run

Given a path to pre-trained (.pt) MODEL, starts a sessions for real-time inference, via OSC. The following OSC routes are available:

Input routes

• /input <list>: Triggers a new prediction, given an list of event features. The number of event features must match either the masked or unmasked event size.

• /sample: Generates a random sample from the learned distribution. Useful for kicking off an auto-regressive loop.

• /reset: Resets the model's hidden state (i.e., it's context), as well as cancelling any scheduled predictions, if the model is temporal.

Output routes

• /output: Predicted event (unmasked). The predicted event can be sent as is back to /input for auto-regression, even if the model is trained on masked features.

tip

Note that, in some cases, running the model on CPU results in lower prediction latency.

Usage

hxmx run <MODEL> [--in-port <INTEGER>] [--out-port <INTEGER>] [--address <TEXT>] [-d <CHOICE>] [--debug] [--help]

Arguments

Name	Type	Required	Default
MODEL	path — must exist, file	✓	—

Options

Name	Type	Default	Description
--in-port	integer	8000	Input OSC port.
--out-port	integer	9000	Output OSC port.
--address	string	127.0.0.1	OSC IP address
--device, -d	choice (auto|cpu| ...)	cpu	Computing device. To list all available devices, run hxmx devices.
--debug	boolean	False	Debug mode. Includes traceback when an error is raised.
-h, --help	boolean	False	Open documentation in browser.