from __future__ import annotations
import argparse
import collections
import contextlib
import copy
import csv
import dataclasses
import functools
import importlib
import itertools
import logging
import os
import pathlib
import random
import shutil
import signal
import subprocess
import sys
import time
import warnings
from contextlib import contextmanager
from typing import Any, Callable, Mapping, NamedTuple, Optional, Tuple, Type
from unittest.mock import MagicMock
import numpy as np
import pandas as pd
import psutil
import torch
import torch._dynamo
import torch._dynamo.utils
import torch.distributed
import torch.fx._pytree as fx_pytree
import torch.multiprocessing as mp
from scipy.stats import gmean, ttest_ind
from torch._dynamo.profiler import fx_insert_profiling, Profiler
from torch._dynamo.testing import dummy_fx_compile, format_speedup, same
from torch._dynamo.utils import clone_inputs, graph_break_reasons
from torch._functorch.aot_autograd import set_model_name
from torch._inductor import config as inductor_config
from torch._inductor.utils import fresh_inductor_cache
from torch._subclasses.fake_tensor import FakeTensorMode
from torch.utils import _pytree as pytree
from torch.utils._pytree import tree_map, tree_map_only
from tqdm.auto import tqdm, trange
try:
import torch_npu
is_npu_available = torch_npu.npu.is_available()
from npu_support import patch_model
from profiler import NPUProfiler
except ImportError:
is_npu_available = False
from profiler import CUDAProfiler
from benchmark.userbenchmark.dynamo.dynamobench.common import (
load_model_from_path, Stats, randomize_input, speedup_experiment_ds,
baselines, null_experiment, DummyGradScaler, cast_to_bf16, cast_to_fp16,
cast_to_fp64, cast_to_fp32, maybe_fresh_cache, maybe_init_distributed,
)
log = logging.getLogger(__name__)
torch.backends.cuda.matmul.allow_tf32 = True
os.environ["KINETO_LOG_LEVEL"] = "5"
current_name = ""
current_device = ""
current_batch_size = None
output_filename = None
MAX_DOWNLOAD_ATTEMPTS = 5
class PathManager:
MAX_PATH_LENGTH = 4096
MAX_FILE_NAME_LENGTH = 255
DATA_FILE_AUTHORITY = 0o640
DATA_DIR_AUTHORITY = 0o750
@classmethod
def check_path_owner_consistent(cls, path: str):
if not os.path.exists(path):
msg = f"The path does not exist: {path}"
raise RuntimeError(msg)
if os.stat(path).st_uid != os.getuid():
warnings.warn(f"Warning: The {path} owner does not match the current user.")
@classmethod
def create_file_safety(cls, path: str):
msg = f"Failed to create file: {path}"
if os.path.islink(path):
raise RuntimeError(msg)
if os.path.exists(path):
return
try:
path = os.path.realpath(path)
os.close(os.open(path, os.O_WRONLY | os.O_CREAT, cls.DATA_FILE_AUTHORITY))
except Exception as err:
raise RuntimeError(msg) from err
@classmethod
def check_directory_path_readable(cls, path):
cls.check_path_owner_consistent(path)
if os.path.islink(path):
msg = f"Invalid path is a soft chain: {path}"
raise RuntimeError(msg)
if not os.access(path, os.R_OK):
msg = f"The path permission check failed: {path}"
raise RuntimeError(msg)
@classmethod
def check_directory_path_writeable(cls, path):
cls.check_path_owner_consistent(path)
if os.path.islink(path):
msg = f"Invalid path is a soft chain: {path}"
raise RuntimeError(msg)
if not os.access(path, os.W_OK):
msg = f"The path permission check failed: {path}"
raise RuntimeError(msg)
callbacks = []
def register_callback(callback):
callbacks.append(callback)
def model_specified_by_path(path_and_class_str):
return ":" in path_and_class_str
def output_csv(filename, headers, row):
abspath = os.path.abspath(filename)
if os.path.exists(filename):
PathManager.check_directory_path_readable(abspath)
with open(filename) as fd:
lines = list(csv.reader(fd)) or [[]]
if headers and len(headers) > len(lines[0]):
lines[0] = headers
else:
headers = lines[0]
else:
lines = [headers]
lines.append([(f"{x:.6f}" if isinstance(x, float) else x) for x in row])
PathManager.create_file_safety(abspath)
PathManager.check_directory_path_writeable(abspath)
with open(filename, "w") as fd:
writer = csv.writer(fd, lineterminator="\n")
for line in lines:
writer.writerow(list(line) + ["0"] * (len(headers) - len(line)))
def nothing(f):
return f
@functools.lru_cache(None)
def patch_torch_manual_seed():
"""Make torch manual seed deterministic. Helps with accuracy testing."""
def deterministic_torch_manual_seed(*args, **kwargs):
from torch._C import default_generator
seed = 1337
if not torch.cuda._is_in_bad_fork():
torch.cuda.manual_seed_all(seed)
if is_npu_available:
torch_npu.npu.manual_seed_all(seed)
return default_generator.manual_seed(seed)
torch.manual_seed = deterministic_torch_manual_seed
def synchronize():
pass
def timed(
model,
model_iter_fn,
example_inputs,
times=1,
return_result=False,
collect_outputs=False,
):
synchronize()
time_total = 0
for _ in range(times):
reset_rng_state()
t_iter_begin = time.perf_counter()
result = model_iter_fn(model, example_inputs, collect_outputs=collect_outputs)
t_iter_end = time.perf_counter()
time_total += t_iter_end - t_iter_begin
t_0 = time.perf_counter()
synchronize()
t_1 = time.perf_counter()
time_total += t_1 - t_0
return (time_total, result) if return_result else time_total
def speedup_experiment(args, model_iter_fn, model, example_inputs, **kwargs):
"""
Measure speedups over eager.
Writes to ./speedups.csv
"""
timings = np.zeros((args.repeat, 2), np.float64)
should_randomize_input = args.randomize_input
from torch._inductor.utils import maybe_profile
@contextlib.contextmanager
def maybe_mark_profile(*args, **kwargs):
prof: torch.profiler.profile = kwargs.pop("p", None)
mark = kwargs.pop("mark", None)
if prof:
with torch.profiler.record_function(mark):
yield
else:
yield
times = args.iterations_per_run
tolerance = 1e-4
torch._dynamo.config.repro_tolerance = tolerance
with maybe_profile(args.export_profiler_trace) as p:
frozen_model_iter_fn = torch._dynamo.run(model_iter_fn)
for rep in trange(args.repeat, desc="running benchmark"):
inputs = (
randomize_input(copy.deepcopy(example_inputs))
if should_randomize_input
else example_inputs
)
with maybe_mark_profile(p=p, mark="expected"):
timings[rep, 0], _ = timed(
model,
model_iter_fn,
inputs,
return_result=True,
times=times,
collect_outputs=args.collect_outputs,
)
with maybe_mark_profile(p=p, mark="actual"):
timings[rep, 1], _ = timed(
model,
frozen_model_iter_fn,
inputs,
return_result=True,
times=times,
collect_outputs=args.collect_outputs,
)
if args.export_profiler_trace:
name = args.profiler_trace_name + "_" + model.name + ".json"
name = os.path.join(torch._dynamo.config.base_dir, name)
p.export_chrome_trace(name)
median = np.median(timings, axis=0)
speedup = median[0] / median[1]
first_headers = ["dev", "name", "batch_size"]
first_fields = [current_device, current_name, current_batch_size]
if "tag" in kwargs:
first_headers.append("tag")
first_fields.append(kwargs["tag"])
headers = first_headers + ["speedup", "abs_latency"]
row = first_fields + [float(speedup), median[1] * 1000]
msg = f"{speedup:.3f}x"
if args.baseline:
headers.extend(
[
"baseline",
"speedup_vs_baseline",
]
)
df = pd.read_csv(args.baseline)
try:
baseline_speedup = df[df["name"] == current_name]["speedup"].item()
row.extend([baseline_speedup, speedup / baseline_speedup])
msg = f"{baseline_speedup:.3f}x -> {speedup:.3f}x [{speedup / baseline_speedup:.3f}x]"
except (KeyError, ZeroDivisionError):
row.extend(
[
0.0,
0.0,
]
)
if "compilation_latency" in kwargs:
headers += [
"compilation_latency",
"compression_ratio",
"eager_peak_mem",
"dynamo_peak_mem",
]
row.append(kwargs["compilation_latency"])
row.append(kwargs["compression_ratio"])
row.append(kwargs["eager_peak_mem"])
row.append(kwargs["dynamo_peak_mem"])
if "dynamo_stats" in kwargs:
for k, v in kwargs["dynamo_stats"].items():
headers.append(k)
row.append(v)
output_csv(
output_filename,
headers,
row,
)
headers, data = torch._dynamo.utils.compile_times(repr="csv", aggregate=True)
if output_filename.find(".csv") <= 0:
raise AssertionError(f"expected output_filename to be a .csv, but got {output_filename}")
output_csv(
output_filename[:-4] + "_compilation_metrics.csv",
first_headers + headers,
first_fields + data,
)
return msg
def read_batch_size_from_file(args, filename, model_name):
batch_size = None
if os.path.exists("benchmarks"):
filename = os.path.join("benchmarks", filename)
if not os.path.exists(filename):
raise AssertionError(filename)
abspath = os.path.abspath(filename)
PathManager.check_directory_path_readable(abspath)
with open(filename) as f:
lines = f.readlines()
lines = [i.split(",") for i in lines if len(i.strip()) > 0]
for val in lines:
cur_name, b = val
if model_name == cur_name:
batch_size = int(b)
if batch_size is None:
log.warning("Could not find batch size for %s", model_name)
elif batch_size == -1:
raise RuntimeError(
f"Batch size is unset for {model_name} in {args.batch_size_file}"
)
print(f"batch size: {batch_size}")
return batch_size
def get_peak_memory():
return torch.cuda.max_memory_allocated() / 10**9
def get_peak_memory_npu():
return torch_npu.npu.max_memory_allocated() / 10**9
def reset_rng_state():
torch.manual_seed(1337)
random.seed(1337)
np.random.seed(1337)
def get_dynamo_stats():
return collections.Counter(
{
"calls_captured": torch._dynamo.utils.counters["stats"]["calls_captured"],
"unique_graphs": torch._dynamo.utils.counters["stats"]["unique_graphs"],
"graph_breaks": sum(torch._dynamo.utils.counters["graph_break"].values()),
"unique_graph_breaks": len(+torch._dynamo.utils.counters["graph_break"]),
}
)
class BenchmarkRunner:
def __init__(self):
self.model_iter_fn = None
self.grad_scaler = DummyGradScaler()
self.autocast = contextlib.nullcontext
self.optimizer = None
self._args = None
def setup_amp(self):
if self.args.only in self.fp32_only_models:
return
if self.args.amp and self.args.devices == ["cuda"]:
self.autocast = torch.cuda.amp.autocast
elif (self.args.bfloat16 or self.args.amp) and self.args.devices == ["cpu"]:
self.autocast = torch.cpu.amp.autocast
elif self.args.amp and self.args.devices == ["npu"]:
self.autocast = torch_npu.npu.amp.autocast
def init_optimizer(self, name, device, params, learning_rate=0.01):
if device == "cuda" and self.args.training:
self.optimizer = torch.optim.SGD(params, lr=learning_rate, foreach=True)
elif device == "npu" and self.args.training:
self.optimizer = torch.optim.SGD(params, lr=learning_rate, foreach=False)
else:
self.optimizer = None
@property
def args(self):
return self._args
@args.setter
def args(self, args):
self._args = args
@property
def skip_models(self):
return set()
@property
def skip_models_for_cuda(self):
return set()
@property
def slow_models(self):
return set()
@property
def very_slow_models(self):
return set()
@property
def non_deterministic_models(self):
return set()
@property
def fp32_only_models(self):
return set()
@property
def force_amp_for_fp16_bf16_models(self):
return set()
@property
def failing_torchinductor_models(self):
return set()
def get_tolerance_and_cosine_flag(self, is_training, curr_device, name):
raise NotImplementedError()
def get_learning_rate(self, is_training, curr_device, name):
raise NotImplementedError()
@property
def equal_nan(self):
equal_nan = True
if self.args.float32:
equal_nan = False
return equal_nan
def iter_models(self, args):
for model_name in self.iter_model_names(args):
for device in args.devices:
try:
yield self.load_model(
device,
model_name,
batch_size=args.batch_size,
)
except NotImplementedError:
continue
def deepcopy_model(self, model):
return copy.deepcopy(model)
def cast_based_on_args(self, model, example_inputs):
if self.args.float32 or self.args.only in self.fp32_only_models:
if not self.args.float32:
log.warning("Model %s supports float32 only", self.args.only)
model, example_inputs = cast_to_fp32(model, example_inputs)
elif self.args.float16:
if self.args.only in self.force_amp_for_fp16_bf16_models:
log.warning(
"Model %s does not support float16, running with amp instead",
self.args.only,
)
self.args.amp = True
self.setup_amp()
else:
model, example_inputs = cast_to_fp16(model, example_inputs)
elif self.args.bfloat16:
if self.args.only in self.force_amp_for_fp16_bf16_models:
log.warning(
"Model %s does not support bfloat16, running with amp instead",
self.args.only,
)
self.args.amp = True
self.setup_amp()
else:
model, example_inputs = cast_to_bf16(model, example_inputs)
return model, example_inputs
def validate_model(self, model, example_inputs):
"""
Runs the eager model with example inputs to ensure that eager passes.
"""
model = self.deepcopy_model(model)
example_inputs = clone_inputs(example_inputs)
model, example_inputs = self.cast_based_on_args(model, example_inputs)
try:
self.model_iter_fn(model, example_inputs)
except Exception as e:
raise NotImplementedError("Eager model failed to run") from e
def maybe_cast(self, model, example_inputs):
model = self.deepcopy_model(model)
example_inputs = clone_inputs(example_inputs)
model, example_inputs = self.cast_based_on_args(model, example_inputs)
return model, example_inputs
def run_n_iterations(self, mod, inputs, run_mode=None):
n = self.args.iterations
if run_mode is None:
for _ in range(n - 1):
self.model_iter_fn(mod, inputs, collect_outputs=False)
return self.model_iter_fn(mod, inputs, collect_outputs=True)
start_step = int(n * 0.3)
end_step = min(int(n * 0.8) - 1, n - 1)
step_times = []
prof_output_dir = os.path.join(self.args.prof_output_path, self.args.only, run_mode)
if is_npu_available:
prof = NPUProfiler(enable=self.args.enable_profiler, warmup=10, active=n, save_path=prof_output_dir)
else:
prof = CUDAProfiler(enable=self.args.enable_profiler, warmup=10, active=n, save_path=prof_output_dir)
prof.start()
for i in range(n):
start = time.perf_counter()
output = self.model_iter_fn(mod, inputs, collect_outputs=(i == n - 1))
synchronize()
end = time.perf_counter()
elapsed_ms = (end - start) * 1000
prof.step()
step_times.append(elapsed_ms)
if i != n - 1:
print(f"[{run_mode}] step: {i+1} step_time: {elapsed_ms} ms loss: {output}")
else:
print(f"[{run_mode}] step: {i+1} step_time: {elapsed_ms} ms")
prof.stop()
steps = step_times[start_step:end_step + 1]
if steps:
total_ms = sum(steps)
avg_ms = total_ms / len(steps)
print(f"[{run_mode}] summary [{start_step+1}-{end_step+1}] "
f"total steps time: {total_ms:.2f} ms, "
f"avg step time: {avg_ms:.2f} ms")
return output
def optimizer_zero_grad(self, mod):
if self.optimizer is not None:
self.optimizer.zero_grad(True)
else:
mod.zero_grad(True)
def optimizer_step(self):
if self.optimizer is not None:
self.optimizer.step()
def deepcopy_and_maybe_ddp(self, model):
model = self.deepcopy_model(model)
if self.args.ddp:
if not torch.distributed.is_available():
raise AssertionError("Can't use DDP without a distributed enabled build")
from torch.nn.parallel import DistributedDataParallel as DDP
model = DDP(model, find_unused_parameters=True)
return model
def check_accuracy(
self, name, model, example_inputs, optimize_ctx, experiment, tag
):
"""
Checks accuracy.
1) Collect the outputs with fp64 datatype. This is useful for error checking.
2) Checks if eager itself has variations.
"""
start_stats = get_dynamo_stats()
lr = self.get_learning_rate(self.args.training, current_device, name)
print(f"learning rate: {lr}")
def record_status(accuracy_status, dynamo_start_stats):
"""
Records the status in the csv file
"""
if current_name in self.non_deterministic_models:
if accuracy_status in (
"pass_accuracy",
"eager_two_runs_differ",
"fail_accuracy",
):
accuracy_status = "pass_accuracy"
headers = ["dev", "name", "batch_size", "accuracy"]
fields = [current_device, current_name, current_batch_size, accuracy_status]
if tag is not None:
headers.insert(3, "tag")
fields.insert(3, tag)
dynamo_stats = get_dynamo_stats()
dynamo_stats.subtract(dynamo_start_stats)
for k, v in dynamo_stats.items():
headers.append(k)
fields.append(v)
output_csv(output_filename, headers, fields)
return accuracy_status
fp64_outputs = None
try:
model_fp64, inputs_fp64 = cast_to_fp64(
self.deepcopy_and_maybe_ddp(model),
clone_inputs(example_inputs),
)
self.init_optimizer(name, current_device, model_fp64.parameters(), lr)
fp64_outputs = self.run_n_iterations(model_fp64, inputs_fp64)
fp64_outputs = tree_map(
lambda x: x.to(torch.float64)
if isinstance(x, torch.Tensor) and x.is_floating_point()
else x,
fp64_outputs,
)
except Exception:
log.warning(
"fp64 golden ref were not generated for %s. Setting accuracy check to cosine",
name,
)
self.args.cosine = True
fp64_outputs = None
tolerance, cos_similarity = self.get_tolerance_and_cosine_flag(
self.args.training, current_device, name
)
print(f"tolerance: {tolerance}")
model, example_inputs = self.maybe_cast(model, example_inputs)
accuracy_status = "pass_accuracy"
with self.pick_grad(name, self.args.training):
reset_rng_state()
try:
model_copy = self.deepcopy_and_maybe_ddp(model)
self.init_optimizer(name, current_device, model_copy.parameters(), lr)
correct_result = self.run_n_iterations(
model_copy, clone_inputs(example_inputs), "eager"
)
except Exception as e:
accuracy_status = (
"eager_1st_run_OOM"
if isinstance(e, torch.cuda.OutOfMemoryError)
else "eager_1st_run_fail"
)
log.exception(e)
return record_status(accuracy_status, dynamo_start_stats=start_stats)
reset_rng_state()
try:
model_copy = self.deepcopy_and_maybe_ddp(model)
self.init_optimizer(name, current_device, model_copy.parameters(), lr)
correct_rerun_result = self.run_n_iterations(
model_copy, clone_inputs(example_inputs)
)
except Exception as e:
accuracy_status = (
"eager_2nd_run_OOM"
if isinstance(e, torch.cuda.OutOfMemoryError)
else "eager_2nd_run_fail"
)
return record_status(accuracy_status, dynamo_start_stats=start_stats)
is_same = True
try:
if (
not same(
correct_result,
correct_rerun_result,
fp64_ref=None,
cos_similarity=False,
tol=0,
equal_nan=self.equal_nan,
)
):
is_same = False
except Exception as e:
is_same = False
if not is_same:
accuracy_status = "eager_two_runs_differ"
return record_status(accuracy_status, dynamo_start_stats=start_stats)
correct_rerun_result = None
reset_rng_state()
torch._dynamo.reset()
try:
model_copy = self.deepcopy_and_maybe_ddp(model)
self.init_optimizer(name, current_device, model_copy.parameters(), lr)
optimized_model = optimize_ctx(model_copy)
new_result = self.run_n_iterations(optimized_model, example_inputs, "compile")
except Exception as e:
log.exception(e)
print(
"TorchDynamo optimized model failed to run because of following error"
)
accuracy_status = (
"OOM"
if isinstance(e, torch.cuda.OutOfMemoryError)
else "fail_to_run"
)
return record_status(accuracy_status, dynamo_start_stats=start_stats)
try:
if not same(
correct_result,
new_result,
fp64_outputs,
equal_nan=self.equal_nan,
cos_similarity=cos_similarity,
tol=tolerance,
):
is_same = False
except Exception as e:
is_same = False
if not is_same:
accuracy_status = "fail_accuracy"
return record_status(accuracy_status, dynamo_start_stats=start_stats)
return record_status(accuracy_status, dynamo_start_stats=start_stats)
def run_performance_test(
self, name, model, example_inputs, optimize_ctx, experiment, tag=None
):
def warmup(fn, model, example_inputs, mode, niters=5):
peak_mem = 0
start_stats = get_dynamo_stats()
try:
if current_device == "cuda":
torch.cuda.reset_peak_memory_stats()
torch.cuda.empty_cache()
elif current_device == "npu":
torch_npu.npu.reset_peak_memory_stats()
torch_npu.npu.empty_cache()
t0 = time.perf_counter()
for _ in range(niters):
fn(model, example_inputs)
t1 = time.perf_counter()
latency = t1 - t0
if current_device == "cuda":
peak_mem = get_peak_memory()
elif current_device == "npu":
peak_mem = get_peak_memory_npu()
elif current_device == "cpu":
total = psutil.virtual_memory().total
percentage = psutil.Process(os.getpid()).memory_percent()
peak_mem = percentage * total / 10**9
except Exception as e:
log.exception("Backend %s failed in warmup()", mode)
raise RuntimeError(f"Backend {mode} failed in warmup()") from e
dynamo_stats = get_dynamo_stats()
dynamo_stats.subtract(start_stats)
return latency, peak_mem, dynamo_stats
model, example_inputs = self.maybe_cast(model, example_inputs)
model = self.deepcopy_and_maybe_ddp(model)
self.init_optimizer(name, current_device, model.parameters())
with self.pick_grad(name, self.args.training):
ok, total = Stats.reset_counters()
experiment_kwargs = {}
if tag is not None:
experiment_kwargs["tag"] = tag
results = []
eager_latency, eager_peak_mem, _ = warmup(
self.model_iter_fn, model, example_inputs, "eager"
)
optimized_model_iter_fn = optimize_ctx(self.model_iter_fn)
dynamo_latency, dynamo_peak_mem, dynamo_stats = warmup(
optimized_model_iter_fn, model, example_inputs, "dynamo"
)
compilation_time = dynamo_latency - eager_latency
compression_ratio = (
eager_peak_mem / dynamo_peak_mem if dynamo_peak_mem else 0.0
)
if experiment.func is speedup_experiment:
experiment_kwargs["compilation_latency"] = compilation_time
experiment_kwargs["compression_ratio"] = compression_ratio
experiment_kwargs["eager_peak_mem"] = eager_peak_mem
experiment_kwargs["dynamo_peak_mem"] = dynamo_peak_mem
experiment_kwargs["dynamo_stats"] = dynamo_stats
if not hasattr(model, name):
model.name = name
results.append(experiment(model, example_inputs, **experiment_kwargs))
return " ".join(map(str, results))
def run_one_model(
self,
name,
model,
example_inputs,
optimize_ctx,
experiment,
tag=None,
):
mode = "train" if self.args.training else "eval"
msg = f"{current_device:4} {mode:5} {current_name:34} "
if tag:
msg += f" {tag:26}"
print(msg, flush=True)
start_stats = get_dynamo_stats()
if self.args.accuracy:
status = self.check_accuracy(
name, model, example_inputs, optimize_ctx, experiment, tag
)
print(status)
if self.args.dump_compile_time:
headers, values = torch._dynamo.utils.compile_times("csv")
for header, value in zip(headers, values):
if header == "async_compile.wait":
numbers = [float(num.strip()) for num in value.split(',') if num.strip()]
op_compile_time = sum(numbers)
print(f"op_compile_time:{op_compile_time * 1e3} ms", )
elif self.args.performance:
status = self.run_performance_test(
name, model, example_inputs, optimize_ctx, experiment, tag
)
print(status)
stats = get_dynamo_stats()
stats.subtract(start_stats)
if self.args.log_graph_breaks or self.args.print_graph_breaks:
filename = f"{output_filename.rstrip('.csv')}_graph_breaks.csv"
def add_double_quotes(x):
return f'"{x}"'
for graph_break in graph_break_reasons:
reason = add_double_quotes(graph_break.reason)
user_stack = add_double_quotes(
", ".join([str(x) for x in graph_break.user_stack])
)
output_csv(
filename,
["model", "reason", "user_stack"],
[current_name, reason, user_stack],
)
if self.args.stats:
Stats.print_summary()
def parse_args(args=None):
parser = argparse.ArgumentParser()
parser.add_argument(
"--devices", "--device", "-d", action="append", help="cpu or cuda"
)
parser.add_argument("--device-index", help="CUDA device index")
parser.add_argument(
"--repeat", "-n", type=int, default=30, help="number of timing runs"
)
iterations_per_run_help = """
Run this may iterations for each time measurement. This is mainly used for
XLA training. We want to run multiple iterations per measurement so the
tracing and computation for different iteartions can overlap with each
other. This makes sure we have an accurate xla baseline.
"""
parser.add_argument(
"--iterations-per-run", type=int, default=1, help=iterations_per_run_help
)
parser.add_argument(
"--randomize-input",
action="store_true",
help="Whether to randomize the input values. Dimensions will be kept the same.",
)
parser.add_argument(
"--nopython", action="store_true", help="Turn graph breaks into errors"
)
parser.add_argument(
"--no-skip",
action="store_true",
help="run models that are in the global SKIP list",
)
parser.add_argument(
"--batch-size", "--batch_size", type=int, help="batch size for benchmarking"
)
parser.add_argument(
"--iterations", type=int, default=50, help="how many iterations to run"
)
parser.add_argument(
"--batch-size-file", type=str, help="String to load batch size from"
)
parser.add_argument("--cosine", action="store_true", help="use cosine similarity")
parser.add_argument(
"--cpp-wrapper", action="store_true", help="turn on cpp/cuda wrapper codegen"
)
parser.add_argument(
"--freezing", action="store_true", help="turn on freezing", default=False
)
parser.add_argument(
"--only",
help="""Run just one model from torchbench. Or
specify the path and class name of the model in format like:
--only=path:<MODEL_FILE_PATH>,class:<CLASS_NAME>
Due to the fact that dynamo changes current working directory,
the path should be an absolute path.
The class should have a method get_example_inputs to return the inputs
for the model. An example looks like
```
class LinearModel(nn.Module):
def __init__(self):
super().__init__()
self.linear = nn.Linear(10, 10)
def forward(self, x):
return self.linear(x)
def get_example_inputs(self):
return (torch.randn(2, 10),)
```
""",
)
parser.add_argument(
"--multiprocess",
action="store_true",
help="Create n processes based on the number of devices (distributed use case).",
)
parser.add_argument(
"--ddp",
action="store_true",
help="Wraps model in DDP before running it, and uses dynamo DDPOptmizer (graph breaks) by default.",
)
parser.add_argument(
"--distributed-master-port",
default="6789",
help="Port to bind for for torch.distributed. Use the default unless it's conflicting with another user",
)
parser.add_argument(
"--dynamic-shapes",
action="store_true",
help="Runs a dynamic shapes version of the benchmark, if available.",
)
parser.add_argument(
"--dynamic-batch-only",
action="store_true",
help="Only assume batch dimension is dynamic. Implies --dynamic-shapes",
)
parser.add_argument(
"--output",
help="Overrides the output filename",
)
parser.add_argument(
"--output-directory",
help="Overrides the directory to place output files.",
)
parser.add_argument(
"--baseline",
help="Compare with a prior --output",
)
parser.add_argument(
"--part",
default=None,
help="Specify the part of the model to run.",
)
parser.add_argument(
"--export-profiler-trace",
action="store_true",
help="exports trace of kineto profiler",
)
parser.add_argument(
"--profiler-trace-name",
"--profiler_trace_name",
help="Overwrites exported trace name",
)
parser.add_argument(
"--tag", default=None, help="Specify a tag to be included in csv files."
)
parser.add_argument(
"--stats",
action="store_true",
help="print graph counter stats",
)
parser.add_argument(
"--cold-start-latency",
"--cold_start_latency",
action="store_true",
help="Use a fresh triton cachedir when running each model, to force cold-start compile.",
)
parser.add_argument(
"--disable-cudagraphs",
action="store_true",
help="Disables cudagraphs for Inductor",
)
parser.add_argument(
"--disable-split-reductions",
action="store_true",
help="Disables split reductions for Inductor",
)
parser.add_argument(
"--disable-persistent-reductions",
action="store_true",
help="Disables split reductions for Inductor",
)
parser.add_argument(
"--inductor-compile-mode",
default=None,
help="torch.compile mode argument for inductor runs.",
)
parser.add_argument(
"--print-graph-breaks",
action="store_true",
help="Show a warning whenever graph break",
)
parser.add_argument(
"--log-graph-breaks",
action="store_true",
help="log graph breaks in a file",
)
parser.add_argument(
"--collect-outputs",
action="store_true",
help="""Whether to collect outputs for training. Set this to true if we
want to verify the numerical correctness of graidents. But that may
cause time measurement not accurate""",
)
parser.add_argument(
"--timeout",
type=int,
default=2000,
help="timeout (second) for benchmarking.",
)
parser.add_argument(
"--enable-profiler",
action="store_true",
help="Enable profile for NPU and GPU."
)
parser.add_argument(
"--prof-output-path",
help="Overrides the profile output path",
default="./profile"
)
parser.add_argument(
"--dump-compile-time",
action="store_true",
help="dump compile time",
)
group_prec = parser.add_mutually_exclusive_group()
group_prec.add_argument("--float16", action="store_true", help="cast model to fp16")
group_prec.add_argument(
"--bfloat16", action="store_true", help="cast model to bf16"
)
group_prec.add_argument("--float32", action="store_true", help="cast model to fp32")
group_prec.add_argument(
"--amp", action="store_true", help="use automatic mixed precision"
)
group = parser.add_mutually_exclusive_group()
group.add_argument(
"--inductor",
action="store_true",
help="Measure speedup with TorchInductor",
)
group.add_argument(
"--backend",
choices=torch._dynamo.list_backends(exclude_tags=None),
help="measure speedup with a given backend",
)
group.add_argument("--nothing", action="store_true", help="A no-op experiment useful for making sure TorchBenchark alone works properly")
mode_group = parser.add_mutually_exclusive_group(required=True)
mode_group.add_argument(
"--accuracy",
action="store_true",
help="Checks accuracy with small batch size and eval mode",
)
mode_group.add_argument(
"--performance", action="store_true", help="Measures performance speedup"
)
run_mode_group = parser.add_mutually_exclusive_group(required=True)
run_mode_group.add_argument(
"--training",
action="store_true",
help="Performs training",
)
run_mode_group.add_argument(
"--inference", action="store_true", help="Performs inference"
)
return parser.parse_args(args)
def process_entry(rank, runner, original_dir, args):
args.rank = rank
with maybe_init_distributed(
args.use_distributed,
rank=rank,
world_size=args.world_size,
port=args.distributed_master_port,
):
return maybe_fresh_cache(
run, (args.cold_start_latency and args.only)
)(runner, args, original_dir)
def main(runner, original_dir=None):
if original_dir:
os.chdir(original_dir)
args = parse_args()
if args.baseline:
args.baseline = os.path.abspath(args.baseline)
if is_npu_available and args.only:
patch_model(args.only)
args.use_distributed = (args.ddp) and args.only
if args.multiprocess:
device_count = torch.cuda.device_count()
if device_count <= 1:
log.warning(
"The use multiprocess flag is set but there are <= 1 devices available."
)
args.world_size = device_count
mp.spawn(process_entry, args=(runner, original_dir, args), nprocs=device_count)
else:
args.world_size = 1
process_entry(0, runner, original_dir, args)
def run(runner, args, original_dir=None):
runner.args = args
if args.inductor:
if args.backend is not None:
raise AssertionError
args.backend = "inductor"
if args.dynamic_batch_only:
args.dynamic_shapes = True
torch._dynamo.config.assume_static_by_default = True
if args.dynamic_shapes:
if not args.dynamic_batch_only:
torch._dynamo.config.assume_static_by_default = False
if args.ddp:
if not args.accuracy:
raise AssertionError("DDP benchmark is currently only hooked up to --accuracy bench")
if not args.training:
raise AssertionError("DDP benchmark requires --training mode")
if args.accuracy:
if args.batch_size is None:
if runner.suite_name == "huggingface":
args.batch_size = 1
elif runner.suite_name == "torchbench":
args.batch_size = 4
else:
if runner.suite_name != "timm_models":
raise AssertionError
args.batch_size = 8
inductor_config.fallback_random = True
if args.only is not None and args.only not in {
"alexnet",
"Background_Matting",
"pytorch_CycleGAN_and_pix2pix",
"pytorch_unet",
"Super_SloMo",
"vgg16",
"Wav2Vec2ForCTC",
"Wav2Vec2ForPreTraining",
"sam",
}:
torch.use_deterministic_algorithms(True)
else:
log.warning("Currently, all models keep deterministic open on npu. "
"But on gpu, this model does not support use_deterministic_algorithms. "
"Please check it to prevent bugs.")
torch.use_deterministic_algorithms(True)
os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8"
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.allow_tf32 = False
torch.backends.cudnn.benchmark = False
torch.backends.cuda.matmul.allow_tf32 = False
patch_torch_manual_seed()
if "CUDA_VISIBLE_DEVICES" not in os.environ:
args.device_index = "0"
if args.device_index is not None:
os.environ["CUDA_VISIBLE_DEVICES"] = args.device_index
def __check_if_transfer_to_npu():
return torch.cuda.is_available == is_npu_available
if not args.devices:
if torch.cuda.is_available():
if __check_if_transfer_to_npu():
args.devices = ["npu"]
else:
args.devices = ["cuda"]
elif is_npu_available:
args.devices = ["npu"]
else:
log.warning("torch.cuda.is_available() == False, using CPU")
args.devices = ["cpu"]
if args.devices != ["cpu"] and torch.cuda.is_available():
global synchronize
synchronize = torch.cuda.synchronize
elif is_npu_available:
synchronize = torch_npu.npu.synchronize
if (
args.devices == ["cuda"]
and torch.cuda.get_device_properties(0).total_memory < 25 * 2**30
):
runner.skip_models.update(
{
"hf_Longformer",
"timm_nfnet",
"timm_efficientdet",
}
)
if args.training:
runner.skip_models.add("hf_T5")
if args.print_graph_breaks:
torch._dynamo.config.print_graph_breaks = True
if args.training:
runner.model_iter_fn = runner.forward_and_backward_pass
else:
runner.model_iter_fn = runner.forward_pass
if args.devices == ["cpu"]:
runner.skip_models.update(runner.very_slow_models)
if args.no_skip:
runner.skip_models.clear()
experiment = null_experiment
global current_name, current_device, current_batch_size, output_filename
optimize_ctx = contextlib.nullcontext()
if args.inductor:
optimize_ctx = functools.partial(
torch.compile,
backend="inductor",
fullgraph=args.nopython,
mode=args.inductor_compile_mode,
)
experiment = speedup_experiment
output_filename = "inductor.csv"
elif args.nothing:
optimize_ctx = nothing
experiment = speedup_experiment
output_filename = "nothing.csv"
elif args.backend:
optimize_ctx = compile_with_backend(args)
experiment = speedup_experiment
if args.accuracy:
output_filename = f"accuracy_{args.backend}.csv"
else:
output_filename = f"speedup_{args.backend}.csv"
if args.only is not None and args.only not in {"hf_Bart", "torch_multimodal_clip", "timm_vision_transformer"}:
if args.inductor or args.backend == "inductor" or args.export_aot_inductor:
inductor_config.triton.cudagraphs = not args.disable_cudagraphs
runner.setup_amp()
if args.output:
output_filename = args.output
if output_filename:
if args.output_directory:
output_filename = os.path.join(args.output_directory, output_filename)
else:
output_filename = os.path.join(
os.path.dirname(os.path.abspath(__file__)), output_filename
)
if args.export_profiler_trace:
if args.profiler_trace_name is None:
if args.backend:
args.profiler_trace_name = args.backend
elif args.inductor:
args.profiler_trace_name = "inductor"
else:
args.profiler_trace_name = "profile"
else:
args.profiler_trace_name = args.profiler_trace_name
experiment = functools.partial(experiment, args, runner.model_iter_fn)
if args.only:
if os.environ.get("USE_ACLOP", "0").upper() in ["1", "ON"]:
torch_npu.npu.set_compile_mode(jit_compile=True)
model_name = args.only
for device in args.devices:
batch_size = args.batch_size
if args.batch_size_file:
batch_size = read_batch_size_from_file(
args, args.batch_size_file, model_name
)
if model_specified_by_path(args.only):
model, example_inputs = load_model_from_path(args.only)
name = model.__class__.__name__
model = model.to(device=device)
example_inputs = tree_map_only(
torch.Tensor, lambda x: x.to(device=device), example_inputs
)
else:
try:
with tqdm(desc="loading model"):
if args.part:
(
device,
name,
model,
example_inputs,
batch_size,
) = runner.load_model(
device,
model_name,
batch_size=batch_size,
part=args.part,
)
else:
(
device,
name,
model,
example_inputs,
batch_size,
) = runner.load_model(
device, model_name, batch_size=batch_size
)
except NotImplementedError as e:
print(e)
import traceback
print(traceback.format_exc())
logging.warning("%s failed to load", args.only)
continue
current_name = name
current_device = device
current_batch_size = batch_size
set_model_name(name)
marked = False
def detect_and_mark_batch(t, target_size=batch_size):
nonlocal marked
for i, s in enumerate(t.size()):
if s == target_size:
torch._dynamo.mark_dynamic(t, i)
marked = True
break
if (
args.dynamic_batch_only
and batch_size > 1
):
tree_map_only(torch.Tensor, detect_and_mark_batch, example_inputs)
if not marked:
raise AssertionError(f"nothing in example_inputs had a dim with {batch_size}")
model, example_inputs = runner.cast_based_on_args(model, example_inputs)
runner.run_one_model(
name,
model,
example_inputs,
optimize_ctx,
experiment,
tag=args.tag,
)
for fn in callbacks:
fn()
else:
if output_filename and os.path.exists(output_filename):
os.unlink(output_filename)
if original_dir:
os.chdir(original_dir)
model_names = list(runner.iter_model_names(args))
nmodels = len(model_names)
for i, name in enumerate(model_names):
current_name = name
placeholder_batch_size = 0
print(f"Running model {i+1}/{nmodels}", flush=True)
def write_csv(status, name=name, placeholder_batch_size=placeholder_batch_size):
if args.accuracy:
headers = ["dev", "name", "batch_size", "accuracy"]
rows = [
[device, name, placeholder_batch_size, status]
for device in args.devices
]
elif args.performance:
headers = ["dev", "name", "batch_size", "speedup", "abs_latency"]
rows = [
[device, name, placeholder_batch_size, 0.0, 0.0]
for device in args.devices
]
else:
headers = []
rows = [
[device, name, placeholder_batch_size, 0.0]
for device in args.devices
]
for row in rows:
output_csv(output_filename, headers, row)
try:
subprocess.check_call(
[sys.executable] + sys.argv + [f"--only={name}"], timeout=args.timeout
)
except subprocess.TimeoutExpired:
print("TIMEOUT", file=sys.stderr)
write_csv("timeout")
except subprocess.SubprocessError:
print("ERROR", file=sys.stderr)
write_csv("infra_error")
def compile_with_backend(args):
return torch._dynamo.optimize(args.backend, nopython=args.nopython)
if __name__ == "__main__":
raise RuntimeError(
f"You shouldn't run {sys.argv[0]} directly, instead try torchbench.py"
)
