Evaluator-AgentLoop(AgentLoop)-阿里云帮助中心

Parameter	Type	Description
	object	The evaluator configuration in an evaluation task. Used to reference a previously created evaluator or to describe an inline evaluator for the task.
evaluatorRef	string	The reference name of a registered evaluator. When specified, the evaluator definition is loaded by this reference with higher priority. Both built-in evaluators and custom evaluators are supported.
name	string	The evaluator name. Required for inline evaluators when evaluatorRef is not specified. The evaluatorRef or name must be unique within the same task.
type	string	The evaluator type. Defaults to LLM if not specified. Inline CODE evaluators are currently not supported. For CODE type evaluators, reference a previously created evaluator by using evaluatorRef.
resultName	string	The field name for the evaluation result. Required for inline evaluators. When referencing an existing evaluator, the metricName defined in the evaluator definition is used if this parameter is not specified.
resultType	string	The evaluation result type. Required for inline evaluators. Defaults to score when referencing an existing evaluator and this parameter is not specified.
config	object	The evaluator runtime configuration. For inline LLM evaluators, this must include configurations such as prompt. When referencing an existing evaluator, this parameter is typically not required and is only specified when runtime parameters such as version need to be set.
filters	object	The evaluator-level data filter conditions. These take effect together with the task-level dataFilter.query.
variableMapping	object	The variable mapping that maps evaluator variables to evaluation data fields. Required for LLM/AGENT inline evaluators. When referencing an existing evaluator, the variable names must exist in the evaluator definition.
	string	The value of a single variable mapping, indicating the evaluation data field path from which the evaluator variable reads.

object

The evaluator configuration in an evaluation task. Used to reference a previously created evaluator or to describe an inline evaluator for the task.

evaluatorRef

string

The reference name of a registered evaluator. When specified, the evaluator definition is loaded by this reference with higher priority. Both built-in evaluators and custom evaluators are supported.

name

string

The evaluator name. Required for inline evaluators when evaluatorRef is not specified. The evaluatorRef or name must be unique within the same task.

type

string

The evaluator type. Defaults to LLM if not specified. Inline CODE evaluators are currently not supported. For CODE type evaluators, reference a previously created evaluator by using evaluatorRef.

resultName

string

The field name for the evaluation result. Required for inline evaluators. When referencing an existing evaluator, the metricName defined in the evaluator definition is used if this parameter is not specified.

resultType

string

The evaluation result type. Required for inline evaluators. Defaults to score when referencing an existing evaluator and this parameter is not specified.

config

object

The evaluator runtime configuration. For inline LLM evaluators, this must include configurations such as prompt. When referencing an existing evaluator, this parameter is typically not required and is only specified when runtime parameters such as version need to be set.

filters

object

The evaluator-level data filter conditions. These take effect together with the task-level dataFilter.query.

variableMapping

object

The variable mapping that maps evaluator variables to evaluation data fields. Required for LLM/AGENT inline evaluators. When referencing an existing evaluator, the variable names must exist in the evaluator definition.

string

The value of a single variable mapping, indicating the evaluation data field path from which the evaluator variable reads.