scorecard-ai 1.0.0 → 1.1.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +21 -0
- package/README.md +48 -19
- package/client.d.mts +2 -2
- package/client.d.mts.map +1 -1
- package/client.d.ts +2 -2
- package/client.d.ts.map +1 -1
- package/client.js +3 -1
- package/client.js.map +1 -1
- package/client.mjs +3 -1
- package/client.mjs.map +1 -1
- package/package.json +1 -1
- package/resources/index.d.mts +1 -1
- package/resources/index.d.mts.map +1 -1
- package/resources/index.d.ts +1 -1
- package/resources/index.d.ts.map +1 -1
- package/resources/index.js.map +1 -1
- package/resources/index.mjs.map +1 -1
- package/resources/metrics.d.mts +192 -2
- package/resources/metrics.d.mts.map +1 -1
- package/resources/metrics.d.ts +192 -2
- package/resources/metrics.d.ts.map +1 -1
- package/resources/metrics.js +20 -1
- package/resources/metrics.js.map +1 -1
- package/resources/metrics.mjs +20 -1
- package/resources/metrics.mjs.map +1 -1
- package/src/client.ts +10 -3
- package/src/resources/index.ts +1 -1
- package/src/resources/metrics.ts +245 -2
- package/src/version.ts +1 -1
- package/version.d.mts +1 -1
- package/version.d.ts +1 -1
- package/version.js +1 -1
- package/version.mjs +1 -1
package/src/resources/metrics.ts
CHANGED
|
@@ -7,7 +7,8 @@ import { path } from '../internal/utils/path';
|
|
|
7
7
|
|
|
8
8
|
export class Metrics extends APIResource {
|
|
9
9
|
/**
|
|
10
|
-
* Create a new Metric for evaluating system outputs.
|
|
10
|
+
* Create a new Metric for evaluating system outputs. The structure of a metric
|
|
11
|
+
* depends on the evalType and outputType of the metric.
|
|
11
12
|
*
|
|
12
13
|
* @example
|
|
13
14
|
* ```ts
|
|
@@ -29,6 +30,25 @@ export class Metrics extends APIResource {
|
|
|
29
30
|
create(projectID: string, body: MetricCreateParams, options?: RequestOptions): APIPromise<Metric> {
|
|
30
31
|
return this._client.post(path`/projects/${projectID}/metrics`, { body, ...options });
|
|
31
32
|
}
|
|
33
|
+
|
|
34
|
+
/**
|
|
35
|
+
* Update an existing Metric. You must specify the evalType and outputType of the
|
|
36
|
+
* metric. The structure of a metric depends on the evalType and outputType of the
|
|
37
|
+
* metric.
|
|
38
|
+
*
|
|
39
|
+
* @example
|
|
40
|
+
* ```ts
|
|
41
|
+
* const metric = await client.metrics.update('321', {
|
|
42
|
+
* evalType: 'ai',
|
|
43
|
+
* outputType: 'boolean',
|
|
44
|
+
* promptTemplate:
|
|
45
|
+
* 'Using the following guidelines, evaluate the response: {{ guidelines }}\n\nResponse: {{ outputs.response }}\n\nIdeal answer: {{ expected.idealResponse }}',
|
|
46
|
+
* });
|
|
47
|
+
* ```
|
|
48
|
+
*/
|
|
49
|
+
update(metricID: string, body: MetricUpdateParams, options?: RequestOptions): APIPromise<Metric> {
|
|
50
|
+
return this._client.patch(path`/metrics/${metricID}`, { body, ...options });
|
|
51
|
+
}
|
|
32
52
|
}
|
|
33
53
|
|
|
34
54
|
/**
|
|
@@ -520,6 +540,229 @@ export declare namespace MetricCreateParams {
|
|
|
520
540
|
}
|
|
521
541
|
}
|
|
522
542
|
|
|
543
|
+
export type MetricUpdateParams =
|
|
544
|
+
| MetricUpdateParams.AIIntMetric
|
|
545
|
+
| MetricUpdateParams.HumanIntMetric
|
|
546
|
+
| MetricUpdateParams.HeuristicIntMetric
|
|
547
|
+
| MetricUpdateParams.AIBooleanMetric
|
|
548
|
+
| MetricUpdateParams.HumanBooleanMetric
|
|
549
|
+
| MetricUpdateParams.HeuristicBooleanMetric;
|
|
550
|
+
|
|
551
|
+
export declare namespace MetricUpdateParams {
|
|
552
|
+
export interface AIIntMetric {
|
|
553
|
+
/**
|
|
554
|
+
* AI-based evaluation type.
|
|
555
|
+
*/
|
|
556
|
+
evalType: 'ai';
|
|
557
|
+
|
|
558
|
+
/**
|
|
559
|
+
* Integer output type.
|
|
560
|
+
*/
|
|
561
|
+
outputType: 'int';
|
|
562
|
+
|
|
563
|
+
/**
|
|
564
|
+
* The description of the Metric.
|
|
565
|
+
*/
|
|
566
|
+
description?: string | null;
|
|
567
|
+
|
|
568
|
+
/**
|
|
569
|
+
* The AI model to use for evaluation.
|
|
570
|
+
*/
|
|
571
|
+
evalModelName?: string;
|
|
572
|
+
|
|
573
|
+
/**
|
|
574
|
+
* Guidelines for AI evaluation on how to score the metric.
|
|
575
|
+
*/
|
|
576
|
+
guidelines?: string | null;
|
|
577
|
+
|
|
578
|
+
/**
|
|
579
|
+
* The name of the Metric.
|
|
580
|
+
*/
|
|
581
|
+
name?: string;
|
|
582
|
+
|
|
583
|
+
/**
|
|
584
|
+
* The threshold for determining pass/fail from integer scores (1-5).
|
|
585
|
+
*/
|
|
586
|
+
passingThreshold?: number;
|
|
587
|
+
|
|
588
|
+
/**
|
|
589
|
+
* The complete prompt template for AI evaluation. Should include placeholders for
|
|
590
|
+
* dynamic content.
|
|
591
|
+
*/
|
|
592
|
+
promptTemplate?: string;
|
|
593
|
+
|
|
594
|
+
/**
|
|
595
|
+
* The temperature for AI evaluation (0-2).
|
|
596
|
+
*/
|
|
597
|
+
temperature?: number;
|
|
598
|
+
}
|
|
599
|
+
|
|
600
|
+
export interface HumanIntMetric {
|
|
601
|
+
/**
|
|
602
|
+
* Human-based evaluation type.
|
|
603
|
+
*/
|
|
604
|
+
evalType: 'human';
|
|
605
|
+
|
|
606
|
+
/**
|
|
607
|
+
* Integer output type.
|
|
608
|
+
*/
|
|
609
|
+
outputType: 'int';
|
|
610
|
+
|
|
611
|
+
/**
|
|
612
|
+
* The description of the Metric.
|
|
613
|
+
*/
|
|
614
|
+
description?: string | null;
|
|
615
|
+
|
|
616
|
+
/**
|
|
617
|
+
* Guidelines for human evaluators.
|
|
618
|
+
*/
|
|
619
|
+
guidelines?: string;
|
|
620
|
+
|
|
621
|
+
/**
|
|
622
|
+
* The name of the Metric.
|
|
623
|
+
*/
|
|
624
|
+
name?: string;
|
|
625
|
+
|
|
626
|
+
/**
|
|
627
|
+
* The threshold for determining pass/fail from integer scores (1-5).
|
|
628
|
+
*/
|
|
629
|
+
passingThreshold?: number;
|
|
630
|
+
}
|
|
631
|
+
|
|
632
|
+
export interface HeuristicIntMetric {
|
|
633
|
+
/**
|
|
634
|
+
* Heuristic-based evaluation type.
|
|
635
|
+
*/
|
|
636
|
+
evalType: 'heuristic';
|
|
637
|
+
|
|
638
|
+
/**
|
|
639
|
+
* Integer output type.
|
|
640
|
+
*/
|
|
641
|
+
outputType: 'int';
|
|
642
|
+
|
|
643
|
+
/**
|
|
644
|
+
* The description of the Metric.
|
|
645
|
+
*/
|
|
646
|
+
description?: string | null;
|
|
647
|
+
|
|
648
|
+
/**
|
|
649
|
+
* Optional guidelines for heuristic evaluation logic.
|
|
650
|
+
*/
|
|
651
|
+
guidelines?: string;
|
|
652
|
+
|
|
653
|
+
/**
|
|
654
|
+
* The name of the Metric.
|
|
655
|
+
*/
|
|
656
|
+
name?: string;
|
|
657
|
+
|
|
658
|
+
/**
|
|
659
|
+
* The threshold for determining pass/fail from integer scores (1-5).
|
|
660
|
+
*/
|
|
661
|
+
passingThreshold?: number;
|
|
662
|
+
}
|
|
663
|
+
|
|
664
|
+
export interface AIBooleanMetric {
|
|
665
|
+
/**
|
|
666
|
+
* AI-based evaluation type.
|
|
667
|
+
*/
|
|
668
|
+
evalType: 'ai';
|
|
669
|
+
|
|
670
|
+
/**
|
|
671
|
+
* Boolean output type.
|
|
672
|
+
*/
|
|
673
|
+
outputType: 'boolean';
|
|
674
|
+
|
|
675
|
+
/**
|
|
676
|
+
* The description of the Metric.
|
|
677
|
+
*/
|
|
678
|
+
description?: string | null;
|
|
679
|
+
|
|
680
|
+
/**
|
|
681
|
+
* The AI model to use for evaluation.
|
|
682
|
+
*/
|
|
683
|
+
evalModelName?: string;
|
|
684
|
+
|
|
685
|
+
/**
|
|
686
|
+
* Guidelines for AI evaluation on how to score the metric.
|
|
687
|
+
*/
|
|
688
|
+
guidelines?: string | null;
|
|
689
|
+
|
|
690
|
+
/**
|
|
691
|
+
* The name of the Metric.
|
|
692
|
+
*/
|
|
693
|
+
name?: string;
|
|
694
|
+
|
|
695
|
+
/**
|
|
696
|
+
* The complete prompt template for AI evaluation. Should include placeholders for
|
|
697
|
+
* dynamic content.
|
|
698
|
+
*/
|
|
699
|
+
promptTemplate?: string;
|
|
700
|
+
|
|
701
|
+
/**
|
|
702
|
+
* The temperature for AI evaluation (0-2).
|
|
703
|
+
*/
|
|
704
|
+
temperature?: number;
|
|
705
|
+
}
|
|
706
|
+
|
|
707
|
+
export interface HumanBooleanMetric {
|
|
708
|
+
/**
|
|
709
|
+
* Human-based evaluation type.
|
|
710
|
+
*/
|
|
711
|
+
evalType: 'human';
|
|
712
|
+
|
|
713
|
+
/**
|
|
714
|
+
* Boolean output type.
|
|
715
|
+
*/
|
|
716
|
+
outputType: 'boolean';
|
|
717
|
+
|
|
718
|
+
/**
|
|
719
|
+
* The description of the Metric.
|
|
720
|
+
*/
|
|
721
|
+
description?: string | null;
|
|
722
|
+
|
|
723
|
+
/**
|
|
724
|
+
* Guidelines for human evaluators.
|
|
725
|
+
*/
|
|
726
|
+
guidelines?: string;
|
|
727
|
+
|
|
728
|
+
/**
|
|
729
|
+
* The name of the Metric.
|
|
730
|
+
*/
|
|
731
|
+
name?: string;
|
|
732
|
+
}
|
|
733
|
+
|
|
734
|
+
export interface HeuristicBooleanMetric {
|
|
735
|
+
/**
|
|
736
|
+
* Heuristic-based evaluation type.
|
|
737
|
+
*/
|
|
738
|
+
evalType: 'heuristic';
|
|
739
|
+
|
|
740
|
+
/**
|
|
741
|
+
* Boolean output type.
|
|
742
|
+
*/
|
|
743
|
+
outputType: 'boolean';
|
|
744
|
+
|
|
745
|
+
/**
|
|
746
|
+
* The description of the Metric.
|
|
747
|
+
*/
|
|
748
|
+
description?: string | null;
|
|
749
|
+
|
|
750
|
+
/**
|
|
751
|
+
* Optional guidelines for heuristic evaluation logic.
|
|
752
|
+
*/
|
|
753
|
+
guidelines?: string;
|
|
754
|
+
|
|
755
|
+
/**
|
|
756
|
+
* The name of the Metric.
|
|
757
|
+
*/
|
|
758
|
+
name?: string;
|
|
759
|
+
}
|
|
760
|
+
}
|
|
761
|
+
|
|
523
762
|
export declare namespace Metrics {
|
|
524
|
-
export {
|
|
763
|
+
export {
|
|
764
|
+
type Metric as Metric,
|
|
765
|
+
type MetricCreateParams as MetricCreateParams,
|
|
766
|
+
type MetricUpdateParams as MetricUpdateParams,
|
|
767
|
+
};
|
|
525
768
|
}
|
package/src/version.ts
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
export const VERSION = '1.
|
|
1
|
+
export const VERSION = '1.1.0'; // x-release-please-version
|
package/version.d.mts
CHANGED
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
export declare const VERSION = "1.
|
|
1
|
+
export declare const VERSION = "1.1.0";
|
|
2
2
|
//# sourceMappingURL=version.d.mts.map
|
package/version.d.ts
CHANGED
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
export declare const VERSION = "1.
|
|
1
|
+
export declare const VERSION = "1.1.0";
|
|
2
2
|
//# sourceMappingURL=version.d.ts.map
|
package/version.js
CHANGED
package/version.mjs
CHANGED
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
export const VERSION = '1.
|
|
1
|
+
export const VERSION = '1.1.0'; // x-release-please-version
|
|
2
2
|
//# sourceMappingURL=version.mjs.map
|