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%|████████▉ | 40.0M/44.7M [00:00<00:00, 419MB/s]
100%|██████████| 44.7M/44.7M [00:00<00:00, 420MB/s]
/usr/local/lib/python3.10/dist-packages/torch/_inductor/compile_fx.py:236: UserWarning:

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

AUTOTUNE convolution(2x3x224x224, 64x3x7x7)
  convolution 0.1013 ms 100.0%
  triton_convolution2d_4 0.1505 ms 67.3% 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=8
  triton_convolution2d_0 0.1536 ms 65.9% 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=4
  triton_convolution2d_3 0.1966 ms 51.5% 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 0.2386 ms 42.4% 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 0.3052 ms 33.2% 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 0.4096 ms 24.7% 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=4
SingleProcess AUTOTUNE benchmarking takes 0.4283 seconds and 0.0062 seconds precompiling for 7 choices
AUTOTUNE convolution(2x64x56x56, 64x64x3x3)
  triton_convolution2d_11 0.0369 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_6 0.0378 ms 97.5% 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_10 0.0399 ms 92.3% 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 0.0410 ms 90.0% 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
  convolution 0.0563 ms 65.5%
  triton_convolution2d_7 0.0604 ms 61.0% 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 0.0655 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
  triton_convolution2d_8 0.1208 ms 30.5% 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 0.1806 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x64x56x56, 128x64x3x3)
  triton_convolution2d_38 0.0307 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 0.0358 ms 85.7% 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 0.0492 ms 62.5% 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 0.0573 ms 53.6% 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
  triton_convolution2d_40 0.0614 ms 50.0% 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 47.6% 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 0.0676 ms 45.5%
  triton_convolution2d_36 0.1126 ms 27.3% 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 0.1797 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x128x28x28, 128x128x3x3)
  triton_convolution2d_45 0.0492 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
  convolution 0.0541 ms 90.8%
  triton_convolution2d_46 0.0737 ms 66.7% 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
  triton_convolution2d_41 0.0891 ms 55.2% 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 0.1096 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 0.1147 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 0.1197 ms 41.0% 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 0.2273 ms 21.6% 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 0.2107 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x64x56x56, 128x64x1x1)
  triton_convolution2d_52 0.0082 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_53 0.0093 ms 87.7% 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_48 0.0113 ms 72.7% 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_49 0.0133 ms 61.5% 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 0.0133 ms 61.5% 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_54 0.0133 ms 61.5% 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
  convolution 0.0143 ms 57.1%
  triton_convolution2d_50 0.0184 ms 44.4% 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
SingleProcess AUTOTUNE benchmarking takes 0.1485 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x128x28x28, 256x128x3x3)
  triton_convolution2d_73 0.0461 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 0.0481 ms 95.7%
  triton_convolution2d_69 0.1065 ms 43.3% 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 0.1065 ms 43.3% 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 0.1085 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 0.1157 ms 39.8% 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 0.1208 ms 38.1% 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 0.1782 ms 25.9% 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 0.2121 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x256x14x14, 256x256x3x3)
  convolution 0.0717 ms 100.0%
  triton_convolution2d_80 0.0922 ms 77.8% 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 0.2038 ms 35.2% 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=1, STRIDE_W=1, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_76 0.2048 ms 35.0% 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 0.2130 ms 33.7% 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 0.2130 ms 33.7% 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 0.2232 ms 32.1% 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 0.2324 ms 30.8% 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 0.2766 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x128x28x28, 256x128x1x1)
  triton_convolution2d_87 0.0093 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
  convolution 0.0174 ms 53.5%
  triton_convolution2d_86 0.0184 ms 50.5% 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_88 0.0184 ms 50.5% 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_89 0.0184 ms 50.5% 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_83 0.0205 ms 45.5% 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 0.0225 ms 41.3% 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 0.0256 ms 36.4% 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
SingleProcess AUTOTUNE benchmarking takes 0.1466 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x256x14x14, 512x256x3x3)
  convolution 0.0543 ms 100.0%
  triton_convolution2d_108 0.0901 ms 60.2% 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 0.2068 ms 26.2% 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 0.2099 ms 25.9% 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 0.2150 ms 25.2% 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_110 0.2324 ms 23.3% 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=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_107 0.2365 ms 22.9% 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=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=8
  triton_convolution2d_105 0.2733 ms 19.9% 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=2, STRIDE_W=2, UNROLL=False, num_stages=2, num_warps=4
SingleProcess AUTOTUNE benchmarking takes 0.2773 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x512x7x7, 512x512x3x3)
  convolution 0.0942 ms 100.0%
  triton_convolution2d_115 0.1824 ms 51.6% 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_113 0.2161 ms 43.6% 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 0.2693 ms 35.0% 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_112 0.2867 ms 32.9% 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 0.4116 ms 22.9% 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 0.4220 ms 22.3% 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 0.4424 ms 21.3% 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 0.2970 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE convolution(2x256x14x14, 512x256x1x1)
  triton_convolution2d_122 0.0143 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 0.0256 ms 56.0% 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 0.0276 ms 51.9% 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_118 0.0287 ms 50.0% 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_124 0.0297 ms 48.3% 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_121 0.0307 ms 46.8% 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_123 0.0316 ms 45.4% 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 0.0338 ms 42.4%
SingleProcess AUTOTUNE benchmarking takes 0.1548 seconds and 0.0002 seconds precompiling for 8 choices
AUTOTUNE addmm(2x1000, 2x512, 512x1000)
  addmm 0.0143 ms 100.0%
  triton_mm_142 0.0215 ms 66.7% ACC_TYPE='tl.float32', ALLOW_TF32=False, BLOCK_K=32, BLOCK_M=16, BLOCK_N=32, EVEN_K=True, GROUP_M=8, num_stages=5, num_warps=2
  triton_mm_143 0.0225 ms 63.6% ACC_TYPE='tl.float32', ALLOW_TF32=False, BLOCK_K=128, BLOCK_M=16, BLOCK_N=32, EVEN_K=True, GROUP_M=8, num_stages=5, num_warps=2
  triton_mm_140 0.0236 ms 60.9% ACC_TYPE='tl.float32', ALLOW_TF32=False, BLOCK_K=128, BLOCK_M=16, BLOCK_N=32, EVEN_K=True, GROUP_M=8, num_stages=2, num_warps=2
  triton_mm_146 0.0297 ms 48.3% ACC_TYPE='tl.float32', ALLOW_TF32=False, BLOCK_K=64, BLOCK_M=16, BLOCK_N=64, EVEN_K=True, GROUP_M=8, num_stages=3, num_warps=4
  triton_mm_152 0.0297 ms 48.3% ACC_TYPE='tl.float32', ALLOW_TF32=False, BLOCK_K=32, BLOCK_M=16, BLOCK_N=64, EVEN_K=True, GROUP_M=8, num_stages=3, num_warps=4
  triton_mm_153 0.0297 ms 48.3% ACC_TYPE='tl.float32', ALLOW_TF32=False, BLOCK_K=32, BLOCK_M=16, BLOCK_N=64, EVEN_K=True, GROUP_M=8, num_stages=4, num_warps=4
  triton_mm_141 0.0307 ms 46.7% ACC_TYPE='tl.float32', ALLOW_TF32=False, BLOCK_K=32, BLOCK_M=16, BLOCK_N=64, EVEN_K=True, GROUP_M=8, num_stages=5, num_warps=4
  triton_mm_139 0.0328 ms 43.8% ACC_TYPE='tl.float32', ALLOW_TF32=False, BLOCK_K=16, BLOCK_M=16, BLOCK_N=32, EVEN_K=True, GROUP_M=8, num_stages=1, num_warps=2
  triton_mm_145 0.0369 ms 38.9% ACC_TYPE='tl.float32', ALLOW_TF32=False, BLOCK_K=32, BLOCK_M=16, BLOCK_N=64, EVEN_K=True, GROUP_M=8, num_stages=2, num_warps=4
SingleProcess AUTOTUNE benchmarking takes 0.4975 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 3.38 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 3399.16 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.