TensorRT-LLMs/tests/integration/defs/perf/data.py
Kaiyu Xie 2631f21089
Update (#2978)
Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>
2025-03-23 16:39:35 +08:00

148 lines
4.2 KiB
Python

# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
File to house data classes related to perf runs.
Compatible with data_export.py for exporting data.
"""
class Data:
"""Base class to support easier ways to interface and assign data to class."""
def __init__(self, **kwargs):
"""
Simple constructor that supports optional params
to hook to class objects.
"""
for key, value in kwargs.items():
if hasattr(self, key):
setattr(self, key, value)
else:
raise KeyError("Unable to assign value to PerfResult: " + key)
def __str__(self):
return str(self.__dict__)
def __repr__(self):
return str(self.__dict__)
def get(self, key, *args, **kwargs):
"""
Args:
key - Get key of the data
default (optional) - Returns given default value if key is None.
"""
if len(args) > 1:
raise RuntimeError(
"Expected 1-2 parameters, but {} were given: {}".format(
len(args), args[1:]))
has_default = False
default = None
if "default" in kwargs and len(args) == 1:
raise RuntimeError(
"Expected default={} but did not expect {} in parameters".
format(kwargs["default"], args[0]))
if "default" in kwargs:
has_default = True
default = kwargs["default"]
elif len(args) == 1:
has_default = True
default = args[0]
# Used so that the class can act like a dictionary
# Works for now instead of meta class
return getattr(self, key, default) if has_default else getattr(
self, key)
def update(self, dct):
for key, value in dct.items():
setattr(self, key, value)
class SessionData(Data):
"""
Class to store session specific results for perf runs.
"""
start_timestamp = None
end_timestamp = None
# Native NRSU specific collection
os_properties = None
cpu_properties = None
gpu_properties = None
nvidia_device_count = None
nvidia_driver_version = None
# OS Specific
username = None
hostname = None
ip = None
# TensorRT specific properties
trt_change_id = None
trt_branch = None
commit_timestamp = None
cuda_version = None
cublas_version = None
cudnn_version = None
#trt_version = None
class PerfResult(Data):
"""
Stores all relevant data from a perf run. Using class over dictionary
for easier documentation and reference to what values are stored
inside class. Can be more intuitive via key mappings with meta class.
"""
# Class attributes can be set in initialization
start_time = None
end_time = None
total_time = None
engine_build_time = None
engine_load_time = None
engine_file_size = None
throughput = None
perf_time = None
trt_peak_cpu_mem = None
trt_peak_gpu_mem = None
build_engine_allocated_cpu_mem = None
build_engine_allocated_gpu_mem = None
deserialize_engine_allocated_cpu_mem = None
deserialize_engine_allocated_gpu_mem = None
execution_context_allocated_cpu_mem = None
execution_context_allocated_gpu_mem = None
state = None
gpu_monitor = None
# Session specific information
sm_clk = None
mem_clk = None
gpu_idx = None
raw_result = None
command_str = None
# ARIA v3 Specific
network_name = None
framework = None
network_hash = None
flags = None
# Log File Location, Used for Reading Later
log_file = None
return_code = None