torch.compile End-to-End Tutorial">

intermediate/torch_compile_full_example

Run in Google Colab

Colab

Download Notebook

Notebook

View on GitHub

GitHub

torch.compile End-to-End Tutorial

Author: William Wen

torch.compile is the new way to speed up your PyTorch code! torch.compile makes PyTorch code run faster by JIT-compiling PyTorch code into optimized kernels, while requiring minimal code changes.

This tutorial covers an end-to-end example of training and evaluating a real model with torch.compile. For a gentle introduction to torch.compile, please check out the introduction to torch.compile tutorial.

Required pip Dependencies

• torch >= 2.0

• torchvision

What you will learn

• How to apply torch.compile to a real model

• torch.compile speedups on a real model

• torch.compile’s first few iterations are expected to be slower due to compilation overhead

Prerequisites

• Introduction to torch.compile

# NOTE: a modern NVIDIA GPU (H100, A100, or V100) is recommended for this tutorial in
# order to reproduce the speedup numbers shown below and documented elsewhere.

import torch
import warnings

gpu_ok = False
if torch.cuda.is_available():
    device_cap = torch.cuda.get_device_capability()
    if device_cap in ((7, 0), (8, 0), (9, 0)):
        gpu_ok = True

if not gpu_ok:
    warnings.warn(
        "GPU is not NVIDIA V100, A100, or H100. Speedup numbers may be lower "
        "than expected."
    )

/var/lib/workspace/intermediate_source/torch_compile_full_example.py:51: UserWarning:

GPU is not NVIDIA V100, A100, or H100. Speedup numbers may be lower than expected.

Let’s demonstrate how using torch.compile can speed up a real model. We will compare standard eager mode and torch.compile by evaluating and training a torchvision model on random data.

Before we start, we need to define some utility functions.

# Returns the result of running `fn()` and the time it took for `fn()` to run,
# in seconds. We use CUDA events and synchronization for the most accurate
# measurements.
def timed(fn):
    start = torch.cuda.Event(enable_timing=True)
    end = torch.cuda.Event(enable_timing=True)
    start.record()
    result = fn()
    end.record()
    torch.cuda.synchronize()
    return result, start.elapsed_time(end) / 1000


# Generates random input and targets data for the model, where `b` is
# batch size.
def generate_data(b):
    return (
        torch.randn(b, 3, 128, 128).to().cuda(),
        torch.randint(1000, (b,)).cuda(),
    )


N_ITERS = 10

from torchvision.models import densenet121


def init_model():
    return densenet121().cuda()

First, let’s compare inference.

Note that in the call to torch.compile, we have the additional mode argument, which we will discuss below.

model = init_model()

# Note that we generally recommend directly compiling a torch.nn.Module by calling
# its .compile() method.
model_opt = init_model()
model_opt.compile(mode="reduce-overhead")

inp = generate_data(16)[0]
with torch.no_grad():
    print("eager:", timed(lambda: model(inp))[1])
    print("compile:", timed(lambda: model_opt(inp))[1])

eager: 0.3626025085449219
/usr/local/lib/python3.10/dist-packages/torch/backends/cuda/__init__.py:131: UserWarning:

Please use the new API settings to control TF32 behavior, such as torch.backends.cudnn.conv.fp32_precision = 'tf32' or torch.backends.cuda.matmul.fp32_precision = 'ieee'. Old settings, e.g, torch.backends.cuda.matmul.allow_tf32 = True, torch.backends.cudnn.allow_tf32 = True, allowTF32CuDNN() and allowTF32CuBLAS() will be deprecated after Pytorch 2.9. Please see https://pytorch.org/docs/main/notes/cuda.html#tensorfloat-32-tf32-on-ampere-and-later-devices (Triggered internally at /pytorch/aten/src/ATen/Context.cpp:80.)

/usr/local/lib/python3.10/dist-packages/torch/_inductor/compile_fx.py:312: UserWarning:

TensorFloat32 tensor cores for float32 matrix multiplication available but not enabled. Consider setting `torch.set_float32_matmul_precision('high')` for better performance.

compile: 51.39578125

Notice that torch.compile takes a lot longer to complete compared to eager. This is because torch.compile compiles the model into optimized kernels as it executes. In our example, the structure of the model doesn’t change, and so recompilation is not needed. So if we run our optimized model several more times, we should see a significant improvement compared to eager.

eager_times = []
for i in range(N_ITERS):
    inp = generate_data(16)[0]
    with torch.no_grad():
        _, eager_time = timed(lambda: model(inp))
    eager_times.append(eager_time)
    print(f"eager eval time {i}: {eager_time}")

print("~" * 10)

compile_times = []
for i in range(N_ITERS):
    inp = generate_data(16)[0]
    with torch.no_grad():
        _, compile_time = timed(lambda: model_opt(inp))
    compile_times.append(compile_time)
    print(f"compile eval time {i}: {compile_time}")
print("~" * 10)

import numpy as np

eager_med = np.median(eager_times)
compile_med = np.median(compile_times)
speedup = eager_med / compile_med
assert speedup > 1
print(
    f"(eval) eager median: {eager_med}, compile median: {compile_med}, speedup: {speedup}x"
)
print("~" * 10)

eager eval time 0: 0.018395135879516602
eager eval time 1: 0.01702400016784668
eager eval time 2: 0.016776191711425782
eager eval time 3: 0.016738304138183592
eager eval time 4: 0.016716863632202147
eager eval time 5: 0.016763904571533202
eager eval time 6: 0.01661235237121582
eager eval time 7: 0.016699392318725585
eager eval time 8: 0.016652288436889647
eager eval time 9: 0.01660518455505371
~~~~~~~~~~
compile eval time 0: 0.061088768005371094
compile eval time 1: 0.007836671829223632
compile eval time 2: 0.008395744323730468
compile eval time 3: 0.007539711952209473
compile eval time 4: 0.007554048061370849
compile eval time 5: 0.007521279811859131
compile eval time 6: 0.007561215877532959
compile eval time 7: 0.007514111995697022
compile eval time 8: 0.007546879768371582
compile eval time 9: 0.007513088226318359
~~~~~~~~~~
(eval) eager median: 0.016727583885192868, compile median: 0.0075504639148712156, speedup: 2.2154378954446248x
~~~~~~~~~~

And indeed, we can see that running our model with torch.compile results in a significant speedup. Speedup mainly comes from reducing Python overhead and GPU read/writes, and so the observed speedup may vary on factors such as model architecture and batch size. For example, if a model’s architecture is simple and the amount of data is large, then the bottleneck would be GPU compute and the observed speedup may be less significant.

You may also see different speedup results depending on the chosen mode argument. The "reduce-overhead" mode uses CUDA graphs to further reduce the overhead of Python. For your own models, you may need to experiment with different modes to maximize speedup. You can read more about modes here.

You may might also notice that the second time we run our model with torch.compile is significantly slower than the other runs, although it is much faster than the first run. This is because the "reduce-overhead" mode runs a few warm-up iterations for CUDA graphs.

Now, let’s consider comparing training.

model = init_model()
opt = torch.optim.Adam(model.parameters())


def train(mod, data):
    opt.zero_grad(True)
    pred = mod(data[0])
    loss = torch.nn.CrossEntropyLoss()(pred, data[1])
    loss.backward()
    opt.step()


eager_times = []
for i in range(N_ITERS):
    inp = generate_data(16)
    _, eager_time = timed(lambda: train(model, inp))
    eager_times.append(eager_time)
    print(f"eager train time {i}: {eager_time}")
print("~" * 10)

model = init_model()
opt = torch.optim.Adam(model.parameters())

# Note that because we are compiling a regular Python function, we do not
# call any .compile() method.
train_opt = torch.compile(train, mode="reduce-overhead")

compile_times = []
for i in range(N_ITERS):
    inp = generate_data(16)
    _, compile_time = timed(lambda: train_opt(model, inp))
    compile_times.append(compile_time)
    print(f"compile train time {i}: {compile_time}")
print("~" * 10)

eager_med = np.median(eager_times)
compile_med = np.median(compile_times)
speedup = eager_med / compile_med
assert speedup > 1
print(
    f"(train) eager median: {eager_med}, compile median: {compile_med}, speedup: {speedup}x"
)
print("~" * 10)

eager train time 0: 0.29086712646484375
eager train time 1: 0.05220249557495117
eager train time 2: 0.050874366760253906
eager train time 3: 0.05087027359008789
eager train time 4: 0.7923558349609375
eager train time 5: 0.05110784149169922
eager train time 6: 0.05072588729858398
eager train time 7: 0.05056716918945312
eager train time 8: 0.051146751403808595
eager train time 9: 0.05076377487182617
~~~~~~~~~~
compile train time 0: 153.296140625
compile train time 1: 2.97493408203125
compile train time 2: 0.02395955276489258
compile train time 3: 0.02142207908630371
compile train time 4: 0.020762624740600585
compile train time 5: 0.020744192123413087
compile train time 6: 0.020756479263305663
compile train time 7: 0.020754432678222655
compile train time 8: 0.02079020881652832
compile train time 9: 0.020716543197631835
~~~~~~~~~~
(train) eager median: 0.05099110412597656, compile median: 0.020776416778564455, speedup: 2.454278072558948x
~~~~~~~~~~

Again, we can see that torch.compile takes longer in the first iteration, as it must compile the model, but in subsequent iterations, we see significant speedups compared to eager.

We remark that the speedup numbers presented in this tutorial are for demonstration purposes only. Official speedup values can be seen at the TorchInductor performance dashboard.

Conclusion

In this tutorial, we applied torch.compile to training and inference on a real model, demonstrating speedups.

Importantly, we note that the first few iterations of a compiled model are slower than eager mode due to compilation overhead, but subsequent iterations are expected to have speedups.

For a gentle introduction to torch.compile, please check out the introduction to torch.compile tutorial.

To troubleshoot issues and to gain a deeper understanding of how to apply torch.compile to your code, check out the torch.compile programming model.

We hope that you will give torch.compile a try!