MAX FAQ
We tried to anticipate your questions about MAX Engine on this page. If this page doesn't answer all your questions, please ask us on our Discord channel.
Distribution
What are the system requirements?
- Mac
- Linux
- Windows
- Apple silicon (M1/M2/M3 processor)
- macOS Ventura (13) or later
- Python 3.9 - 3.12
- Xcode or Xcode Command Line Tools
- Homebrew
- Ubuntu 22.04 LTS
- x86-64 CPU (with SSE4.2 or newer) or AWS Graviton2/3 CPU
- Minimum 8 GiB RAM
- Python 3.9 - 3.12
- g++ or clang++ C++ compiler
Windows support is still in development.
In the meantime, you can use MAX on Windows with WSL, using a compatible version of Ubuntu (see our requirements for Linux).
What versions of torch are supported?
MAX supports torch from version 2.2.2 through 2.4.0.
MAX currently defaults to use CPU-only torch, but CUDA/ROCm supporting torch versions should work as intended.
Will it be open-sourced?
We want to contribute a lot to open source, but we also want to do it right. Our team has decades of experience building open-source projects, and we have learned that the important thing is to create an inclusive and vibrant community – and that takes a lot of work. We will need to figure out the details, but as we do so, we will share more. Please stay tuned.
Why bundle Mojo with MAX?
Integrating Mojo and MAX into a single package is the best way to ensure interoperability between Mojo and MAX for all users, and avoid version conflicts that happen when installing them separately.
We built Mojo as a core technology for MAX, and you can use it to extend MAX Engine, so MAX clearly depends on Mojo. On the other hand, writing Mojo code that runs on both CPUs and GPUs (and other accelerators) requires runtime components and orchestration logic that falls outside the domain of Mojo, and into the domain of MAX. That is, MAX isn't just a framework for AI development, it's also a framework for general heterogeneous compute. As such, writing Mojo programs that can execute across heterogeneous hardware depends on MAX.
Nothing has changed for Mojo developers—you can still build and develop in Mojo like you always have. The only difference is that you're now able to seamlessly step into general-purpose GPU programming (coming soon).
Does the MAX SDK collect telemetry?
Yes, the MAX SDK collects basic system information, session durations, compiler events, and crash reports that enable us to identify, analyze, and prioritize issues.
This telemetry is crucial to help us quickly identify problems and improve our products. Without this telemetry, we would have to rely on user-submitted bug reports, and in our decades of experience building developer products, we know that most people don’t do that. The telemetry provides us the insights we need to build better products for you.
You can opt-out of some telemetry, such as compiler events and crash reports. However, package install/update/uninstall events, basic system information,and session durations (the amount of time spent running MAX Engine) cannot be disabled (see the MAX SDK terms).
If you installed with magic
When using Magic, telemetry is configured separately for each project. To disable telemetry for compiler events and crash reports, run this command in your project environment:
magic telemetry --disable
magic telemetry --disable
For details, see the magic telemetry
docs.
If you installed with modular
(deprecated)
To disable crash reports, use this command:
modular config-set crash_reporting.enabled=false
modular config-set crash_reporting.enabled=false
To reduce other telemetry to only the required telemetry events, use this command:
modular config-set telemetry.level=0
modular config-set telemetry.level=0
There are 3 telemetry levels: 0
records only the required details such as
hardware information and session durations; 1
records high-level events such
as when the compiler is invoked; and 2
records more specific events such as
when the compiler uses a framework fallback op.
Installation
Can I install both stable and nightly builds?
Yes, when you use the Magic CLI, it's safe and easy to use the stable and nightly builds for different projects, each with their own virtual environment and package dependencies.
To install nightlies, see how to create a project with MAX or get started with Mojo.
How do I uninstall everything?
It depends on whether you installed using the modular
CLI
(deprecated) or the magic
CLI.
If you installed with magic
To remove the MAX/Mojo packages you've installed with magic
, delete the
project paths that you created with magic init
or the paths that contain a
pixi.toml
or mojoproject.toml
file.
To remove the magic
tool, delete the magic
binary:
rm ~/.modular/bin/magic
rm ~/.modular/bin/magic
That's it.
If you installed with modular
(deprecated)
To uninstall everything, including the modular
tool itself, first use the
modular uninstall
command to remove the max
or mojo
packages and configurations:
modular uninstall max
modular uninstall max
That removes the executable tools and configurations, but it leaves the MAX Python package, the Modular CLI tool, and some other files.
To remove everything else, use these commands:
-
Uninstall the MAX Python package. You can do this in one of two ways:
-
Delete your virtual environment completely, which removes the
max-engine
Python packages along with any other dependencies. For example:- venv
- conda
rm -rf ~/max-venv
rm -rf ~/max-venv
conda env remove --name max
conda env remove --name max
-
Or, keep the virtual environment and delete just the
max-engine
packages inside:- venv
- conda
source ~/max-venv/bin/activate
source ~/max-venv/bin/activate
conda activate max
conda activate max
pip uninstall max-engine max-engine-libs -y
pip uninstall max-engine max-engine-libs -y
-
-
Uninstall the
modular
CLI:- Mac
- Linux
brew uninstall modular
brew uninstall modular
sudo apt update && sudo apt remove modular -y
sudo apt update && sudo apt remove modular -y
-
Delete the Modular home directory:
rm -rf ~/.modular
rm -rf ~/.modular
That should do it for situations in which you're troubleshooting your installation and want to reinstall from scratch.
If you want to remove absolutely all trace of MAX, here are a few more actions:
-
Delete the lines that mention
MODULAR_HOME
and.modular
in~/.bash_rc
,~/.bash_profile
, or~/.zshrc
. -
Remove the Modular repository name:
- Mac
- Linux
brew untap modularml/packages/modular
brew untap modularml/packages/modular
sudo rm -rf /etc/apt/sources.list.d/modular-installer.list
sudo rm -rf /etc/apt/sources.list.d/modular-installer.list
-
If you used Jupyter notebooks with Mojo, delete the Mojo kernels:
jupyter kernelspec list
jupyter kernelspec list
jupyter kernelspec remove <KERNEL_NAME>
jupyter kernelspec remove <KERNEL_NAME>
-
Uninstall the Mojo VS Code extension via the VS Code extensions tab.
Functionality
How do I use MAX Engine?
You can execute a trained model using our Python, C, and Mojo API libraries. Learn more about our APIs and view sample code in the MAX Engine docs.
Why does MAX Engine take so long to "load" a model?
The first time you load your model (such as with the Python
load()
function), MAX Engine must first compile the model.
This might seem strange if you're used to "eager execution" in PyTorch or TensorFlow, but this compilation step is how MAX Engine optimizes the graph to deliver more performance. This is an up-front cost that occurs only when you first load the model, and it pays dividends with major latency savings provided by our next-generation graph compiler. So it's worth the wait. :)
What types of models does MAX Engine support and from which training frameworks?
MAX Engine supports most models from PyTorch (in TorchScript format), ONNX, and TensorFlow (in SavedModel format).
However, TensorFlow support is available for enterprise customers only and no longer included in the MAX SDK. We removed TensorFlow support in the MAX SDK because industry-wide TensorFlow usage has declined significantly, especially for the latest AI innovations. Also, it cuts our package size by over 50% and accelerates the development of other customer-requested features. If you have a production use-case for a TensorFlow model, please contact us.
What hardware is currently supported by MAX Engine?
MAX Engine currently supports all CPU types from major vendors (Intel, AMD, Graviton).
Support for NVIDIA GPUs is in the works.
Does MAX Engine support generic ARM architectures?
Yes, but we officially support only Graviton because it’s the most commonly used ARM chip for server deployments, and our benchmarks are designed to match what users use most often in production.
We also support Apple Silicon (M1/M2) for local development with MAX.
Which programming languages does it support? Can I use Mojo?
We currently provide MAX Engine API bindings in Python, C, and—yes—Mojo. If there are other languages you’d like to see us support, please share in our Discord.
How quickly will it support new model architectures as they become available in ML frameworks?
One of the design principles of MAX Engine is full compatibility with models from supported ML frameworks. Hence, we will quickly support new operators as soon as a stable version of the training framework is available.
Will I need different packages depending on the platform or hardware where my code needs to be deployed?
No. We believe forcing developers to deploy and manage different packages for different deployment targets is a major friction factor that shouldn’t exist. The same engine package works regardless of the hardware available in your deployment platform.
Will I need to change my code if I want to switch deployment to hardware from a different manufacturer?
No. With MAX Engine, you write your code once and deploy anywhere. For example, if you are currently running on an Intel instance in AWS but want to experiment with a Graviton instance, that’s no problem. Just redeploy the same engine package to the Graviton instance, and you’re good to go.
Can I use Modular’s MAX Engine on existing cloud platforms?
Yes. You can deploy MAX Engine to infrastructure provided by any major cloud platform (AWS, GCP, Azure) via traditional container solutions. For more information, read about MAX Serve.
Does MAX Engine support quantization and sparsity?
It supports some quantized models today (models with Int data types, and we are working to add support for more) and we will be adding support for sparsity soon.
If you're building a model with MAX Graph, you can quantize your model with several different quantization encodings.
Will MAX Engine support distributed inference of large models?
Yes, it will support executing large models that do not fit into the memory of a single device. This isn't available yet, so stay tuned!
Will MAX Engine support mobile?
We are currently focused on server deployment, but we plan to support deployment to many different platforms, including mobile. We will share more about our mobile support in the future, so stay tuned!
Can I extend MAX Engine with a new hardware backend?
Yes, you can. MAX Engine can support other hardware backends, including specialized accelerators, via Mojo, which can talk directly to MLIR and LLVM abstractions (for an example of Mojo talking directly to MLIR, see our low-level IR in Mojo notebook). The exact details about how to add additional hardware support are still being ironed out, but you can read our vision for pluggable hardware.
Can I integrate ops written in Mojo into MAX Engine?
We're working on it. Learn more about our extensibility API for MAX Engine.
Can the runtime be separated from model compilation (for edge deployment)?
Yes, our runtime (and our entire stack) is designed to be modular. It scales down very well, supports heterogeneous configs, and scales up to distributed settings as well. That being said, this isn't available yet, but we'll share details about more deployment scenarios we support over time.
Performance
Why does MAX Engine perform slowly on my computer compared to PyTorch?
So far, we've been focused on building optimizations for data center CPUs. In many cases, these optimizations carry over to desktop x86 CPUs. However, modern desktop/laptop CPUs include specialty compute cores not found in data center CPUs, such as performance cores ("P cores," which are often paired with efficiency "E cores" in an "asymmetric core"). If your CPU includes these P/E cores, MAX Engine simply doesn't use them thoughtfully yet. We have plans to fix this.
Additionally, we haven't optimized for systems with 32+ cores yet. This will also be fixed soon.
If you're seeing slow results that aren't explained here, please let us know.
Why do you only show performance numbers on the CPU?
Our stack is completely extensible to any type of hardware architecture, from commodity silicon to newer types of AI-specific accelerators. We are starting with CPUs because many real-world inference workloads still heavily depend on CPUs. Our stack currently supports x86-64 CPUs from all major hardware vendors, such as Intel and AMD. It also supports the Graviton CPUs available in AWS. We are actively bringing up support for GPU execution, starting with NVIDIA’s GPU. We will share more about our GPU support soon.
Why test with batch size 1?
We started our benchmarking with batch size 1 for a couple reasons: 1) it’s a common batch size for production inference scenarios and 2) it puts runtime efficiencies front-and-center, which helps ensure we are building the most performant possible stack. We have also tested and seen the same relative performance improvements with larger batch sizes. We’ll be releasing those results on our performance dashboard in the near future.
Do you have any benchmarks on GPUs?
We are currently working on adding GPU support. Stay tuned for benchmarks in the near future.
Why are you benchmarking across so many different sequence lengths?
Production NLP deployment scenarios typically involve variable sequence lengths. One of the defining features of MAX Engine is that it supports full dynamic shapes, meaning that it’s not padding shorter sequence lengths or having to recompile when sequence lengths change. Therefore, we benchmark on a variety of sequence lengths to show the relative speedups you should expect depending on the distribution of sequence lengths in your data.
Future work
Your launch keynote also mentioned a “cloud serving platform” – what’s the deal with that?
We are excited to bring state of the art innovations across many layers of the AI lifecycle, including layers that are necessary to serve increasingly large AI models on cloud infrastructure. To that end—as mentioned in our launch keynote—we are on a journey to build a next-generation cloud compute platform that will significantly improve server utilization by distributing inference across many nodes. It will effectively scale out and down to meet dynamic changes in traffic volume, and overall significantly reduce the amount of time and effort required to bring up and maintain large models in production. Stay tuned for more information about our plans here.
Do you also intend to support training?
Right now we are focused on inference because it’s a more fragmented landscape than training and because it is where organizations have a majority of their AI operating expenses. That being said, there’s no reason why the technology we’ve built for MAX Engine can’t scale to support training workloads with similar performance improvements. Stay tuned!
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!