torch.export AOTInductor Tutorial for Python runtime (Beta)

Created On: Aug 23, 2024 | Last Updated: Jan 24, 2025 | Last Verified: Nov 05, 2024

Author: Ankith Gunapal, Bin Bao, Angela Yi

Warning

torch._inductor.aoti_compile_and_package and torch._inductor.aoti_load_package are in Beta status and are subject to backwards compatibility breaking changes. This tutorial provides an example of how to use these APIs for model deployment using Python runtime.

It has been shown previously how AOTInductor can be used to do Ahead-of-Time compilation of PyTorch exported models by creating an artifact that can be run in a non-Python environment. In this tutorial, you will learn an end-to-end example of how to use AOTInductor for Python runtime.

Contents

Prerequisites

• PyTorch 2.6 or later

• Basic understanding of torch.export and AOTInductor

• Complete the AOTInductor: Ahead-Of-Time Compilation for Torch.Export-ed Models tutorial

What you will learn

• How to use AOTInductor for Python runtime.

• How to use torch._inductor.aoti_compile_and_package() along with torch.export.export() to generate a compiled artifact

• How to load and run the artifact in a Python runtime using torch._export.aot_load().

• When to you use AOTInductor with a Python runtime

Model Compilation

We will use the TorchVision pretrained ResNet18 model as an example.

The first step is to export the model to a graph representation using torch.export.export(). To learn more about using this function, you can check out the docs or the tutorial.

Once we have exported the PyTorch model and obtained an ExportedProgram, we can apply torch._inductor.aoti_compile_and_package() to AOTInductor to compile the program to a specified device, and save the generated contents into a “.pt2” artifact.

Note

This API supports the same available options that torch.compile() has, such as mode and max_autotune (for those who want to enable CUDA graphs and leverage Triton based matrix multiplications and convolutions)

import os
import torch
import torch._inductor
from torchvision.models import ResNet18_Weights, resnet18

model = resnet18(weights=ResNet18_Weights.DEFAULT)
model.eval()

with torch.inference_mode():
    inductor_configs = {}

    if torch.cuda.is_available():
        device = "cuda"
        inductor_configs["max_autotune"] = True
    else:
        device = "cpu"

    model = model.to(device=device)
    example_inputs = (torch.randn(2, 3, 224, 224, device=device),)

    exported_program = torch.export.export(
        model,
        example_inputs,
    )
    path = torch._inductor.aoti_compile_and_package(
        exported_program,
        package_path=os.path.join(os.getcwd(), "resnet18.pt2"),
        inductor_configs=inductor_configs
    )

Downloading: "https://download.pytorch.org/models/resnet18-f37072fd.pth" to /var/lib/ci-user/.cache/torch/hub/checkpoints/resnet18-f37072fd.pth

  0%|          | 0.00/44.7M [00:00<?, ?B/s]
 90%|######### | 40.2M/44.7M [00:00<00:00, 422MB/s]
100%|##########| 44.7M/44.7M [00:00<00:00, 424MB/s]
AUTOTUNE convolution(2x3x224x224, 64x3x7x7)
  convolution 0.0902 ms 100.0%
  triton_convolution2d_4 0.1507 ms 59.9% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=7, KERNEL_W=7, PADDING_H=3, PADDING_W=3, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_0 0.1546 ms 58.3% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=7, KERNEL_W=7, PADDING_H=3, PADDING_W=3, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_3 0.1962 ms 46.0% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=128, BLOCK_N=64, GROUPS=1, KERNEL_H=7, KERNEL_W=7, PADDING_H=3, PADDING_W=3, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_5 0.2390 ms 37.7% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=7, KERNEL_W=7, PADDING_H=3, PADDING_W=3, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_2 0.3036 ms 29.7% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=1024, BLOCK_N=16, GROUPS=1, KERNEL_H=7, KERNEL_W=7, PADDING_H=3, PADDING_W=3, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=1, num_warps=8
  triton_convolution2d_1 0.4442 ms 20.3% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=7, KERNEL_W=7, PADDING_H=3, PADDING_W=3, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
SingleProcess AUTOTUNE benchmarking takes 0.5348 seconds and 0.0003 seconds precompiling for 7 choices
AUTOTUNE convolution(2x64x56x56, 64x64x3x3)
  triton_convolution2d_11 0.0372 ms 100.0% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_10 0.0389 ms 95.6% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_9 0.0423 ms 88.0% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_6 0.0524 ms 71.1% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_7 0.0605 ms 61.6% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_12 0.0661 ms 56.3% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  convolution 0.0840 ms 44.3%
  triton_convolution2d_8 0.1208 ms 30.8% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=1024, BLOCK_N=16, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=1, num_warps=8
SingleProcess AUTOTUNE benchmarking takes 0.1953 seconds and 0.0003 seconds precompiling for 8 choices
AUTOTUNE convolution(2x64x56x56, 128x64x3x3)
  triton_convolution2d_38 0.0345 ms 100.0% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_39 0.0362 ms 95.3% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=128, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_34 0.0507 ms 68.1% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=128, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_35 0.0580 ms 59.5% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_40 0.0622 ms 55.5% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_37 0.0645 ms 53.5% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=128, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  convolution 0.0995 ms 34.7%
  triton_convolution2d_36 0.1124 ms 30.7% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=1024, BLOCK_N=16, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=1, num_warps=8
SingleProcess AUTOTUNE benchmarking takes 0.1769 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x128x28x28, 128x128x3x3)
  triton_convolution2d_45 0.0494 ms 100.0% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_46 0.0750 ms 65.9% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=128, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  convolution 0.0816 ms 60.6%
  triton_convolution2d_41 0.0910 ms 54.3% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=128, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_42 0.1102 ms 44.9% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_44 0.1151 ms 42.9% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=128, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_47 0.1198 ms 41.3% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_43 0.2269 ms 21.8% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=1024, BLOCK_N=16, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=1, num_warps=8
SingleProcess AUTOTUNE benchmarking takes 0.2137 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x64x56x56, 128x64x1x1)
  triton_convolution2d_49 0.0207 ms 100.0% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=4
  triton_convolution2d_51 0.0240 ms 86.0% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=128, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=8
  triton_convolution2d_53 0.0241 ms 85.8% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=128, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=8
  triton_convolution2d_52 0.0248 ms 83.5% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=4
  triton_convolution2d_48 0.0250 ms 82.7% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=128, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=4
  triton_convolution2d_54 0.0260 ms 79.7% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=8
  triton_convolution2d_50 0.0272 ms 76.0% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=1024, BLOCK_N=16, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=1, num_warps=8
  convolution 0.0608 ms 34.0%
SingleProcess AUTOTUNE benchmarking takes 0.1594 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x128x28x28, 256x128x3x3)
  triton_convolution2d_73 0.0463 ms 100.0% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
  convolution 0.0831 ms 55.8%
  triton_convolution2d_69 0.1066 ms 43.5% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_74 0.1073 ms 43.2% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_70 0.1092 ms 42.5% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_75 0.1161 ms 39.9% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_72 0.1205 ms 38.4% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=128, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_71 0.1785 ms 26.0% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=1024, BLOCK_N=16, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=1, num_warps=8
SingleProcess AUTOTUNE benchmarking takes 0.2148 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x256x14x14, 256x256x3x3)
  convolution 0.0897 ms 100.0%
  triton_convolution2d_80 0.0917 ms 97.8% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_81 0.2036 ms 44.0% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_76 0.2058 ms 43.6% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_77 0.2132 ms 42.1% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_78 0.2136 ms 42.0% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=512, BLOCK_N=16, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=1, num_warps=8
  triton_convolution2d_79 0.2229 ms 40.2% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=128, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_82 0.2332 ms 38.5% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
SingleProcess AUTOTUNE benchmarking takes 0.2775 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x128x28x28, 256x128x1x1)
  triton_convolution2d_87 0.0212 ms 100.0% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=4
  triton_convolution2d_89 0.0276 ms 76.7% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=8
  triton_convolution2d_88 0.0280 ms 75.9% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=8
  triton_convolution2d_86 0.0284 ms 74.8% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=128, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=8
  triton_convolution2d_83 0.0298 ms 71.3% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=4
  triton_convolution2d_84 0.0312 ms 67.9% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=4
  triton_convolution2d_85 0.0353 ms 60.1% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=1024, BLOCK_N=16, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=1, num_warps=8
  convolution 0.0614 ms 34.5%
SingleProcess AUTOTUNE benchmarking takes 0.1650 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x256x14x14, 512x256x3x3)
  convolution 0.0860 ms 100.0%
  triton_convolution2d_108 0.0900 ms 95.5% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_104 0.2078 ms 41.4% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_106 0.2095 ms 41.1% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=512, BLOCK_N=16, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=1, num_warps=8
  triton_convolution2d_109 0.2153 ms 40.0% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_110 0.2332 ms 36.9% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_107 0.2361 ms 36.4% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=128, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_105 0.2724 ms 31.6% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
SingleProcess AUTOTUNE benchmarking takes 0.2814 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x512x7x7, 512x512x3x3)
  convolution 0.1125 ms 100.0%
  triton_convolution2d_115 0.1840 ms 61.1% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_113 0.2153 ms 52.3% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=128, BLOCK_N=16, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=1, num_warps=8
  triton_convolution2d_117 0.2712 ms 41.5% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_112 0.2868 ms 39.2% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=128, BLOCK_N=64, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_111 0.4092 ms 27.5% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=4
  triton_convolution2d_116 0.4235 ms 26.6% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_114 0.4426 ms 25.4% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=128, GROUPS=1, KERNEL_H=3, KERNEL_W=3, PADDING_H=1, PADDING_W=1, STRIDE_H=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
SingleProcess AUTOTUNE benchmarking takes 0.3012 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x256x14x14, 512x256x1x1)
  triton_convolution2d_122 0.0207 ms 100.0% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=64, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=4
  triton_convolution2d_120 0.0348 ms 59.5% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=512, BLOCK_N=16, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=1, num_warps=8
  triton_convolution2d_119 0.0372 ms 55.7% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=4
  triton_convolution2d_118 0.0385 ms 53.7% ALLOW_TF32=True, BLOCK_K=16, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=4
  triton_convolution2d_121 0.0404 ms 51.3% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=128, BLOCK_N=128, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=8
  triton_convolution2d_124 0.0405 ms 51.1% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=256, BLOCK_N=64, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=8
  triton_convolution2d_123 0.0408 ms 50.8% ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=64, BLOCK_N=256, GROUPS=1, KERNEL_H=1, KERNEL_W=1, PADDING_H=0, PADDING_W=0, STRIDE_H=2, STRIDE_W=2, UNROLL=True, num_stages=2, num_warps=8
  convolution 0.0851 ms 24.3%
SingleProcess AUTOTUNE benchmarking takes 0.1633 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE addmm(2x1000, 2x512, 512x1000)
  triton_mm_141 0.0202 ms 100.0% ACC_TYPE='tl.float32', ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=16, BLOCK_N=64, EVEN_K=True, GROUP_M=8, num_stages=5, num_warps=4
  triton_mm_142 0.0208 ms 97.2% ACC_TYPE='tl.float32', ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=16, BLOCK_N=32, EVEN_K=True, GROUP_M=8, num_stages=5, num_warps=2
  triton_mm_146 0.0211 ms 96.0% ACC_TYPE='tl.float32', ALLOW_TF32=True, BLOCK_K=64, BLOCK_M=16, BLOCK_N=64, EVEN_K=True, GROUP_M=8, num_stages=3, num_warps=4
  triton_mm_153 0.0211 ms 95.8% ACC_TYPE='tl.float32', ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=16, BLOCK_N=64, EVEN_K=True, GROUP_M=8, num_stages=4, num_warps=4
  triton_mm_149 0.0220 ms 91.7% ACC_TYPE='tl.float32', ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=16, BLOCK_N=128, EVEN_K=True, GROUP_M=8, num_stages=4, num_warps=8
  triton_mm_154 0.0242 ms 83.6% ACC_TYPE='tl.float32', ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=16, BLOCK_N=128, EVEN_K=True, GROUP_M=8, num_stages=2, num_warps=8
  triton_mm_150 0.0248 ms 81.4% ACC_TYPE='tl.float32', ALLOW_TF32=True, BLOCK_K=64, BLOCK_M=16, BLOCK_N=128, EVEN_K=True, GROUP_M=8, num_stages=3, num_warps=4
  triton_mm_140 0.0252 ms 80.2% ACC_TYPE='tl.float32', ALLOW_TF32=True, BLOCK_K=128, BLOCK_M=16, BLOCK_N=32, EVEN_K=True, GROUP_M=8, num_stages=2, num_warps=2
  triton_mm_143 0.0255 ms 79.2% ACC_TYPE='tl.float32', ALLOW_TF32=True, BLOCK_K=128, BLOCK_M=16, BLOCK_N=32, EVEN_K=True, GROUP_M=8, num_stages=5, num_warps=2
  triton_mm_152 0.0262 ms 77.2% ACC_TYPE='tl.float32', ALLOW_TF32=True, BLOCK_K=32, BLOCK_M=16, BLOCK_N=64, EVEN_K=True, GROUP_M=8, num_stages=3, num_warps=4
SingleProcess AUTOTUNE benchmarking takes 0.4992 seconds and 0.0002 seconds precompiling for 18 choices

The result of aoti_compile_and_package() is an artifact “resnet18.pt2” which can be loaded and executed in Python and C++.

The artifact itself contains a bunch of AOTInductor generated code, such as a generated C++ runner file, a shared library compiled from the C++ file, and CUDA binary files, aka cubin files, if optimizing for CUDA.

Structure-wise, the artifact is a structured .zip file, with the following specification:

We can use the following command to inspect the artifact contents:

$ unzip -l resnet18.pt2

Archive:  resnet18.pt2
  Length      Date    Time    Name
---------  ---------- -----   ----
        1  01-08-2025 16:40   version
        3  01-08-2025 16:40   archive_format
    10088  01-08-2025 16:40   data/aotinductor/model/cagzt6akdaczvxwtbvqe34otfe5jlorktbqlojbzqjqvbfsjlge4.cubin
    17160  01-08-2025 16:40   data/aotinductor/model/c6oytfjmt5w4c7onvtm6fray7clirxt7q5xjbwx3hdydclmwoujz.cubin
    16616  01-08-2025 16:40   data/aotinductor/model/c7ydp7nocyz323hij4tmlf2kcedmwlyg6r57gaqzcsy3huneamu6.cubin
    17776  01-08-2025 16:40   data/aotinductor/model/cyqdf46ordevqhiddvpdpp3uzwatfbzdpl3auj2nx23uxvplnne2.cubin
    10856  01-08-2025 16:40   data/aotinductor/model/cpzfebfgrusqslui7fxsuoo4tvwulmrxirc5tmrpa4mvrbdno7kn.cubin
    14608  01-08-2025 16:40   data/aotinductor/model/c5ukeoz5wmaszd7vczdz2qhtt6n7tdbl3b6wuy4rb2se24fjwfoy.cubin
    11376  01-08-2025 16:40   data/aotinductor/model/csu3nstcp56tsjfycygaqsewpu64l5s6zavvz7537cm4s4cv2k3r.cubin
    10984  01-08-2025 16:40   data/aotinductor/model/cp76lez4glmgq7gedf2u25zvvv6rksv5lav4q22dibd2zicbgwj3.cubin
    14736  01-08-2025 16:40   data/aotinductor/model/c2bb5p6tnwz4elgujqelsrp3unvkgsyiv7xqxmpvuxcm4jfl7pc2.cubin
    11376  01-08-2025 16:40   data/aotinductor/model/c6eopmb2b4ngodwsayae4r5q6ni3jlfogfbdk3ypg56tgpzhubfy.cubin
    11624  01-08-2025 16:40   data/aotinductor/model/chmwe6lvoekzfowdbiizitm3haiiuad5kdm6sd2m6mv6dkn2zk32.cubin
    15632  01-08-2025 16:40   data/aotinductor/model/c3jop5g344hj3ztsu4qm6ibxyaaerlhkzh2e6emak23rxfje6jam.cubin
    25472  01-08-2025 16:40   data/aotinductor/model/chaiixybeiuuitm2nmqnxzijzwgnn2n7uuss4qmsupgblfh3h5hk.cubin
   139389  01-08-2025 16:40   data/aotinductor/model/cvk6qzuybruhwxtfblzxiov3rlrziv5fkqc4mdhbmantfu3lmd6t.cpp
       27  01-08-2025 16:40   data/aotinductor/model/cvk6qzuybruhwxtfblzxiov3rlrziv5fkqc4mdhbmantfu3lmd6t_metadata.json
 47195424  01-08-2025 16:40   data/aotinductor/model/cvk6qzuybruhwxtfblzxiov3rlrziv5fkqc4mdhbmantfu3lmd6t.so
---------                     -------
 47523148                     18 files

Model Inference in Python

To load and run the artifact in Python, we can use torch._inductor.aoti_load_package().

import os
import torch
import torch._inductor

model_path = os.path.join(os.getcwd(), "resnet18.pt2")

compiled_model = torch._inductor.aoti_load_package(model_path)
example_inputs = (torch.randn(2, 3, 224, 224, device=device),)

with torch.inference_mode():
    output = compiled_model(example_inputs)

When to use AOTInductor with a Python Runtime

There are mainly two reasons why one would use AOTInductor with a Python Runtime:

• torch._inductor.aoti_compile_and_package generates a singular serialized artifact. This is useful for model versioning for deployments and tracking model performance over time.

• With torch.compile() being a JIT compiler, there is a warmup cost associated with the first compilation. Your deployment needs to account for the compilation time taken for the first inference. With AOTInductor, the compilation is done ahead of time using torch.export.export and torch._inductor.aoti_compile_and_package. At deployment time, after loading the model, running inference does not have any additional cost.

The section below shows the speedup achieved with AOTInductor for first inference

We define a utility function timed to measure the time taken for inference

import time
def timed(fn):
    # Returns the result of running `fn()` and the time it took for `fn()` to run,
    # in seconds. We use CUDA events and synchronization for accurate
    # measurement on CUDA enabled devices.
    if torch.cuda.is_available():
        start = torch.cuda.Event(enable_timing=True)
        end = torch.cuda.Event(enable_timing=True)
        start.record()
    else:
        start = time.time()

    result = fn()
    if torch.cuda.is_available():
        end.record()
        torch.cuda.synchronize()
    else:
        end = time.time()

    # Measure time taken to execute the function in miliseconds
    if torch.cuda.is_available():
        duration = start.elapsed_time(end)
    else:
        duration = (end - start) * 1000

    return result, duration

Lets measure the time for first inference using AOTInductor

torch._dynamo.reset()

model = torch._inductor.aoti_load_package(model_path)
example_inputs = (torch.randn(1, 3, 224, 224, device=device),)

with torch.inference_mode():
    _, time_taken = timed(lambda: model(example_inputs))
    print(f"Time taken for first inference for AOTInductor is {time_taken:.2f} ms")

Time taken for first inference for AOTInductor is 3.86 ms

Lets measure the time for first inference using torch.compile

torch._dynamo.reset()

model = resnet18(weights=ResNet18_Weights.DEFAULT).to(device)
model.eval()

model = torch.compile(model)
example_inputs = torch.randn(1, 3, 224, 224, device=device)

with torch.inference_mode():
    _, time_taken = timed(lambda: model(example_inputs))
    print(f"Time taken for first inference for torch.compile is {time_taken:.2f} ms")

Time taken for first inference for torch.compile is 4578.57 ms

We see that there is a drastic speedup in first inference time using AOTInductor compared to torch.compile

Conclusion

In this recipe, we have learned how to effectively use the AOTInductor for Python runtime by compiling and loading a pretrained ResNet18 model. This process demonstrates the practical application of generating a compiled artifact and running it within a Python environment. We also looked at the advantage of using AOTInductor in model deployments, with regards to speed up in first inference time.