Getting Started with TVM command line driver - TVMC

Authors: Leandro Nunes, Matthew Barrett

This tutorial is an introduction to working with TVMC, the TVM command line driver. TVMC is a tool that exposes TVM features such as auto-tuning, compiling, profiling and execution of models, via a command line interface.

In this tutorial we are going to use TVMC to compile, run and tune a ResNet-50 on a x86 CPU.

We are going to start by downloading ResNet 50 V2. Then, we are going to use TVMC to compile this model into a TVM module, and use the compiled module to generate predictions. Finally, we are going to experiment with the auto-tuning options, that can be used to help the compiler to improve network performance.

The final goal is to give an overview of TVMC’s capabilities and also some guidance on where to look for more information.

Using TVMC

TVMC is a Python application, part of the TVM Python package. When you install TVM using a Python package, you will get TVMC as as a command line application called tvmc.

Alternatively, if you have TVM as a Python module on your $PYTHONPATH,you can access the command line driver functionality via the executable python module, python -m tvm.driver.tvmc.

For simplicity, this tutorial will mention TVMC command line using tvmc <options>, but the same results can be obtained with python -m tvm.driver.tvmc <options>.

You can check the help page using:

tvmc --help

As you can see in the help page, the main features are accessible via the subcommands tune, compile and run. To read about specific options under a given subcommand, use tvmc <subcommand> --help.

In the following sections we will use TVMC to tune, compile and run a model. But first, we need a model.

Obtaining the model

We are going to use ResNet-50 V2 as an example to experiment with TVMC. The version below is in ONNX format. To download the file, you can use the command below:



Supported model formats

TVMC supports models created with Keras, ONNX, TensorFlow, TFLite and Torch. Use the option``–model-format`` if you need to explicitly provide the model format you are using. See tvmc compile --help for more information.

Compiling the model

The next step once we’ve downloaded ResNet-50, is to compile it, To accomplish that, we are going to use tvmc compile. The output we get from the compilation process is a TAR package, that can be used to run our model on the target device.

tvmc compile \
  --target "llvm" \
  --output compiled_module.tar \

Once compilation finishes, the output compiled_module.tar will be created. This can be directly loaded by your application and run via the TVM runtime APIs.


Defining the correct target

Specifying the correct target (option --target) can have a huge impact on the performance of the compiled module, as it can take advantage of hardware features available on the target. For more information, please refer to Auto-tuning a convolutional network for x86 CPU.

In the next step, we are going to use the compiled module, providing it with some inputs, to generate some predictions.

Input pre-processing

In order to generate predictions, we will need two things:

  • the compiled module, which we just produced;

  • a valid input to the model

Each model is particular when it comes to expected tensor shapes, formats and data types. For this reason, most models require some pre and post processing, to ensure the input(s) is valid and to interpret the output(s).

In TVMC, we adopted NumPy’s .npz format for both input and output data. This is a well-supported NumPy format to serialize multiple arrays into a file.

We will use the usual cat image, similar to other TVM tutorials:

For our ResNet 50 V2 model, the input is expected to be in ImageNet format. Here is an example of a script to pre-process an image for ResNet 50 V2.

from import download_testdata
from PIL import Image
import numpy as np

img_url = ""
img_path = download_testdata(img_url, "imagenet_cat.png", module="data")

# Resize it to 224x224
resized_image =, 224))
img_data = np.asarray(resized_image).astype("float32")

# ONNX expects NCHW input, so convert the array
img_data = np.transpose(img_data, (2, 0, 1))

# Normalize according to ImageNet
imagenet_mean = np.array([0.485, 0.456, 0.406])
imagenet_stddev = np.array([0.229, 0.224, 0.225])
norm_img_data = np.zeros(img_data.shape).astype("float32")
for i in range(img_data.shape[0]):
    norm_img_data[i, :, :] = (img_data[i, :, :] / 255 - imagenet_mean[i]) / imagenet_stddev[i]

# Add batch dimension
img_data = np.expand_dims(norm_img_data, axis=0)

# Save to .npz (outputs imagenet_cat.npz)
np.savez("imagenet_cat", data=img_data)


File /workspace/.tvm_test_data/data/imagenet_cat.png exists, skip.

Running the compiled module

With both the compiled module and input file in hand, we can run it by invoking tvmc run.

tvmc run \
  --inputs imagenet_cat.npz \
  --output predictions.npz \

When running the above command, a new file predictions.npz should be produced. It contains the output tensors.

In this example, we are running the model on the same machine that we used for compilation. In some cases we might want to run it remotely via an RPC Tracker. To read more about these options please check tvmc run --help.

Output post-processing

As previously mentioned, each model will have its own particular way of providing output tensors.

In our case, we need to run some post-processing to render the outputs from ResNet 50 V2 into a more human-readable form.

The script below shows an example of the post-processing to extract labels from the output of our compiled module.

import os.path
import numpy as np

from scipy.special import softmax

from import download_testdata

# Download a list of labels
labels_url = ""
labels_path = download_testdata(labels_url, "synset.txt", module="data")

with open(labels_path, "r") as f:
    labels = [l.rstrip() for l in f]

output_file = "predictions.npz"

# Open the output and read the output tensor
if os.path.exists(output_file):
    with np.load(output_file) as data:
        scores = softmax(data["output_0"])
        scores = np.squeeze(scores)
        scores = np.argsort(scores)[::-1]

        for i in scores[0:5]:
            print("class='%s' with probability=%f" % (labels[i], scores[i]))


File /workspace/.tvm_test_data/data/synset.txt exists, skip.

When running the script, a list of predictions should be printed similar the the example below.

$ python
class=n02123045 tabby, tabby cat ; probability=446.000000
class=n02123159 tiger cat ; probability=675.000000
class=n02124075 Egyptian cat ; probability=836.000000
class=n02129604 tiger, Panthera tigris ; probability=917.000000
class=n04040759 radiator ; probability=213.000000

Tuning the model

In some cases, we might not get the expected performance when running inferences using our compiled module. In cases like this, we can make use of the auto-tuner, to find a better configuration for our model and get a boost in performance.

Tuning in TVM refers to the process by which a model is optimized to run faster on a given target. This differs from training or fine-tuning in that it does not affect the accuracy of the model, but only the runtime performance.

As part of the tuning process, TVM will try running many different operator implementation variants to see which perform best. The results of these runs are stored in a tuning records file, which is ultimately the output of the tune subcommand.

In the simplest form, tuning requires you to provide three things:

  • the target specification of the device you intend to run this model on;

  • the path to an output file in which the tuning records will be stored, and finally,

  • a path to the model to be tuned.

The example below demonstrates how that works in practice:

tvmc tune \
  --target "llvm" \
  --output autotuner_records.json \

Tuning sessions can take a long time, so tvmc tune offers many options to customize your tuning process, in terms of number of repetitions (--repeat and --number, for example), the tuning algorithm to be use, and so on. Check tvmc tune --help for more information.

As an output of the tuning process above, we obtained the tuning records stored in autotuner_records.json. This file can be used in two ways:

  • as an input to further tuning (via tvmc tune --tuning-records), or

  • as an input to the compiler

The compiler will use the results to generate high performance code for the model on your specified target. To do that we can use tvmc compile --tuning-records. Check tvmc compile --help for more information.

Final Remarks

In this tutorial, we presented TVMC, a command line driver for TVM. We demonstrated how to compile, run and tune a model, as well as discussed the need for pre and post processing of inputs and outputs.

Here we presented a simple example using ResNet 50 V2 locally. However, TVMC supports many more features including cross-compilation, remote execution and profiling/benchmarking.

To see what other options are available, please have a look at tvmc --help.

Gallery generated by Sphinx-Gallery