FAQ#

Install#

How do I install the SDK on Windows?#

If you are on Windows, you can use WSL to install Ubuntu Linux as an app on your computer. Once installed, use it to follow the instructions in the Install section.

Note that if you haven’t used WSL before, you will need to install some basic packages first:

$ sudo apt update
$ sudo apt install python3.9-venv python3-pip

To use graphical tools (e.g., mx_viewer), you also need

$ sudo apt install qt5dxcb-plugin python3-tk

How do I install the SDK on Mac?#

If you are on macOS, it’s a little more complicated. You’ll need to either use a lima-vm or (if on an Intel Mac) set up a virtual machine in VirtualBox or VMWare or Parallels. Then login to Linux and follow the Install steps.

DFP#

Why are DFP files so big?#

A single .dfp file is actually made up of 2 separate data structures:

  1. Simulator Config

  2. MXA Config

The MXA Config section is what is actually used to program the chips. It contains the compiled model weights and MXA configs (like “instructions” for cores) and is modestly sized.

The Simulator Config, on the other is multiple times larger than the MXA Config, because of the nature of the hardware-accurate Simulator. It can be over 4x larger than the MXA Config.

To see the breakdown of Simulator vs. MXA Config, use the dfp_inspect tool. It will show a breakdown of Simulator vs. MXA sizes in megabytes.

Simulator Config is disabled by default when using the command line. In the NeuralCompiler API, it is controlled with the no_sim_dfp argument. True will disable Simulator Config, decreasing the DFP file size.

Logging and Telemetry#

How do I know the information is being sent anonymously?#

The only customer specific data collected is a randomized source ID which can be changed at will by the customer

How do I change my ID or Y/N choice regarding sharing data for optimizing performance & enabling debug support?#

The ID and permission to enable telemetry are stored in a JSON file whose location depends on the OS. For Linux, the file path is ~/.config/memx_config.json. This file can be opened in a text editor and the ID can be changed or telemetry can be disabled by setting the value to false

{
   "anonymous_id": "998150d0fe6611eebcf6bd5b872ec7f5",
   "permit_telemetry": true
}

Is any of this information used for advertising or sent to any 3rd party company?#

No. As a reminder, the customer remains anonymous, and no information collected is used outside of MemryX

Why are you collecting data regarding model compilation?#

Rather than send MemryX a screenshot of compilation messages for debug, additional information is collected which can help us debug/optimize AI model(s)

Do you collect the model structure/weights so MemryX can rebuild any model compiled?#

No; All customer models all remain private to each customer. No weights/structure are collected. We do collect a list of operators used and basic model information so we can be sure our compiler can support each operation with optimal performance, but it is not possible for anyone to rebuild a model with information collected.

Are you collecting any system information?#

We collect basic system information to make sure the host system is compatible with the MemryX AI accelerator; this also helps us troubleshoot problems which may occur due to the system configuration, rather than AI Accelerator configuration

On this page