Have you ever ever copy-pasted chunks of utility code between initiatives, leading to a number of variations of the identical code dwelling in numerous repositories? Or, maybe, you needed to make pull requests to tens of initiatives after the title of the GCP bucket through which you retailer your information was up to date?

Conditions described above come up means too typically in ML groups, and their penalties range from a single developer’s annoyance to the crew’s lack of ability to ship their code as wanted. Fortunately, there’s a treatment.

Let’s dive into the world of monorepos, an structure broadly adopted in main tech corporations like Google, and the way they’ll improve your ML workflows. A monorepo gives a plethora of benefits which, regardless of some drawbacks, make it a compelling selection for managing complicated machine studying ecosystems. 

We’ll briefly debate monorepos’ deserves and demerits, look at why it’s a wonderful structure selection for machine studying groups, and peek into how BigTech is utilizing it. Lastly, we’ll see tips on how to harness the ability of the Pants construct system to prepare your machine studying monorepo into a sturdy CI/CD construct system. 

Strap in as we embark on this journey to streamline your ML mission administration.

What’s a monorepo?

What is ML monorepo? (short for monolithic repository)
Machine studying monorepo | Supply: Writer

A monorepo (quick for monolithic repository) is a software program improvement technique the place code for a lot of initiatives is saved in the identical repository. The concept may be as broad as all of the corporate code written in quite a lot of programming languages saved collectively (did someone say Google?) or as slim as a few Python initiatives developed by a small crew thrown right into a single repository. 

On this weblog submit, we deal with repositories storing machine studying code.

Monorepos vs. polyrepos

Monorepos are in stark distinction to the polyrepos strategy, the place every particular person mission or part has its personal separate repository. Lots has been mentioned in regards to the benefits and downsides of each approaches, and we received’t go down this rabbit gap too deep. Let’s simply put the fundamentals on the desk.

The monorepo structure gives the next benefits:

Monorepo architecture
Monorepo structure | Supply: Writer
  • Single CI/CD pipeline, which means no hidden deployment data unfold throughout particular person contributors to completely different repositories;
  • Atomic commits, given that each one initiatives reside in the identical repository, builders could make cross-project adjustments that span throughout a number of initiatives however are merged as a single commit;
  • Simple sharing of utilities and templates throughout initiatives;
  • Simple unification of coding requirements and approaches;
  • Higher code discoverability.

Naturally, there aren’t any free lunches. We have to pay for the above goodies, and the value comes within the type of:

  • Scalability challenges: Because the codebase grows, managing a monorepo can turn into more and more troublesome. At a very massive scale, you’ll want highly effective instruments and servers to deal with operations like cloning, pulling, and pushing adjustments, which may take a big period of time and sources.
  • Complexity: A monorepo may be extra complicated to handle, significantly with regard to dependencies and versioning. A change in a shared part might probably affect many initiatives, so additional warning is required to keep away from breaking adjustments.
  • Visibility and entry management: With everybody understanding of the identical repository, it may be troublesome to manage who has entry to what. Whereas not a drawback as such, it might pose issues of a authorized nature in circumstances the place code is topic to a really strict NDA.

The choice as as to if the benefits a monorepo gives are value paying the value is to be decided by every group or crew individually. Nonetheless, until you’re working at a prohibitively massive scale or are coping with top-secret missions, I might argue that – at the least relating to my space of experience, the machine studying initiatives – a monorepo is an efficient structure selection most often. 

Let’s discuss why that’s.

Machine studying with monorepos 

There are at the least six the reason why monorepos are significantly appropriate for machine studying initiatives.

  • 1
    Knowledge pipeline integration
  • 2
    Consistency throughout experiments
  • 3
    Simplified mannequin versioning
  • 4
    Cross-functional collaboration
  • 5
    Atomic adjustments
  • 6
    Unification of coding requirements

Knowledge pipeline integration

Machine studying initiatives typically contain information pipelines that preprocess, remodel, and feed information into the mannequin. These pipelines is likely to be tightly built-in with the ML code. Protecting the information pipelines and ML code in the identical repo helps keep this tight integration and streamline the workflow.

Consistency throughout experiments

Machine studying improvement includes a lot of experimentation. Having all experiments in a monorepo ensures constant setting setups and reduces the danger of discrepancies between completely different experiments because of various code or information variations.

Simplified mannequin versioning

In a monorepo, the code and mannequin variations are in sync as a result of they’re checked into the identical repository. This makes it simpler to handle and hint mannequin variations, which may be particularly necessary in initiatives the place ML reproducibility is crucial. 

Simply take the commit SHA at any given time limit, and it offers the data on the state of all fashions and companies.

Cross-functional collaboration

Machine studying initiatives typically contain collaboration between information scientists, ML engineers, and software program engineers. A monorepo facilitates this cross-functional collaboration by offering a single supply of fact for all project-related code and sources.

Atomic adjustments

Within the context of ML, a mannequin’s efficiency can rely on varied interconnected elements like information preprocessing, function extraction, mannequin structure, and post-processing. A monorepo permits for atomic adjustments – a change to a number of of those parts may be dedicated as one, making certain that interdependencies are at all times in sync.

Unification of coding requirements

Lastly, machine studying groups typically embody members and not using a software program engineering background. These mathematicians, statisticians, and econometricians are brainy of us with sensible concepts and the talents to coach fashions that resolve enterprise issues. Nonetheless, writing code that’s clear, straightforward to learn, and keep may not at all times be their strongest aspect. 

A monorepo helps by mechanically checking and implementing coding requirements throughout all initiatives, which not solely ensures excessive code high quality but in addition helps the much less engineering-inclined crew members be taught and develop.

How they do it in trade: well-known monorepos

Within the software program improvement panorama, a number of the largest and most profitable corporations on the planet use monorepos. Listed below are just a few notable examples.

  • Google: Google has lengthy been a staunch advocate for the monorepo strategy. Their total codebase, estimated to comprise 2 billion traces of code, is contained in a single, huge repository. They even published a paper about it.
  • Meta: Meta additionally employs a monorepo for his or her huge codebase. They created a model management system referred to as “Mercurial” to deal with the scale and complexity of their monorepo.
  • Twitter: Twitter has been managing their monorepo for a very long time utilizing Pants, the construct system we’ll discuss subsequent!

Many different corporations resembling Microsoft, Uber, Airbnb, and Stripe are using the monorepo approach at the least for some elements of their codebases, too.

Sufficient of the speculation! Let’s check out tips on how to truly construct a machine studying monorepo. As a result of simply throwing what was once separate repositories into one folder doesn’t do the job.

Tips on how to arrange ML monorepo with Python?

All through this part, we’ll base our dialogue on a sample machine learning repository I’ve created for this text. It’s a easy monorepo holding only one mission, or module: a hand-written digits classifier referred to as mnist, after the well-known dataset it makes use of. 

All you might want to know proper now could be that within the monorepo’s root there’s a listing referred to as mnist, and in it, there’s some Python code for coaching the mannequin, the corresponding unit assessments, and a Dockerfile to run coaching in a container.

ML monorepo: mnist directory

We might be utilizing this small instance to maintain issues easy, however in a bigger monorepo, mnist could be simply one of many many mission folders within the repo’s root, every of which is able to comprise supply code, assessments, dockerfiles, and requirement recordsdata as a minimum.

Construct system: Why do you want one and the way to decide on it?

The Why?

Take into consideration all of the actions, aside from writing code, that the completely different groups growing completely different initiatives inside the monorepo take as a part of their improvement workflow. They’d run linters in opposition to their code to make sure adherence to fashion requirements, run unit assessments, construct artifacts resembling docker containers and Python wheels, push them to exterior artifact repositories, and deploy them to manufacturing. 

Take testing. 

You’ve made a change in a utility perform you keep, ran the assessments, and all’s inexperienced. However how are you going to ensure your change is just not breaking code for different groups that is likely to be importing your utility? You need to run their check suite, too, in fact. 

However to do that, you might want to know precisely the place the code you modified is getting used. Because the codebase grows, discovering this out manually doesn’t scale nicely. After all, as a substitute, you may at all times execute all of the assessments, however once more: that strategy doesn’t scale very nicely.

Setting up ML monorepo and why do you need a system (testing)
Why do you want a system: testing | Supply: Writer

One other instance, manufacturing deployment

Whether or not you deploy weekly, every day, or repeatedly, when the time comes, you’ll construct all of the companies within the monorepo and push them to manufacturing. However hey, do you might want to construct all of them on every event? That may very well be time-consuming and costly at scale. 

Some initiatives may not have been up to date for weeks. Then again, the shared utility code they use might need obtained updates. How can we determine what to construct? Once more, it’s all about dependencies. Ideally, we might solely construct companies which have been affected by the current adjustments.

Setting up ML monorepo and why do you need a system (deployment)
Why do you want a system: deployment | Supply: Writer

All of this may be dealt with with a easy shell script with a small codebase, however because it scales and initiatives begin sharing code, challenges emerge, a lot of which revolve round dependency administration. 

Choosing the right system

The entire above is just not an issue anymore in the event you put money into a correct construct system. A construct system’s main job is to construct code. And it ought to achieve this in a intelligent means: the developer ought to solely want to inform it what to construct (“construct docker photos affected by my newest commit”, or “run solely these assessments that cowl code which makes use of the strategy I’ve up to date”), however the how needs to be left for the system to determine.

There are a few nice open-source construct methods on the market. Since most machine studying is finished in Python, let’s deal with those with the most effective Python assist. The 2 hottest selections on this regard are Bazel and Pants

Bazel is an open-source model of Google’s inside construct system, Blaze. Pants can be closely impressed by Blaze and it goals for related technical design objectives as Bazel. An reader will discover a good comparability of Pants vs. Bazel on this blog post (however take note it comes from the Pants devs). The desk on the backside of monorepo.tools gives yet one more comparability.

Each methods are nice, and it’s not my intention to declare a “higher” resolution right here. That being mentioned, Pants is usually described as simpler to arrange, extra approachable, and well-optimized for Python, which makes it an ideal match for machine studying monorepos. 

In my private expertise, the decisive issue that made me go along with Pants was its lively and useful group. Every time you have got questions or doubts, simply submit on the group Slack channel, and a bunch of supportive of us will assist you out quickly.

Introducing Pants 

Alright, time to get to the meat of it! We’ll go step-by-step, introducing completely different Pants’ functionalities and tips on how to implement them. Once more, you may take a look at the related pattern repo here.


Pants is installable with pip. On this tutorial, we’ll use the latest secure model as of this writing, 2.15.1.

pip set up pantsbuild.pants==2.15.1

Pants is configurable via a worldwide grasp config file named pants.toml. In it, we will configure Pants’ personal habits in addition to the settings of downstream instruments it depends on, resembling pytest or mypy.

Let’s begin with a naked minimal pants.toml:


pants_version = "2.15.1"

backend_packages = [




root_patterns = ["/"]


interpreter_constraints = ["==3.9.*"]

Within the world part, we outline the Pants model and the backend packages we want. These packages are Pants’ engines that assist completely different options. For starters, we solely embody the Python backend.

Within the supply part, we set the supply to the repository’s root. Since model 2.15, to verify that is picked up, we additionally want so as to add an empty BUILD_ROOT file on the repository’s root.

Lastly, within the Python part, we select the Python model to make use of. Pants will browse our system seeking a model that matches the circumstances specified right here, so be sure to have this model put in.

That’s a begin! Subsequent, let’s check out any construct system’s coronary heart: the BUILD recordsdata.

Construct recordsdata

Construct recordsdata are configuration recordsdata used to outline targets (what to construct) and their dependencies (what they should work) in a declarative means. 

You’ll be able to have a number of construct recordsdata at completely different ranges of the listing tree. The extra there are, the extra granular the management over dependency administration. The truth is, Google has a construct file in nearly each listing of their repo. 

In our instance, we’ll use three construct recordsdata:

  • mnist/BUILD – within the mission listing, this construct file will outline the python necessities for the mission and the docker container to construct;
  • mnist/src/BUILD – within the supply code listing, this construct file will outline python sources, that’s, recordsdata to be coated by python-specific checks;
  • mnist/assessments/BUILD – within the assessments listing, this construct file will outline which recordsdata to run with Pytest and what dependencies are wanted for these assessments to run.

Let’s check out the mnist/src/BUILD:






On the similar time, mnist/BUILD appears to be like like this:






The 2 entries within the construct recordsdata are known as targets. First, now we have a Python sources goal, which we aptly name python, though the title may very well be something. We outline our Python sources as all .py recordsdata within the listing. That is relative to the construct file’s location, that’s: even when we had Python recordsdata outdoors of the mnist/src listing, these sources solely seize the contents of the mnist/src folder. There may be additionally a resolve filed; we’ll discuss it in a second.

Subsequent, now we have the Python necessities goal. It tells Pants the place to search out the necessities wanted to execute our Python code (once more, relative to the construct file’s location, which is within the mnist mission’s root on this case).

That is all we have to get began. To ensure the construct file definition is right, let’s run:

pants tailor --check update-build-files --check ::

As anticipated, we get: “No required adjustments to BUILD recordsdata discovered.” because the output. Good!

Let’s spend a bit extra time on this command. In a nutshell, a naked pants tailor can mechanically create construct recordsdata. Nonetheless, it generally tends so as to add too many for one’s wants, which is why I have a tendency so as to add them manually, adopted by the command above that checks their correctness. 

The double semicolon on the finish is a Pants notation that tells it to run the command over all the monorepo. Alternatively, we might have changed it with mnist: to run solely in opposition to the mnist module. 

Dependencies and lockfiles

To do environment friendly dependency administration, pants depends on lockfiles. Lockfiles file the particular variations and sources of all dependencies utilized by every mission. This consists of each direct and transitive dependencies. 

By capturing this data, lockfiles be certain that the identical variations of dependencies are used constantly throughout completely different environments and builds. In different phrases, they function a snapshot of the dependency graph, making certain reproducibility and consistency throughout builds.

To generate a lockfile for our mnist module, we want the next addition to pants.toml:

interpreter_constraints = ["==3.9.*"]
enable_resolves = true
default_resolve = "mnist"

mnist = "mnist/mnist.lock"

We allow the resolves (Pants time period for lockfiles’ environments) and outline one for mnist passing a file path. We additionally select it because the default one. That is the resolve now we have handed to Python sources and Python necessities goal earlier than: that is how they know what dependencies are wanted. We will now run:

to get:

Accomplished: Generate lockfile for mnist
Wrote lockfile for the resolve `mnist` to mnist/mnist.lock

This has created a file at mnist/mnist.lock. This file needs to be checked with git in the event you intend to make use of Pants in your distant CI/CD. And naturally, it must be up to date each time you replace the necessities.txt file. 

With extra initiatives within the monorepo, you’ll quite generate the lockfiles selectively for the mission that wants it, e.g. pants generate-lockfiles mnist: .

That’s it for the setup! Now let’s use Pants to do one thing helpful for us.

Unifying code fashion with Pants

Pants natively helps quite a few Python linters and code formatting instruments resembling Black, yapf, Docformatter, Autoflake, Flake8, isort, Pyupgrade, or Bandit. They’re all utilized in the identical means; in our instance, let’s implement Black and Docformatter.

To take action, we add acceptable two backends to pants.toml:

pants_version = "2.15.1"
colours = true
backend_packages = [

We might configure each instruments if we needed to by including further sections under within the toml file, however let’s keep on with the defaults now.

To make use of the formatters, we have to execute what’s referred to as a Pants purpose. On this case, two objectives are related.

First, the lint purpose will run each instruments (within the order through which they’re listed in backend packages, so Docformatter first, Black second) within the verify mode.

pants lint ::

Accomplished: Format with docformatter - docformatter made no adjustments.
Accomplished: Format with Black - black made no adjustments.

✓ black succeeded.
✓ docformatter succeeded.

It appears to be like like our code adheres to the requirements of each formatters! Nonetheless, if that was not the case, we might execute the fmt (quick for “format”) purpose that adapts the code appropriately:

In apply, you would possibly need to use greater than these two formatters. On this case, you might have to replace every formatter’s config to make sure that it’s appropriate with the others. As an example, if you’re utilizing Black with its default config as now we have executed right here, it’ll count on code traces to not exceed 88 characters. 

However in the event you then need to add isort to mechanically type your imports, they’ll conflict: isort truncates traces after 79 characters. To make isort appropriate with Black, you would wish to incorporate the next part within the toml file:

args = [

All formatters may be configured in the identical means in pants.toml by passing the arguments to their underlying instrument.

Testing with Pants

Let’s run some assessments! To do that, we want two steps.

First, we add the suitable sections to pants.toml:

output = "all"
report = false
use_coverage = true

global_report = true

args = ["-vv", "-s", "-W ignore::DeprecationWarning", "--no-header"]

These settings ensure that because the assessments are run, a check protection report is produced. We additionally cross a few customized pytest choices to adapt its output.

Subsequent, we have to return to our mnist/assessments/BUILD file and add a Python assessments goal:


We name it assessments and specify the resolve (i.e. lockfile) to make use of. Sources are the areas the place pytest might be let in to search for assessments to run; right here, we explicitly cross all .py recordsdata prefixed with “test_”. 

Now we will run:

pants check ::

to get:

✓ mnist/assessments/test_data.py:../assessments succeeded in 3.83s.
✓ mnist/assessments/test_model.py:../assessments succeeded in 2.26s.

Identify                               Stmts   Miss  Cowl
__global_coverage__/no-op-exe.py       0      0   100%
mnist/src/information.py                     14      0   100%
mnist/src/mannequin.py                    15      0   100%
mnist/assessments/test_data.py              21      1    95%
mnist/assessments/test_model.py             20      1    95%
TOTAL                                 70      2    97%

As you may see, it took round three seconds to run this check suite. Now, if we re-run it once more, we’ll get the outcomes instantly:

✓ mnist/assessments/test_data.py:../assessments succeeded in 3.83s (memoized).
✓ mnist/assessments/test_model.py:../assessments succeeded in 2.26s (memoized).

Discover how Pants tells us these outcomes are memoized, or cached. Since no adjustments have been made to the assessments, the code being examined, or the necessities,  there isn’t a want to really re-run the assessments – their outcomes are assured to be the identical, so they’re simply served from the cache. 

Checking static typing with Pants

Let’s add yet one more code high quality verify. Pants enable utilizing mypy to verify static typing in Python. All we have to do is add the mypy backend in pants.toml: “pants.backend.python.typecheck.mypy”.

You may additionally need to configure mypy to make its output extra readable and informative by additionally including the next config part:

args = [

With this, we will run pants verify :: to get:

Accomplished: Typecheck utilizing MyPy - mypy - mypy succeeded.
Success: no points discovered in 6 supply recordsdata

✓ mypy succeeded.

Delivery ML fashions with Pants

Let’s speak transport. Most machine studying initiatives contain a number of docker containers, for instance, processing coaching information, coaching a mannequin, or serving it by way of an API utilizing Flask or FastAPI. In our toy mission, we even have a container for model training.

Pants assist computerized constructing and pushing of docker photos. Let’s see the way it works.

First, we add the docker backend in pants.toml: pants.backend.docker. We may even configure our docker, passing it quite a few setting variables and a construct arg which is able to come in useful in a second:


build_args = ["SHORT_SHA"]

env_vars = ["DOCKER_CONFIG=%(env.HOME)s/.docker", "HOME", "USER", "PATH"]

Now, in the mnist/BUILD file, we'll add two extra targets: a recordsdata goal and a docker picture goal.










    image_tags=["latest", "{build_args.SHORT_SHA}"],


We name the docker goal “train_mnist”. As a dependency, we have to cross it the listing of recordsdata to be included within the container. Probably the most handy means to do that is to outline this listing as a separated recordsdata goal. Right here, we merely embody all of the recordsdata within the mnist mission in a goal referred to as module_files, and cross it as a dependency to the docker picture goal.

Naturally, if you recognize that just some subset of recordsdata might be wanted by the container, it’s a good suggestion to cross solely them as a dependency. It’s important as a result of these dependencies are utilized by Pants to deduce whether or not a container has been affected by a change and desires a rebuild. Right here, with module_files together with all recordsdata, if any file within the mnist folder adjustments (even a readme!), Pants will see the train_mnist docker picture as affected by this variation.

Lastly, we will additionally set the exterior registry and repository to which the picture may be pushed, and the tags with which will probably be pushed: right here, I might be pushing the picture to my private dockerhub repo, at all times with two tags: “newest”, and the quick commit SHA which might be handed as a construct arg.

With this, we will construct a picture. Only one other thing: since Pants is working in its remoted environments, it can not learn env vars from the host. Therefore, to construct or push the picture that requires the SHORT_SHA variable, we have to cross it along with the Pants command.

We will construct the picture like this:

SHORT_SHA=$(git rev-parse --short HEAD) pants bundle mnist:train_mnist 

to get:

Accomplished: Constructing docker picture docker.io/michaloleszak/mnist:newest +1 further tag.
Constructed docker photos: 
  * docker.io/michaloleszak/mnist:newest
  * docker.io/michaloleszak/mnist:0185754

A fast verify reveals that the photographs have certainly been constructed:

docker photos 

REPOSITORY            TAG       IMAGE ID       CREATED              SIZE
michaloleszak/mnist   0185754   d86dca9fb037   A couple of minute in the past   3.71GB
michaloleszak/mnist   newest    d86dca9fb037   A couple of minute in the past   3.71GB

We will additionally construct and push photos in a single go utilizing Pants. All it takes is changing the bundle command with the publish command.

SHORT_SHA=$(git rev-parse --short HEAD) pants publish mnist:train_mnist 

This constructed the photographs and pushed them to my dockerhub, where they have indeed landed.

Pants in CI/CD

The identical instructions now we have simply manually run regionally may be executed as elements of a CI/CD pipeline. You’ll be able to run them by way of companies resembling GitHub Actions or Google CloudBuild, as an example as a PR verify earlier than a function department is allowed to be merged to the primary department, or after the merge, to validate it’s inexperienced and construct & push containers.

In our toy repo, I’ve carried out a pre-push commit hook that runs Pants instructions on git push and solely lets it via if all of them cross. In it, we’re working the next instructions:

pants tailor --check update-build-files --check ::
pants lint ::
pants --changed-since=fundamental --changed-dependees=transitive verify
pants check ::

You’ll be able to see some new flags for pants verify, that’s the typing verify with mypy. They be certain that the verify is simply run on recordsdata which have modified in comparison with the primary department and their transitive dependencies. That is helpful since mypy tends to take a while to run. Limiting its scope to what’s truly wanted accelerates the method.

How would a docker construct & push look in a CI/CD pipeline? Considerably like this:

pants --changed-since=HEAD^ --changed-dependees=transitive --filter-target-type=docker_image publish

We use the publish command as earlier than, however with three further arguments:

  • –changed-since=HEAD^ and –changed-dependees=transitive ensure that solely the containers affected by the adjustments in comparison with the earlier commit are constructed; that is helpful for executing on the primary department after the merge.
  • –filter-target-type=docker_image makes certain that the one issues Pants does is construct and push docker; it’s because the pants publish command can confer with targets aside from docker: for instance, it may be used to publish helm charts to OCI registries. 

The identical goes for the pants bundle: on prime of constructing docker photos, it will possibly additionally create a Python bundle; for that motive, it’s a superb apply to cross the –filter-target-type choice.


Monorepos are most of the time a fantastic structure selection for machine studying groups. Managing them at scale, nevertheless, requires funding in a correct construct system. One such system is Pants: it’s straightforward to arrange and use and gives native assist for a lot of Python and Docker options that machine studying groups typically use. 

On prime of that, it’s an open-source mission with a big and useful group. I hope after studying this text you’ll go forward and check out it out. Even in the event you don’t presently have a monolithic repository, Pants can nonetheless streamline and facilitate many points of your every day work!


Discover extra content material matters:

Leave a Reply

Your email address will not be published. Required fields are marked *