFactory.instance().get_analyser('timeline',
profiler_dir_abs,
device_id)
summary = analyser.get_timeline_summary(device_type)
return summary
def get_timeline_detail():
"""
Get timeline detail.
Returns:
Response, the detail information of timeline.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/timeline
"""
summary_dir = request.args.get("dir")
profiler_dir_abs = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
check_train_job_and_profiler_dir(profiler_dir_abs)
device_id = request.args.get("device_id", default='0')
to_int(device_id, 'device_id')
device_type = request.args.get("device_type", default='ascend')
scope_name_num = request.args.get("scope_name_num", default='0')
if device_type not in ['gpu', 'ascend']:
logger.info(
"Invalid device_type, device_type should be gpu or ascend.")
raise ParamValueError("Invalid device_type.")
analyser = AnalyserFactory.instance().get_analyser('timeline',
profiler_dir_abs,
device_id)
timeline = analyser.get_display_timeline(device_type, scope_name_num)
return jsonify(timeline)
def get_training_trace_graph():
"""
Get training trace info of one step.
Returns:
Response, the training trace info of one step.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/training-trace/graph
"""
summary_dir = request.args.get("dir")
profiler_dir_abs = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
check_train_job_and_profiler_dir(profiler_dir_abs)
graph_type = request.args.get("type", default='0')
graph_type = to_int(graph_type, 'graph_type')
device_id = request.args.get("device_id", default='0')
_ = to_int(device_id, 'device_id')
graph_info = {}
try:
analyser = AnalyserFactory.instance().get_analyser(
'step_trace', profiler_dir_abs, device_id)
except ProfilerFileNotFoundException:
return jsonify(graph_info)
graph_info = analyser.query(
{'filter_condition': {
'mode': 'step',
'step_id': graph_type
}})
graph_info['summary'] = analyser.summary
graph_info['point_info'] = analyser.point_info
return jsonify(graph_info)
def get_target_time_info():
"""
Get all the time information of the specified column.
Returns:
Response, all the time information of the specified column.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/training-trace/target-time-info
"""
summary_dir = request.args.get("dir")
profiler_dir_abs = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
check_train_job_and_profiler_dir(profiler_dir_abs)
proc_name = request.args.get("type")
validate_ui_proc(proc_name)
device_id = request.args.get("device_id", default='0')
to_int(device_id, 'device_id')
analyser = AnalyserFactory.instance().get_analyser('step_trace',
profiler_dir_abs,
device_id)
target_time_info = analyser.query(
{'filter_condition': {
'mode': 'proc',
'proc_name': proc_name
}})
target_time_info['summary'] = analyser.summary
return jsonify(target_time_info)
def get_timeline_detail():
"""
Get timeline detail.
Returns:
Response, the detail information of timeline.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/timeline
"""
summary_dir = request.args.get("dir")
profiler_dir = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
if not os.path.exists(profiler_dir):
raise ProfilerDirNotFoundException(msg=summary_dir)
device_id = request.args.get("device_id", default='0')
_ = to_int(device_id, 'device_id')
device_type = request.args.get("device_type", default='ascend')
if device_type not in ['gpu', 'ascend']:
logger.info(
"Invalid device_type, device_type should be gpu or ascend.")
raise ParamValueError("Invalid device_type.")
analyser = AnalyserFactory.instance().get_analyser('timeline',
profiler_dir, device_id)
timeline = analyser.get_display_timeline(device_type)
return jsonify(timeline)
def get_training_trace_graph():
"""
Get training trace info of one step.
Returns:
Response, the training trace info of one step.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/training-trace/graph
"""
summary_dir = request.args.get("dir")
profiler_dir_abs = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
check_train_job_and_profiler_dir(profiler_dir_abs)
graph_type = request.args.get("type", default='0')
graph_type = to_int(graph_type, 'graph_type')
device_id = request.args.get("device_id", default='0')
to_int(device_id, 'device_id')
graph_info = {}
try:
analyser = AnalyserFactory.instance().get_analyser(
'step_trace', profiler_dir_abs, device_id)
except ProfilerFileNotFoundException:
return jsonify(graph_info)
graph_info = analyser.query(
{'filter_condition': {
'mode': 'step',
'step_id': graph_type
}})
graph_info['summary'] = analyser.summary
graph_info['point_info'] = analyser.point_info(graph_type)
graph_info['is_heterogeneous'] = False
# In heterogeneous training scene, do not display step trace data.
cpu_op_type_file_name = f"cpu_op_type_info_{device_id}.csv"
if cpu_op_type_file_name in os.listdir(profiler_dir_abs):
graph_info = {'is_heterogeneous': True}
return jsonify(graph_info)
def get_timeline_detail():
"""
Get timeline detail.
Returns:
Response, the detail information of timeline.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/timeline
"""
summary_dir = request.args.get("dir")
profiler_dir = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
device_id = request.args.get("device_id", default='0')
_ = to_int(device_id, 'device_id')
analyser = AnalyserFactory.instance().get_analyser('timeline',
profiler_dir, device_id)
timeline = analyser.get_display_timeline()
return jsonify(timeline)
def get_timeline_summary():
"""
Get timeline summary info.
Returns:
Response, the timeline summary info.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/timeline-summary
"""
summary_dir = request.args.get("dir")
profiler_dir = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
device_id = request.args.get("device_id", default='0')
_ = to_int(device_id, 'device_id')
analyser = AnalyserFactory.instance().get_analyser('timeline',
profiler_dir, device_id)
summary = analyser.get_timeline_summary()
return summary
def get_flops_summary():
"""
Get flops summary info.
Returns:
Response, the flops summary info.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/flops-summary
"""
train_id = request.args.get("train_id")
profiler_dir_abs = validate_and_normalize_profiler_path(
train_id, settings.SUMMARY_BASE_DIR)
check_train_job_and_profiler_dir(profiler_dir_abs)
device_id = request.args.get("device_id", default='0')
to_int(device_id, 'device_id')
analyser = AnalyserFactory.instance().get_analyser('flops',
profiler_dir_abs,
device_id)
summary = analyser.get_flops_summary()
return summary
def get_timeline_summary():
"""
Get timeline summary info.
Returns:
Response, the timeline summary info.
Examples:
>>> GET http://xxxx/v1/mindinsight/profile/timeline-summary
"""
summary_dir = request.args.get("dir")
profiler_dir = validate_and_normalize_profiler_path(
summary_dir, settings.SUMMARY_BASE_DIR)
if not os.path.exists(profiler_dir):
msg = 'The profiler dir is not found!'
raise ProfilerDirNotFoundException(msg=msg)
device_id = request.args.get("device_id", default='0')
_ = to_int(device_id, 'device_id')
analyser = AnalyserFactory.instance().get_analyser('timeline',
profiler_dir, device_id)
summary = analyser.get_timeline_summary()
return summary
