Note
Go to the end to download the full example code.
201: Evaluate DORT¶
It compares DORT to eager mode and onnxrt backend.
To run the script:
python _doc/examples/plot_torch_dort --help
Some helpers¶
import warnings
try:
with warnings.catch_warnings():
warnings.simplefilter("ignore")
import onnxruntime
has_cuda = "CUDAExecutionProvider" in onnxruntime.get_available_providers()
except ImportError:
print("onnxruntime not available.")
import sys
sys.exit(0)
import torch._dynamo
import contextlib
import itertools
import os
import gc
import platform
# import pickle
import pprint
import multiprocessing
import time
import cProfile
import pstats
import io
import logging
from pstats import SortKey
import numpy as np
import matplotlib.pyplot as plt
import pandas
import onnx
from onnx_array_api.profiling import profile2graph
import torch
from torch import nn
import torch.nn.functional as F
import experimental_experiment
from experimental_experiment.plotting.memory import memory_peak_plot
from experimental_experiment.ext_test_case import measure_time, get_figure
from experimental_experiment.args import get_parsed_args
from experimental_experiment.memory_peak import start_spying_on
from experimental_experiment.torch_models.training_helper import make_aot_ort
from tqdm import tqdm
has_cuda = has_cuda and torch.cuda.is_available()
logging.disable(logging.ERROR)
def system_info():
obs = {}
obs["processor"] = platform.processor()
obs["cores"] = multiprocessing.cpu_count()
try:
obs["cuda"] = 1 if torch.cuda.is_available() else 0
obs["cuda_count"] = torch.cuda.device_count()
obs["cuda_name"] = torch.cuda.get_device_name()
obs["cuda_capa"] = torch.cuda.get_device_capability()
except (RuntimeError, AssertionError):
# no cuda
pass
return obs
pprint.pprint(system_info())
{'cores': 20,
'cuda': 1,
'cuda_capa': (8, 9),
'cuda_count': 1,
'cuda_name': 'NVIDIA GeForce RTX 4060 Laptop GPU',
'processor': 'x86_64'}
Scripts arguments
script_args = get_parsed_args(
"plot_torch_dort",
description=__doc__,
scenarios={
"small": "small model to test",
"middle": "55Mb model",
"large": "1Gb model",
},
warmup=5,
repeat=5,
repeat1=(1, "repeat for the first iteration"),
maxtime=(
2,
"maximum time to run a model to measure the computation time, "
"it is 0.1 when scenario is small",
),
expose="scenarios,repeat,repeat1,warmup",
)
if script_args.scenario in (None, "small"):
script_args.maxtime = 0.1
print(f"scenario={script_args.scenario or 'small'}")
print(f"warmup={script_args.warmup}")
print(f"repeat={script_args.repeat}")
print(f"repeat1={script_args.repeat1}")
print(f"maxtime={script_args.maxtime}")
scenario=small
warmup=5
repeat=5
repeat1=1
maxtime=0.1
The model¶
A simple model to convert.
class MyModelClass(nn.Module):
def __init__(self, scenario=script_args.scenario):
super().__init__()
if scenario == "middle":
self.large = False
self.conv1 = nn.Conv2d(1, 32, 5)
# self.conv2 = nn.Conv2d(128, 16, 5)
self.fc1 = nn.Linear(30752, 1024)
self.fcs = []
self.fc2 = nn.Linear(1024, 128)
self.fc3 = nn.Linear(128, 10)
elif scenario in (None, "small"):
self.large = False
self.conv1 = nn.Conv2d(1, 16, 5)
# self.conv2 = nn.Conv2d(16, 16, 5)
self.fc1 = nn.Linear(144, 512)
self.fcs = []
self.fc2 = nn.Linear(512, 128)
self.fc3 = nn.Linear(128, 10)
elif scenario in (None, "large"):
self.large = True
self.conv1 = nn.Conv2d(1, 32, 5)
# self.conv2 = nn.Conv2d(128, 16, 5)
self.fc1 = nn.Linear(30752, 4096)
# torch script does not support loops.
self.fca = nn.Linear(4096, 4096)
self.fcb = nn.Linear(4096, 4096)
self.fcc = nn.Linear(4096, 4096)
self.fcd = nn.Linear(4096, 4096)
self.fce = nn.Linear(4096, 4096)
self.fcf = nn.Linear(4096, 4096)
self.fcg = nn.Linear(4096, 4096)
self.fch = nn.Linear(4096, 4096)
self.fci = nn.Linear(4096, 4096)
# end of the unfolded loop.
self.fc2 = nn.Linear(4096, 128)
self.fc3 = nn.Linear(128, 10)
else:
raise ValueError(f"Unsupported scenario={scenario!r}.")
def forward(self, x):
x = F.max_pool2d(F.relu(self.conv1(x)), (4, 4))
# x = F.max_pool2d(F.relu(self.conv2(x)), 2)
x = torch.flatten(x, 1)
x = F.relu(self.fc1(x))
if self.large:
# loop
x = F.relu(self.fca(x))
x = F.relu(self.fcb(x))
x = F.relu(self.fcc(x))
x = F.relu(self.fcd(x))
x = F.relu(self.fce(x))
x = F.relu(self.fcf(x))
x = F.relu(self.fcg(x))
x = F.relu(self.fch(x))
x = F.relu(self.fci(x))
# end of the loop
x = F.relu(self.fc2(x))
y = self.fc3(x)
return y
def create_model_and_input(scenario=script_args.scenario):
if scenario == "middle":
shape = [1, 1, 128, 128]
elif scenario in (None, "small"):
shape = [1, 1, 16, 16]
elif scenario == "large":
shape = [1, 1, 128, 128]
else:
raise ValueError(f"Unsupported scenario={scenario!r}.")
input_tensor = torch.rand(*shape).to(torch.float32)
model = MyModelClass(scenario=scenario)
assert model(input_tensor) is not None
return model, input_tensor
def torch_model_size(model):
size_model = 0
for param in model.parameters():
size = param.numel() * torch.finfo(param.data.dtype).bits / 8
size_model += size
return size_model
model, input_tensor = create_model_and_input()
model_size = torch_model_size(model)
print(f"model size={model_size / 2 ** 20} Mb")
model size=0.5401992797851562 Mb
Backends¶
def get_torch_eager(model, *args):
def my_compiler(gm, example_inputs):
return gm.forward
with contextlib.redirect_stdout(io.StringIO()):
with warnings.catch_warnings():
warnings.simplefilter("ignore")
optimized_mod = torch.compile(model, fullgraph=True, backend=my_compiler)
optimized_mod(*args)
return optimized_mod
def get_torch_default(model, *args):
with contextlib.redirect_stdout(io.StringIO()):
with warnings.catch_warnings():
warnings.simplefilter("ignore")
optimized_mod = torch.compile(model, fullgraph=True, mode="reduce-overhead")
optimized_mod(*args)
return optimized_mod
def get_torch_dort(model, *args):
with contextlib.redirect_stdout(io.StringIO()):
with warnings.catch_warnings():
warnings.simplefilter("ignore")
local_aot_ort, _ = make_aot_ort(dynamic=True, rewrite=True)
optimized_mod = torch.compile(model, backend=local_aot_ort, fullgraph=True)
optimized_mod(*args)
return optimized_mod
def get_torch_opti(model, *args):
with contextlib.redirect_stdout(io.StringIO()):
with warnings.catch_warnings():
warnings.simplefilter("ignore")
local_aot_ort, _ = make_aot_ort(dynamic=True, rewrite=True)
optimized_mod = torch.compile(model, backend=local_aot_ort, fullgraph=True)
optimized_mod(*args)
return optimized_mod
Let’s check they are working.
export_functions = [
get_torch_eager,
get_torch_default,
get_torch_dort,
# get_torch_opti,
]
exporters = {f.__name__.replace("get_", ""): f for f in export_functions}
supported_exporters = {}
for k, v in exporters.items():
print(f"run function {k}")
filename = f"plot_torch_dort_{k}.onnx"
torch._dynamo.reset()
model, input_tensor = create_model_and_input()
try:
v(model, input_tensor)
except Exception as e:
print(f"skipped due to {str(e)[:1000]}")
continue
supported_exporters[k] = v
del model
gc.collect()
time.sleep(1)
run function torch_eager
run function torch_default
run function torch_dort
Compile and Memory¶
def flatten(ps):
obs = ps["cpu"].to_dict(unit=2**20)
if "gpus" in ps:
for i, g in enumerate(ps["gpus"]):
for k, v in g.to_dict(unit=2**20).items():
obs[f"gpu{i}_{k}"] = v
return obs
data = []
for k, v in supported_exporters.items():
print(f"run compile for memory {k} on cpu")
filename = f"plot_torch_dort_{k}.onnx"
if has_cuda:
torch.cuda.set_device(0)
torch._dynamo.reset()
# CPU
model, input_tensor = create_model_and_input()
stat = start_spying_on(cuda=1 if has_cuda else 0)
v(model, input_tensor)
obs = flatten(stat.stop())
print("done.")
obs.update(dict(export=k, p="cpu"))
data.append(obs)
del model
gc.collect()
time.sleep(1)
if not has_cuda:
continue
if k in {"torch_default"}:
print(f"skip compile for memory {k} on cuda")
continue
torch._dynamo.reset()
# CUDA
model, input_tensor = create_model_and_input()
model = model.cuda()
input_tensor = input_tensor.cuda()
print(f"run compile for memory {k} on cuda")
stat = start_spying_on(cuda=1 if has_cuda else 0)
v(model, input_tensor)
obs = flatten(stat.stop())
print("done.")
obs.update(dict(export=k, p="cuda"))
data.append(obs)
del model
gc.collect()
time.sleep(1)
run compile for memory torch_eager on cpu
done.
run compile for memory torch_eager on cuda
done.
run compile for memory torch_default on cpu
done.
skip compile for memory torch_default on cuda
run compile for memory torch_dort on cpu
done.
run compile for memory torch_dort on cuda
done.
The result.
df1 = pandas.DataFrame(data)
df1.to_csv("plot_torch_dort_1_memory.csv", index=False)
df1.to_excel("plot_torch_dort_1_memory.xlsx", index=False)
print(df1)
for p in ["cpu", "cuda"]:
if not has_cuda and p == "cuda":
continue
ax = memory_peak_plot(
df1[df1["p"] == p],
key=("export",),
bars=[model_size * i / 2**20 for i in range(1, 5)],
suptitle=f"Memory Consumption of the Compilation on {p}\n"
f"model size={model_size / 2**20:1.0f} Mb",
)
get_figure(ax).savefig(f"plot_torch_dort_1_memory_{p}.png")
peak mean n begin end gpu0_peak gpu0_mean gpu0_n gpu0_begin gpu0_end export p
0 3946.113281 3943.472957 13 3940.308594 3946.113281 1045.617188 1045.617188 13 1045.617188 1045.617188 torch_eager cpu
1 6143.183594 5042.613445 358 3942.214844 6143.183594 1499.617188 1258.723333 358 1045.617188 1499.617188 torch_eager cuda
2 6145.734375 6143.777054 27 6143.675781 6145.734375 1499.617188 1499.617188 27 1499.617188 1499.617188 torch_default cpu
3 6144.332031 6143.607096 48 6143.757812 6140.191406 1499.617188 1499.617188 48 1499.617188 1499.617188 torch_dort cpu
4 6148.609375 6142.613664 51 6142.269531 6148.609375 1517.617188 1500.166207 51 1499.617188 1517.617188 torch_dort cuda
dort first iteration speed¶
data = []
for k, v in supported_exporters.items():
print(f"run dort cpu {k}: {script_args.repeat1}")
times = []
for _ in range(int(script_args.repeat1)):
model, input_tensor = create_model_and_input()
torch._dynamo.reset()
begin = time.perf_counter()
v(model, input_tensor)
duration = time.perf_counter() - begin
times.append(duration)
del model
gc.collect()
time.sleep(1)
print(f"done: {times[-1]}")
data.append(
dict(
export=k,
time=np.mean(times),
min=min(times),
max=max(times),
first=times[0],
last=times[-1],
std=np.std(times),
p="cpu",
)
)
if not has_cuda:
continue
if k in {"torch_dort", "torch_default"}:
print(f"skip dort cuda {k}: {script_args.repeat1}")
continue
print(f"run dort cuda {k}: {script_args.repeat1}")
times = []
for _ in range(int(script_args.repeat1)):
model, input_tensor = create_model_and_input()
model = model.cuda()
input_tensor = input_tensor.cuda()
torch._dynamo.reset()
begin = time.perf_counter()
v(model, input_tensor)
duration = time.perf_counter() - begin
times.append(duration)
del model
gc.collect()
time.sleep(1)
print(f"done: {times[-1]}")
data.append(
dict(
export=k,
time=np.mean(times),
min=min(times),
max=max(times),
first=times[0],
last=times[-1],
std=np.std(times),
p="cuda",
)
)
run dort cpu torch_eager: 1
done: 0.047091833999729715
run dort cuda torch_eager: 1
done: 0.07029420699836919
run dort cpu torch_default: 1
done: 0.2517179459973704
skip dort cuda torch_default: 1
run dort cpu torch_dort: 1
done: 0.3630555009949603
skip dort cuda torch_dort: 1
The result.
df1 = pandas.DataFrame(data)
df1.to_csv("plot_torch_dort_1_time.csv", index=False)
df1.to_excel("plot_torch_dort_1_time.xlsx", index=False)
print(df1)
fig, ax = plt.subplots(1, 1)
dfi = df1[["export", "p", "time", "std"]].set_index(["export", "p"])
dfi["time"].plot.bar(ax=ax, title="Compilation time", yerr=dfi["std"], rot=30)
fig.tight_layout()
fig.savefig("plot_torch_dort_1_time.png")
export time min max first last std p
0 torch_eager 0.047092 0.047092 0.047092 0.047092 0.047092 0.0 cpu
1 torch_eager 0.070294 0.070294 0.070294 0.070294 0.070294 0.0 cuda
2 torch_default 0.251718 0.251718 0.251718 0.251718 0.251718 0.0 cpu
3 torch_dort 0.363056 0.363056 0.363056 0.363056 0.363056 0.0 cpu
Compilation Profiling¶
def clean_text(text):
pathes = [
os.path.abspath(os.path.normpath(os.path.join(os.path.dirname(torch.__file__), ".."))),
os.path.abspath(os.path.normpath(os.path.join(os.path.dirname(onnx.__file__), ".."))),
os.path.abspath(
os.path.normpath(
os.path.join(os.path.dirname(experimental_experiment.__file__), "..")
)
),
]
for p in pathes:
text = text.replace(p, "")
text = text.replace("experimental_experiment", "experimental_experiment".upper())
return text
def profile_function(name, export_function, with_args=True, verbose=False, suffix="export"):
if verbose:
print(f"profile {name}: {export_function}")
if with_args:
model, input_tensor = create_model_and_input()
pr = cProfile.Profile()
pr.enable()
for _ in range(int(script_args.repeat1)):
export_function(model, input_tensor)
pr.disable()
else:
pr = cProfile.Profile()
pr.enable()
for _ in range(int(script_args.repeat1)):
export_function()
pr.disable()
s = io.StringIO()
sortby = SortKey.CUMULATIVE
ps = pstats.Stats(pr, stream=s).sort_stats(sortby)
ps.print_stats()
# with open(f"plot_torch_dort_profile_{name}_{suffix}.pickle", "wb") as f:
# pickle.dump(ps, f)
raw = s.getvalue()
text = "\n".join(raw.split("\n")[:200])
if verbose:
print(text)
with open(f"plot_torch_dort_profile_{name}_{suffix}.txt", "w") as f:
f.write(raw)
root, nodes = profile2graph(ps, clean_text=clean_text)
text = root.to_text()
with open(f"plot_torch_dort_profile_{name}_{suffix}_h.txt", "w") as f:
f.write(text)
if verbose:
print("done.")
model, input_tensor = create_model_and_input()
def function_to_profile(model=model, input_tensor=input_tensor):
return get_torch_dort(model, input_tensor)
profile_function("dort", function_to_profile, verbose=True, suffix="1")
profile dort: <function function_to_profile at 0x7f622b732b60>
1375777 function calls (1344911 primitive calls) in 0.830 seconds
Ordered by: cumulative time
ncalls tottime percall cumtime percall filename:lineno(function)
16/5 0.000 0.000 0.517 0.103 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/lazy.py:166(realize_and_forward)
103/35 0.000 0.000 0.486 0.014 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/symbolic_convert.py:935(step)
1 0.000 0.000 0.419 0.419 /home/xadupre/github/experimental-experiment/experimental_experiment/torch_models/training_helper.py:7(make_aot_ort)
1 0.000 0.000 0.418 0.418 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/onnxruntime.py:763(__init__)
1 0.000 0.000 0.363 0.363 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/_exporter_legacy.py:297(__init__)
1 0.000 0.000 0.244 0.244 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/_exporter_legacy.py:90(__init__)
1 0.001 0.001 0.244 0.244 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/_exporter_legacy.py:114(_initiate_registry_from_torchlib)
1 0.004 0.004 0.239 0.239 /home/xadupre/github/onnxscript/onnxscript/_framework_apis/torch_2_5.py:99(get_torchlib_ops)
184 0.001 0.000 0.234 0.001 /home/xadupre/github/onnxscript/onnxscript/values.py:640(function_ir)
1 0.001 0.001 0.119 0.119 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/fx/decomposition_table.py:73(create_onnx_friendly_decomposition_table)
128 0.018 0.000 0.102 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/functional_tensor.py:352(__torch_dispatch__)
2 0.020 0.010 0.099 0.050 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/fx/decomposition_table.py:14(_create_onnx_supports_op_overload_table)
184 0.001 0.000 0.091 0.000 /home/xadupre/github/onnxscript/onnxscript/_internal/ast_utils.py:16(get_src_and_ast)
184 0.000 0.000 0.073 0.000 /home/xadupre/github/onnxscript/onnxscript/converter.py:1463(translate_function_signature)
184 0.005 0.000 0.073 0.000 /home/xadupre/github/onnxscript/onnxscript/converter.py:1378(_translate_function_signature_common)
1 0.000 0.000 0.070 0.070 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/export/decomp_utils.py:125(items)
1 0.000 0.000 0.070 0.070 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/export/decomp_utils.py:142(_materialize_if_needed)
1 0.000 0.000 0.070 0.070 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/export/decomp_utils.py:129(materialize)
1 0.000 0.000 0.069 0.069 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_export/utils.py:1088(_collect_all_valid_cia_ops)
23 0.001 0.000 0.069 0.003 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_export/utils.py:1071(_collect_all_valid_cia_ops_for_namespace)
184 0.005 0.000 0.065 0.000 /usr/lib/python3.12/inspect.py:1606(getclosurevars)
185 0.000 0.000 0.065 0.000 /usr/lib/python3.12/inspect.py:1279(getsource)
185 0.005 0.000 0.064 0.000 /usr/lib/python3.12/inspect.py:1258(getsourcelines)
23 0.024 0.001 0.063 0.003 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_export/utils.py:1006(_materialize_cpp_cia_ops)
12/6 0.000 0.000 0.061 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/symbolic_convert.py:653(wrapper)
297/276 0.001 0.000 0.060 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/experimental/proxy_tensor.py:1329(__torch_dispatch__)
12/6 0.000 0.000 0.060 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/symbolic_convert.py:2341(CALL)
12/6 0.000 0.000 0.060 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/symbolic_convert.py:2300(_call)
12/6 0.000 0.000 0.059 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/symbolic_convert.py:878(call_function)
8459 0.023 0.000 0.059 0.000 /usr/lib/python3.12/dis.py:434(_get_instructions_bytes)
1 0.000 0.000 0.057 0.057 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_functorch/partitioners.py:1702(min_cut_rematerialization_partition)
156757/151855 0.024 0.000 0.055 0.000 {built-in method builtins.isinstance}
69/54 0.002 0.000 0.053 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/experimental/proxy_tensor.py:762(proxy_call)
184 0.014 0.000 0.051 0.000 /usr/lib/python3.12/inspect.py:1239(getblock)
516/511 0.003 0.000 0.050 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:1788(dispatch)
16630 0.006 0.000 0.047 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/_exporter_legacy.py:199(is_registered_op)
150 0.001 0.000 0.046 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:1348(_cached_dispatch_impl)
28 0.000 0.000 0.043 0.002 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/symbolic_convert.py:1799(LOAD_ATTR)
7569/1816 0.009 0.000 0.043 0.000 /home/xadupre/github/onnxscript/onnxscript/type_annotation.py:131(is_value_type)
35 0.000 0.000 0.042 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/interpreter.py:290(call_function)
16645 0.009 0.000 0.041 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/_exporter_legacy.py:177(get_op_functions)
28 0.000 0.000 0.041 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/symbolic_convert.py:1792(_load_attr)
4 0.000 0.000 0.041 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/nn_module.py:851(call_function)
5/4 0.000 0.000 0.039 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/functions.py:293(call_function)
5/4 0.000 0.000 0.039 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/functions.py:112(call_function)
5/4 0.000 0.000 0.037 0.009 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/symbolic_convert.py:900(inline_user_function_return)
5/4 0.000 0.000 0.037 0.009 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/symbolic_convert.py:3071(inline_call)
5/4 0.000 0.000 0.037 0.009 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/symbolic_convert.py:3108(inline_call_)
14 0.000 0.000 0.035 0.002 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:2594(from_tensor)
533/436 0.000 0.000 0.034 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_ops.py:721(__call__)
27/14 0.000 0.000 0.032 0.002 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:331(from_real_tensor)
1565 0.032 0.000 0.032 0.000 {built-in method builtins.compile}
26790 0.017 0.000 0.032 0.000 /usr/lib/python3.12/tokenize.py:569(_generate_tokens_from_c_tokenizer)
13 0.000 0.000 0.030 0.002 {built-in method torch._to_functional_tensor}
1022 0.001 0.000 0.028 0.000 /home/xadupre/github/onnxscript/onnxscript/type_annotation.py:172(is_valid_type)
420/380 0.001 0.000 0.027 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/experimental/proxy_tensor.py:1231(__torch_function__)
314 0.001 0.000 0.026 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/utils/_pytree.py:1152(tree_map_only)
8 0.000 0.000 0.026 0.003 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph.py:1569(python_code)
486 0.002 0.000 0.026 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph.py:1111(create_node)
141 0.000 0.000 0.026 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/lazy.py:61(realize)
75 0.000 0.000 0.026 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/base.py:445(build)
4/2 0.000 0.000 0.025 0.013 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/fx/fx_onnx_interpreter.py:393(run_node)
2 0.000 0.000 0.025 0.013 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/fx/fx_onnx_interpreter.py:556(placeholder)
2 0.000 0.000 0.025 0.013 /home/xadupre/github/onnxscript/onnxscript/function_libs/torch_lib/graph_building/_graph_building_torch.py:613(add_input)
31/29 0.000 0.000 0.025 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builtin.py:987(call_function)
75 0.000 0.000 0.025 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builder.py:369(__call__)
1308/1164 0.001 0.000 0.025 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/base.py:184(__instancecheck__)
29 0.000 0.000 0.025 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builtin.py:850(builtin_dispatch)
28 0.000 0.000 0.025 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builtin.py:770(call_self_handler)
28 0.000 0.000 0.024 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builtin.py:1632(call_getattr)
4 0.001 0.000 0.024 0.006 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_functorch/partitioners.py:157(_extract_graph_with_inputs_outputs)
40 0.001 0.000 0.024 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builder.py:528(_wrap)
90903 0.023 0.000 0.024 0.000 {built-in method builtins.getattr}
20/2 0.000 0.000 0.024 0.012 /home/xadupre/github/onnxscript/onnxscript/function_libs/torch_lib/graph_building/_graph_building_torch.py:270(_wrap_torch_value_to_tensor)
2 0.024 0.012 0.024 0.012 /home/xadupre/github/onnxscript/onnxscript/function_libs/torch_lib/graph_building/_graph_building_torch.py:199(dtype)
2 0.024 0.012 0.024 0.012 /home/xadupre/github/onnxscript/onnxscript/function_libs/torch_lib/graph_building/_graph_building_torch.py:170(shape)
51/35 0.000 0.000 0.023 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/nn_module.py:1015(var_getattr)
27292/27100 0.007 0.000 0.023 0.000 {built-in method builtins.next}
36 0.000 0.000 0.023 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/lazy.py:20(realize)
35/19 0.000 0.000 0.023 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/user_defined.py:1019(var_getattr)
8 0.000 0.000 0.023 0.003 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph.py:1646(_python_code)
8 0.002 0.000 0.022 0.003 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph.py:408(_gen_python_code)
327 0.001 0.000 0.022 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph.py:1500(node_copy)
184 0.000 0.000 0.021 0.000 /usr/lib/python3.12/ast.py:34(parse)
36326 0.012 0.000 0.021 0.000 {method 'get' of 'dict' objects}
31 0.000 0.000 0.020 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/meta_utils.py:1794(__call__)
9949 0.010 0.000 0.020 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/fx/registration.py:55(from_qualified_name)
31 0.000 0.000 0.020 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/meta_utils.py:836(meta_tensor)
1 0.000 0.000 0.020 0.020 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_functorch/partitioners.py:289(_extract_fwd_bwd_modules)
136 0.000 0.000 0.019 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:1722(_output_from_cache_entry)
9 0.000 0.000 0.019 0.002 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builder.py:1533(wrap_tensor)
18 0.000 0.000 0.019 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/utils.py:2072(wrap_fake_exception)
140 0.002 0.000 0.019 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:1656(_get_output_tensor_from_cache_entry)
150 0.001 0.000 0.018 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:1391(_cache_key)
7569 0.005 0.000 0.018 0.000 /home/xadupre/github/onnxscript/onnxscript/type_annotation.py:123(_is_tensor_type)
12/9 0.000 0.000 0.018 0.002 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/nn/functional.py:1693(relu)
8979/4158 0.010 0.000 0.017 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/node.py:899(map_aggregate)
4938 0.006 0.000 0.017 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/node.py:869(__setattr__)
9 0.000 0.000 0.017 0.002 {built-in method torch.relu}
565/153 0.002 0.000 0.017 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:1467(_prep_args_for_hash)
3 0.000 0.000 0.016 0.005 /home/xadupre/github/onnxscript/onnxscript/rewriter/__init__.py:28(rewrite)
794 0.000 0.000 0.016 0.000 /home/xadupre/github/onnxscript/onnxscript/type_annotation.py:168(is_attr_type)
31 0.001 0.000 0.016 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/autograd/grad_mode.py:273(__exit__)
16645 0.007 0.000 0.016 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/fx/registration.py:45(from_name_parts)
1377 0.003 0.000 0.016 0.000 /home/xadupre/github/onnxscript/onnxscript/converter.py:451(_eval_constant_expr)
102 0.001 0.000 0.016 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/proxy.py:209(create_proxy)
26607 0.008 0.000 0.015 0.000 /usr/lib/python3.12/collections/__init__.py:447(_make)
4 0.000 0.000 0.015 0.004 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph_module.py:908(print_readable)
4 0.000 0.000 0.015 0.004 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph_module.py:297(_print_readable)
12/9 0.000 0.000 0.015 0.002 {built-in method torch._C._nn.linear}
2917 0.002 0.000 0.015 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/node.py:890(map_arg)
4 0.000 0.000 0.014 0.003 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_functorch/aot_autograd.py:500(convert)
539 0.001 0.000 0.014 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/utils/_pytree.py:874(tree_flatten)
9 0.000 0.000 0.014 0.002 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/torch.py:876(call_function)
64175 0.014 0.000 0.014 0.000 {method 'split' of 'str' objects}
9 0.000 0.000 0.013 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builder.py:2833(wrap_to_fake_tensor_and_record)
2434/12 0.001 0.000 0.013 0.001 /home/xadupre/github/onnxscript/onnxscript/ir/serde.py:94(wrapper)
17035 0.011 0.000 0.013 0.000 /usr/lib/python3.12/dis.py:623(_unpack_opargs)
301 0.003 0.000 0.013 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph.py:634(emit_node)
4 0.000 0.000 0.013 0.003 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph_module.py:792(recompile)
1860/539 0.003 0.000 0.013 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/utils/_pytree.py:882(helper)
9 0.000 0.000 0.013 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builder.py:2870(<lambda>)
7 0.000 0.000 0.013 0.002 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph_module.py:437(__init__)
61 0.000 0.000 0.012 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/experimental/proxy_tensor.py:594(track_tensor_tree)
87/71 0.000 0.000 0.012 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/nn/modules/module.py:1932(__setattr__)
8045 0.004 0.000 0.012 0.000 /home/xadupre/github/onnxscript/onnxscript/type_annotation.py:70(_remove_annotation)
4 0.000 0.000 0.012 0.003 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/overrides.py:1669(handle_torch_function)
495 0.002 0.000 0.012 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/node.py:380(prepend)
9 0.000 0.000 0.012 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph_module.py:548(graph)
76/61 0.000 0.000 0.012 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/experimental/proxy_tensor.py:616(wrap_with_proxy)
1643/1548 0.001 0.000 0.012 0.000 /usr/lib/python3.12/contextlib.py:132(__enter__)
1 0.000 0.000 0.011 0.011 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/onnxruntime.py:1099(compile)
191 0.004 0.000 0.011 0.000 /usr/lib/python3.12/dis.py:647(findlabels)
1 0.000 0.000 0.011 0.011 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/passes/infra/partitioner.py:384(partition_and_fuse)
104 0.001 0.000 0.011 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/proxy.py:143(create_node)
1 0.000 0.000 0.011 0.011 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_functorch/partitioners.py:1465(choose_saved_values_set)
1 0.001 0.001 0.011 0.011 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_functorch/partitioners.py:814(solve_min_cut)
1 0.000 0.000 0.010 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_functorch/partitioners.py:545(reordering_to_mimic_autograd_engine)
1 0.000 0.000 0.010 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/passes/infra/partitioner.py:297(fuse_partitions)
1 0.000 0.000 0.010 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/passes/utils/fuser_utils.py:250(fuse_by_partitions)
62/2 0.001 0.000 0.010 0.005 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/diagnostics/infra/decorator.py:66(wrapper)
2 0.000 0.000 0.010 0.005 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/fx/_pass.py:240(run)
448 0.000 0.000 0.010 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/utils/_pytree.py:1101(wrapped)
23 0.010 0.000 0.010 0.000 {built-in method torch._C._dispatch_get_registrations_for_dispatch_key}
1 0.000 0.000 0.010 0.010 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/fx/passes/type_promotion.py:1696(_run)
99867/99754 0.009 0.000 0.009 0.000 {built-in method builtins.len}
7 0.000 0.000 0.009 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph.py:1815(eliminate_dead_code)
18 0.000 0.000 0.009 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builder.py:2151(wrap_fx_proxy)
10045 0.005 0.000 0.009 0.000 /usr/lib/python3.12/typing.py:2340(get_origin)
31 0.001 0.000 0.009 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/autograd/grad_mode.py:269(__enter__)
18 0.000 0.000 0.009 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builder.py:2224(wrap_fx_proxy_cls)
45/40 0.001 0.000 0.009 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_logging/_internal.py:1156(trace_structured)
69/54 0.000 0.000 0.008 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/experimental/proxy_tensor.py:2248(maybe_handle_decomp)
1 0.001 0.001 0.008 0.008 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_functorch/compile_utils.py:42(fx_graph_cse)
182 0.002 0.000 0.008 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:954(_flatten_into)
12 0.002 0.000 0.008 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/graph.py:1706(lint)
5 0.000 0.000 0.008 0.002 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_prims_common/wrappers.py:270(_fn)
27 0.000 0.000 0.008 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/onnx/_internal/fx/passes/type_promotion.py:1601(run_node)
3 0.000 0.000 0.008 0.003 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_decomp/__init__.py:155(_fn)
3 0.000 0.000 0.008 0.003 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_decomp/decompositions.py:220(threshold_backward)
55/13 0.000 0.000 0.008 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/interpreter.py:212(run_node)
185 0.001 0.000 0.008 0.000 /usr/lib/python3.12/inspect.py:1070(findsource)
5 0.000 0.000 0.008 0.002 /home/xadupre/github/onnxscript/onnxscript/ir/serde.py:461(deserialize_model)
9 0.000 0.000 0.008 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/builder.py:2308(_wrap_fx_proxy)
40536/39468 0.007 0.000 0.008 0.000 {built-in method builtins.hash}
1643/1548 0.001 0.000 0.008 0.000 /usr/lib/python3.12/contextlib.py:141(__exit__)
5 0.000 0.000 0.008 0.002 /home/xadupre/github/onnxscript/onnxscript/ir/serde.py:551(_deserialize_graph)
143/29 0.000 0.000 0.008 0.000 /home/xadupre/github/onnxscript/onnxscript/_legacy_ir/visitor.py:799(visit_node)
74 0.000 0.000 0.007 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/experimental/proxy_tensor.py:487(set_meta)
249 0.002 0.000 0.007 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:678(__new__)
43757 0.007 0.000 0.007 0.000 {built-in method __new__ of type object at 0xa20960}
812/42 0.001 0.000 0.007 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/utils/_stats.py:15(wrapper)
496 0.002 0.000 0.007 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/node.py:243(__init__)
14 0.000 0.000 0.007 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:1974(_dispatch_impl)
1 0.000 0.000 0.007 0.007 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/variables/functions.py:336(call_function)
64 0.000 0.000 0.007 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/guards.py:1252(get_guard_manager)
1 0.000 0.000 0.007 0.007 <class 'networkx.utils.decorators.argmap'> compilation 4:1(argmap_minimum_cut_1)
6/1 0.000 0.000 0.007 0.007 /home/xadupre/vv/this312/lib/python3.12/site-packages/networkx/utils/backends.py:959(__call__)
512/42 0.001 0.000 0.007 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/fake_tensor.py:1263(__torch_dispatch__)
31 0.001 0.000 0.007 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_subclasses/meta_utils.py:260(describe_tensor)
1 0.000 0.000 0.007 0.007 /home/xadupre/vv/this312/lib/python3.12/site-packages/networkx/algorithms/flow/maxflow.py:307(minimum_cut)
3 0.000 0.000 0.007 0.002 /home/xadupre/github/onnxscript/onnxscript/rewriter/pattern.py:1500(apply_to_model)
1513/1507 0.001 0.000 0.007 0.000 {built-in method builtins.any}
9 0.000 0.000 0.007 0.001 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_dynamo/utils.py:2468(get_fake_value)
625 0.001 0.000 0.007 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/node.py:576(__update_args_kwargs)
17063 0.004 0.000 0.007 0.000 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/_ops.py:729(__hash__)
1722 0.006 0.000 0.007 0.000 {built-in method builtins.eval}
46 0.000 0.000 0.007 0.000 /home/xadupre/github/onnxscript/onnxscript/optimizer/_legacy/constant_folding.py:165(process_node)
30947 0.006 0.000 0.006 0.000 {built-in method builtins.hasattr}
3 0.000 0.000 0.006 0.002 /home/xadupre/github/onnxscript/onnxscript/rewriter/pattern.py:1467(_apply_to_graph_or_function)
5 0.000 0.000 0.006 0.001 /home/xadupre/github/onnxscript/onnxscript/optimizer/_remove_unused_function.py:64(remove_unused_functions)
1 0.000 0.000 0.006 0.006 <class 'networkx.utils.decorators.argmap'> compilation 8:1(argmap_preflow_push_5)
1 0.000 0.000 0.006 0.006 /home/xadupre/vv/this312/lib/python3.12/site-packages/networkx/algorithms/flow/preflowpush.py:291(preflow_push)
1 0.000 0.000 0.006 0.006 /home/xadupre/vv/this312/lib/python3.12/site-packages/torch/fx/passes/utils/fuser_utils.py:95(fuse_as_graphmodule)
1 0.001 0.001 0.006 0.006 /home/xadupre/vv/this312/lib/python3.12/site-packages/networkx/algorithms/flow/preflowpush.py:22(preflow_push_impl)
done.
Benchmark exported models with ORT¶
def benchmark(shape):
data = []
data_mem_first_run = []
data_mem_run = []
confs = list(
itertools.product(
export_functions,
["CPU", "CUDA"],
)
)
loop = tqdm(confs)
print(f"number of experiments: {len(loop)}")
for export_fct, p in loop:
name = export_fct.__name__.replace("get_torch_", "")
obs = {} # system_info()
obs["name"] = name
obs["compute"] = p
obs["export"] = name
model, input_tensor = create_model_and_input()
if p == "CUDA":
if not has_cuda:
continue
model = model.cuda()
input_tensor = input_tensor.cuda()
try:
exported_model = export_fct(model, input_tensor)
except torch._dynamo.exc.BackendCompilerFailed as e:
# Triton only supports devices of CUDA Capability >= 7.0,
# but your device is of CUDA capability 6.1
obs["error"] = str(e)
data.append(obs)
continue
def call_model(
export_fct=export_fct,
exported_model=exported_model,
input_tensor=input_tensor,
):
res = exported_model(input_tensor).sum()
return res
stat = start_spying_on(cuda=1 if has_cuda else 0)
try:
call_model()
except Exception as e:
loop.set_description(f"ERROR-run: {name} {e}")
obs.update({"error": e, "step": "load"})
data.append(obs)
stat.stop()
continue
memobs = flatten(stat.stop())
memobs.update(obs)
data_mem_first_run.append(memobs)
# memory consumption
stat = start_spying_on(cuda=1 if has_cuda else 0)
for _ in range(0, script_args.warmup):
call_model()
memobs = flatten(stat.stop())
memobs.update(obs)
data_mem_run.append(memobs)
obs.update(
measure_time(
call_model,
max_time=script_args.maxtime,
repeat=script_args.repeat,
number=1,
)
)
profile_function(name, call_model, with_args=False, suffix=f"run_{p}")
loop.set_description(f"{obs['average']} {name} {p}")
data.append(obs)
del model
del exported_model
gc.collect()
time.sleep(1)
df = pandas.DataFrame(data)
df.to_csv("plot_torch_dort_ort_time.csv", index=False)
df.to_excel("plot_torch_dort_ort_time.xlsx", index=False)
dfmemr = pandas.DataFrame(data_mem_run)
dfmemr.to_csv("plot_torch_dort_ort_run_mem.csv", index=False)
dfmemr.to_excel("plot_torch_dort_ort_run_mem.xlsx", index=False)
dfmemfr = pandas.DataFrame(data_mem_first_run)
dfmemfr.to_csv("plot_torch_dort_ort_first_run_mem.csv", index=False)
dfmemfr.to_excel("plot_torch_dort_ort_first_run_mem.xlsx", index=False)
return df, dfmemfr, dfmemr
df, dfmemfr, dfmemr = benchmark(list(input_tensor.shape))
print(df)
0%| | 0/6 [00:00<?, ?it/s]number of experiments: 6
0.003408030238157759 eager CPU: 0%| | 0/6 [00:00<?, ?it/s]
0.003408030238157759 eager CPU: 17%|█▋ | 1/6 [00:01<00:09, 1.89s/it]
0.000306338890255371 eager CUDA: 17%|█▋ | 1/6 [00:02<00:09, 1.89s/it]
0.000306338890255371 eager CUDA: 33%|███▎ | 2/6 [00:03<00:07, 1.81s/it]
0.003683345764561547 default CPU: 33%|███▎ | 2/6 [00:04<00:07, 1.81s/it]
0.003683345764561547 default CPU: 50%|█████ | 3/6 [00:05<00:05, 1.99s/it]
0.00023473138890428664 default CUDA: 50%|█████ | 3/6 [00:07<00:05, 1.99s/it]
0.00023473138890428664 default CUDA: 67%|██████▋ | 4/6 [00:08<00:04, 2.30s/it]
0.0005048716431847082 dort CPU: 67%|██████▋ | 4/6 [00:09<00:04, 2.30s/it]
0.0005048716431847082 dort CPU: 83%|████████▎ | 5/6 [00:10<00:02, 2.21s/it]
0.0006035198889305784 dort CUDA: 83%|████████▎ | 5/6 [00:11<00:02, 2.21s/it]
0.0006035198889305784 dort CUDA: 100%|██████████| 6/6 [00:12<00:00, 2.21s/it]
0.0006035198889305784 dort CUDA: 100%|██████████| 6/6 [00:12<00:00, 2.15s/it]
name compute export average deviation min_exec max_exec repeat number ttime context_size warmup_time
0 eager CPU eager 0.003408 0.000387 0.002068 0.003851 1 42.0 0.143137 64 0.001769
1 eager CUDA eager 0.000306 0.000016 0.000294 0.000450 1 483.0 0.147962 64 0.000975
2 default CPU default 0.003683 0.000530 0.001074 0.003976 1 51.0 0.187851 64 0.001776
3 default CUDA default 0.000235 0.000042 0.000207 0.000635 1 468.0 0.109854 64 0.001223
4 dort CPU dort 0.000505 0.000238 0.000335 0.001100 1 213.0 0.107538 64 0.001846
5 dort CUDA dort 0.000604 0.000027 0.000568 0.000868 1 243.0 0.146655 64 0.001559
Other view
def view_time(df, title, suffix="time"):
piv = pandas.pivot_table(df, index="export", columns=["compute"], values="average")
print(piv)
piv.to_csv(f"plot_torch_dort_{suffix}_compute.csv")
piv.to_excel(f"plot_torch_dort_{suffix}_compute.xlsx")
piv_cpu = pandas.pivot_table(
df[df.compute == "CPU"],
index="export",
columns=["compute"],
values="average",
)
fig, ax = plt.subplots(1, 2, figsize=(12, 4))
fig.suptitle(title)
piv_cpu.plot.barh(ax=ax[0], title="CPU", logx=True)
if has_cuda:
piv_gpu = pandas.pivot_table(
df[df.compute == "CUDA"],
index="export",
columns=["compute"],
values="average",
)
piv_gpu.plot.barh(ax=ax[1], title="CUDA", logx=True)
fig.tight_layout()
fig.savefig(f"plot_torch_dort_{suffix}.png")
return ax
view_time(df, "Compares processing time on backends")
compute CPU CUDA
export
default 0.003683 0.000235
dort 0.000505 0.000604
eager 0.003408 0.000306
array([<Axes: title={'center': 'CPU'}, ylabel='export'>,
<Axes: title={'center': 'CUDA'}, ylabel='export'>], dtype=object)
Memory First Running Time (ORT)¶
for compute in ["CPU", "CUDA"]:
if not has_cuda and compute == "CUDA":
continue
ax = memory_peak_plot(
dfmemfr[dfmemfr.compute == compute],
("export",),
suptitle=f"Memory Consumption of backend, first running time"
f"\nrunning on {compute}",
bars=[model_size * i / 2**20 for i in range(1, 3)],
figsize=(18, 6),
)
get_figure(ax).savefig(f"plot_torch_dort_first_run_mem_{compute}.png")
Memory Running Time (ORT)¶
for compute in ["CPU", "CUDA"]:
if not has_cuda and compute == "CUDA":
continue
ax = memory_peak_plot(
dfmemr[dfmemr.compute == compute],
("export",),
suptitle=f"Memory Consumption of backens, running time\nrunning on {compute}",
bars=[model_size * i / 2**20 for i in range(1, 3)],
figsize=(18, 6),
)
get_figure(ax).savefig(f"plot_torch_dort_run_mem_{compute}.png")
Total running time of the script: (0 minutes 49.816 seconds)
Related examples
201: Evaluate different ways to export a torch model to ONNX
201: Evaluate different ways to export a torch model to ONNX
