Loading TOC...
Matches for cat:function have been highlighted. remove

MarkLogic Server 11.0 Product Documentation
ort.session

ort.session(
   model as Node,
   [options as Object]
) as Object

Summary

Load an ONNX model from the database, as an inference session. The user can then perform "run"s of this session, with different input values/settings.

Parameters
model The binary node representing the ONNX model.
options Options used to create the session. Possible values are:
Key Name Value Type Possible Values Details
graphOptimizationLevel integer 0,1,2,99 0 means disable all optimizations; 1 means enable basic optimizations; 2 means enable an extended set of optimizations; 99 means enable all optimizations.
logId string Any string. The log-id will be appended to the front of logs produced by any run of this session.
logVerbosityLevel string VERBOSE, INFO, WARNING, ERROR, FATAL VERBOSE level logs of ort session will be logged when MarkLogic's log level is Fine or higher. WARNING level logs of ort session will be logged when MarkLogic's log level is Debug or higher. INFO level logs of ort session will be logged when MarkLogic's log level is Info or higher. ERROR level logs of ort session will be logged when MarkLogic's log level is Error or higher. FATAL level logs of ort session will be logged when MarkLogic's log level is Emergency or higher.
device xs:string cpu,gpu# "cpu" means the session will be loaded on CPU memory, while "gpu#" means the session will be loaded on GPU memory, where # means the device ID of the GPU, which typically starts from 0. Defaults to "gpu0" if the host has an NVIDIA GPU with driver installed; to "cpu" otherwise.

Example

  (:after loading an ONNX model into the database as "testmodel.onnx" :)
  ort.session(cts.doc("testmodel.onnx"), {"logId":"testSessionLogId"})
  => an ort:session, possibly described as follows:
  OrtSession(Inputs:["input_0"]=>Outputs:["output_0"])

Stack Overflow iconStack Overflow: Get the most useful answers to questions from the MarkLogic community, or ask your own question.