Class
MLSession
NotCreatableNotReplicated
A machine-learning inference session that wraps a single loaded model and runs
it through MLSession:ForwardAsync().
An MLSession wraps a single loaded neural-network model and runs inference
on it. You don't construct a session directly; instead,
MLService:CreateSessionAsync() loads a model bundle for a given asset
and returns a ready-to-use session. Call
ForwardAsync() on that session to run the
model's forward pass on your input data and receive its output.
Methods 1#
| ForwardAsync | Runs the session's model on the given input and asynchronously returns the model's output.Yields |
ForwardAsync(data: Dictionary): Dictionary#
Yields
Runs inference for the session's loaded model.
If the model declares a dynamic output dimension, the call errors. Calling this method on a session with no valid loaded model also errors.
| Name | Type | Default | Description |
|---|---|---|---|
data | Dictionary | A dictionary mapping each of the model's input names to that input's tensor, given as a nested array of numbers matching the input's shape. |
Returns
Dictionary— A dictionary mapping each of the model's output names to that output's tensor, given as a nested array of numbers matching the output's shape.