@mastra/evals 0.14.3 → 1.0.0-beta.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 +32 -14
- package/README.md +19 -159
- package/dist/{chunk-KHEXN75Q.js → chunk-CCLM7KPF.js} +45 -21
- package/dist/chunk-CCLM7KPF.js.map +1 -0
- package/dist/{chunk-QKR2PMLZ.cjs → chunk-TPQLLHZW.cjs} +46 -21
- package/dist/chunk-TPQLLHZW.cjs.map +1 -0
- package/dist/scorers/code/completeness/index.d.ts +1 -1
- package/dist/scorers/code/completeness/index.d.ts.map +1 -1
- package/dist/scorers/code/content-similarity/index.d.ts +1 -1
- package/dist/scorers/code/content-similarity/index.d.ts.map +1 -1
- package/dist/scorers/code/keyword-coverage/index.d.ts +1 -1
- package/dist/scorers/code/keyword-coverage/index.d.ts.map +1 -1
- package/dist/scorers/code/textual-difference/index.d.ts +1 -1
- package/dist/scorers/code/textual-difference/index.d.ts.map +1 -1
- package/dist/scorers/code/tone/index.d.ts +1 -1
- package/dist/scorers/code/tone/index.d.ts.map +1 -1
- package/dist/scorers/code/tool-call-accuracy/index.d.ts +1 -1
- package/dist/scorers/code/tool-call-accuracy/index.d.ts.map +1 -1
- package/dist/scorers/llm/answer-relevancy/index.d.ts +1 -1
- package/dist/scorers/llm/answer-relevancy/index.d.ts.map +1 -1
- package/dist/scorers/llm/answer-similarity/index.d.ts +2 -2
- package/dist/scorers/llm/answer-similarity/index.d.ts.map +1 -1
- package/dist/scorers/llm/bias/index.d.ts +2 -2
- package/dist/scorers/llm/bias/index.d.ts.map +1 -1
- package/dist/scorers/llm/context-precision/index.d.ts +3 -3
- package/dist/scorers/llm/context-precision/index.d.ts.map +1 -1
- package/dist/scorers/llm/context-relevance/index.d.ts +3 -3
- package/dist/scorers/llm/context-relevance/index.d.ts.map +1 -1
- package/dist/scorers/llm/faithfulness/index.d.ts +2 -2
- package/dist/scorers/llm/faithfulness/index.d.ts.map +1 -1
- package/dist/scorers/llm/hallucination/index.d.ts +2 -2
- package/dist/scorers/llm/hallucination/index.d.ts.map +1 -1
- package/dist/scorers/llm/noise-sensitivity/index.d.ts +1 -1
- package/dist/scorers/llm/noise-sensitivity/index.d.ts.map +1 -1
- package/dist/scorers/llm/prompt-alignment/index.d.ts +2 -2
- package/dist/scorers/llm/prompt-alignment/index.d.ts.map +1 -1
- package/dist/scorers/llm/tool-call-accuracy/index.d.ts +2 -2
- package/dist/scorers/llm/tool-call-accuracy/index.d.ts.map +1 -1
- package/dist/scorers/llm/toxicity/index.d.ts +2 -2
- package/dist/scorers/llm/toxicity/index.d.ts.map +1 -1
- package/dist/scorers/{llm → prebuilt}/index.cjs +479 -62
- package/dist/scorers/prebuilt/index.cjs.map +1 -0
- package/dist/scorers/prebuilt/index.d.ts +3 -0
- package/dist/scorers/prebuilt/index.d.ts.map +1 -0
- package/dist/scorers/{llm → prebuilt}/index.js +419 -15
- package/dist/scorers/prebuilt/index.js.map +1 -0
- package/dist/scorers/utils.cjs +21 -17
- package/dist/scorers/utils.d.ts +21 -11
- package/dist/scorers/utils.d.ts.map +1 -1
- package/dist/scorers/utils.js +1 -1
- package/package.json +12 -58
- package/dist/attachListeners.d.ts +0 -4
- package/dist/attachListeners.d.ts.map +0 -1
- package/dist/chunk-7QAUEU4L.cjs +0 -10
- package/dist/chunk-7QAUEU4L.cjs.map +0 -1
- package/dist/chunk-EMMSS5I5.cjs +0 -37
- package/dist/chunk-EMMSS5I5.cjs.map +0 -1
- package/dist/chunk-G3PMV62Z.js +0 -33
- package/dist/chunk-G3PMV62Z.js.map +0 -1
- package/dist/chunk-IUSAD2BW.cjs +0 -19
- package/dist/chunk-IUSAD2BW.cjs.map +0 -1
- package/dist/chunk-KHEXN75Q.js.map +0 -1
- package/dist/chunk-QKR2PMLZ.cjs.map +0 -1
- package/dist/chunk-QTWX6TKR.js +0 -8
- package/dist/chunk-QTWX6TKR.js.map +0 -1
- package/dist/chunk-YGTIO3J5.js +0 -17
- package/dist/chunk-YGTIO3J5.js.map +0 -1
- package/dist/dist-LDTK3TIP.cjs +0 -16759
- package/dist/dist-LDTK3TIP.cjs.map +0 -1
- package/dist/dist-OWYZEOJK.js +0 -16737
- package/dist/dist-OWYZEOJK.js.map +0 -1
- package/dist/evaluation.d.ts +0 -8
- package/dist/evaluation.d.ts.map +0 -1
- package/dist/index.cjs +0 -93
- package/dist/index.cjs.map +0 -1
- package/dist/index.d.ts +0 -3
- package/dist/index.d.ts.map +0 -1
- package/dist/index.js +0 -89
- package/dist/index.js.map +0 -1
- package/dist/magic-string.es-7ORA5OGR.js +0 -1305
- package/dist/magic-string.es-7ORA5OGR.js.map +0 -1
- package/dist/magic-string.es-NZ2XWFKN.cjs +0 -1311
- package/dist/magic-string.es-NZ2XWFKN.cjs.map +0 -1
- package/dist/metrics/index.d.ts +0 -4
- package/dist/metrics/index.d.ts.map +0 -1
- package/dist/metrics/judge/index.cjs +0 -12
- package/dist/metrics/judge/index.cjs.map +0 -1
- package/dist/metrics/judge/index.d.ts +0 -7
- package/dist/metrics/judge/index.d.ts.map +0 -1
- package/dist/metrics/judge/index.js +0 -3
- package/dist/metrics/judge/index.js.map +0 -1
- package/dist/metrics/llm/answer-relevancy/index.d.ts +0 -16
- package/dist/metrics/llm/answer-relevancy/index.d.ts.map +0 -1
- package/dist/metrics/llm/answer-relevancy/metricJudge.d.ts +0 -20
- package/dist/metrics/llm/answer-relevancy/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/answer-relevancy/prompts.d.ts +0 -19
- package/dist/metrics/llm/answer-relevancy/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/bias/index.d.ts +0 -14
- package/dist/metrics/llm/bias/index.d.ts.map +0 -1
- package/dist/metrics/llm/bias/metricJudge.d.ts +0 -14
- package/dist/metrics/llm/bias/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/bias/prompts.d.ts +0 -14
- package/dist/metrics/llm/bias/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/context-position/index.d.ts +0 -16
- package/dist/metrics/llm/context-position/index.d.ts.map +0 -1
- package/dist/metrics/llm/context-position/metricJudge.d.ts +0 -20
- package/dist/metrics/llm/context-position/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/context-position/prompts.d.ts +0 -17
- package/dist/metrics/llm/context-position/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/context-precision/index.d.ts +0 -16
- package/dist/metrics/llm/context-precision/index.d.ts.map +0 -1
- package/dist/metrics/llm/context-precision/metricJudge.d.ts +0 -20
- package/dist/metrics/llm/context-precision/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/context-precision/prompts.d.ts +0 -17
- package/dist/metrics/llm/context-precision/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/context-relevancy/index.d.ts +0 -16
- package/dist/metrics/llm/context-relevancy/index.d.ts.map +0 -1
- package/dist/metrics/llm/context-relevancy/metricJudge.d.ts +0 -16
- package/dist/metrics/llm/context-relevancy/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/context-relevancy/prompts.d.ts +0 -13
- package/dist/metrics/llm/context-relevancy/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/contextual-recall/index.d.ts +0 -16
- package/dist/metrics/llm/contextual-recall/index.d.ts.map +0 -1
- package/dist/metrics/llm/contextual-recall/metricJudge.d.ts +0 -16
- package/dist/metrics/llm/contextual-recall/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/contextual-recall/prompts.d.ts +0 -13
- package/dist/metrics/llm/contextual-recall/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/faithfulness/index.d.ts +0 -16
- package/dist/metrics/llm/faithfulness/index.d.ts.map +0 -1
- package/dist/metrics/llm/faithfulness/metricJudge.d.ts +0 -22
- package/dist/metrics/llm/faithfulness/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/faithfulness/prompts.d.ts +0 -20
- package/dist/metrics/llm/faithfulness/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/hallucination/index.d.ts +0 -16
- package/dist/metrics/llm/hallucination/index.d.ts.map +0 -1
- package/dist/metrics/llm/hallucination/metricJudge.d.ts +0 -22
- package/dist/metrics/llm/hallucination/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/hallucination/prompts.d.ts +0 -17
- package/dist/metrics/llm/hallucination/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/index.cjs +0 -2481
- package/dist/metrics/llm/index.cjs.map +0 -1
- package/dist/metrics/llm/index.d.ts +0 -12
- package/dist/metrics/llm/index.d.ts.map +0 -1
- package/dist/metrics/llm/index.js +0 -2469
- package/dist/metrics/llm/index.js.map +0 -1
- package/dist/metrics/llm/prompt-alignment/index.d.ts +0 -33
- package/dist/metrics/llm/prompt-alignment/index.d.ts.map +0 -1
- package/dist/metrics/llm/prompt-alignment/metricJudge.d.ts +0 -20
- package/dist/metrics/llm/prompt-alignment/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/prompt-alignment/prompts.d.ts +0 -17
- package/dist/metrics/llm/prompt-alignment/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/summarization/index.d.ts +0 -19
- package/dist/metrics/llm/summarization/index.d.ts.map +0 -1
- package/dist/metrics/llm/summarization/metricJudge.d.ts +0 -34
- package/dist/metrics/llm/summarization/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/summarization/prompts.d.ts +0 -30
- package/dist/metrics/llm/summarization/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/toxicity/index.d.ts +0 -14
- package/dist/metrics/llm/toxicity/index.d.ts.map +0 -1
- package/dist/metrics/llm/toxicity/metricJudge.d.ts +0 -14
- package/dist/metrics/llm/toxicity/metricJudge.d.ts.map +0 -1
- package/dist/metrics/llm/toxicity/prompts.d.ts +0 -10
- package/dist/metrics/llm/toxicity/prompts.d.ts.map +0 -1
- package/dist/metrics/llm/types.d.ts +0 -7
- package/dist/metrics/llm/types.d.ts.map +0 -1
- package/dist/metrics/llm/utils.d.ts +0 -14
- package/dist/metrics/llm/utils.d.ts.map +0 -1
- package/dist/metrics/nlp/completeness/index.d.ts +0 -21
- package/dist/metrics/nlp/completeness/index.d.ts.map +0 -1
- package/dist/metrics/nlp/content-similarity/index.d.ts +0 -18
- package/dist/metrics/nlp/content-similarity/index.d.ts.map +0 -1
- package/dist/metrics/nlp/index.cjs +0 -203
- package/dist/metrics/nlp/index.cjs.map +0 -1
- package/dist/metrics/nlp/index.d.ts +0 -6
- package/dist/metrics/nlp/index.d.ts.map +0 -1
- package/dist/metrics/nlp/index.js +0 -190
- package/dist/metrics/nlp/index.js.map +0 -1
- package/dist/metrics/nlp/keyword-coverage/index.d.ts +0 -13
- package/dist/metrics/nlp/keyword-coverage/index.d.ts.map +0 -1
- package/dist/metrics/nlp/textual-difference/index.d.ts +0 -15
- package/dist/metrics/nlp/textual-difference/index.d.ts.map +0 -1
- package/dist/metrics/nlp/tone/index.d.ts +0 -18
- package/dist/metrics/nlp/tone/index.d.ts.map +0 -1
- package/dist/scorers/code/index.cjs +0 -329
- package/dist/scorers/code/index.cjs.map +0 -1
- package/dist/scorers/code/index.js +0 -315
- package/dist/scorers/code/index.js.map +0 -1
- package/dist/scorers/llm/index.cjs.map +0 -1
- package/dist/scorers/llm/index.js.map +0 -1
package/dist/scorers/utils.d.ts
CHANGED
|
@@ -1,6 +1,12 @@
|
|
|
1
|
-
import {
|
|
2
|
-
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent, ScoringInput } from '@mastra/core/
|
|
3
|
-
import
|
|
1
|
+
import type { MastraDBMessage } from '@mastra/core/agent';
|
|
2
|
+
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent, ScoringInput } from '@mastra/core/evals';
|
|
3
|
+
import { RequestContext } from '@mastra/core/request-context';
|
|
4
|
+
import type { ToolInvocation } from 'ai';
|
|
5
|
+
/**
|
|
6
|
+
* Extract text content from MastraDBMessage
|
|
7
|
+
* Matches the logic used in MessageList.mastraDBMessageToAIV4UIMessage
|
|
8
|
+
*/
|
|
9
|
+
export declare function getTextContentFromMastraDBMessage(message: MastraDBMessage): string;
|
|
4
10
|
export declare const roundToTwoDecimals: (num: number) => number;
|
|
5
11
|
export declare function isCloserTo(value: number, target1: number, target2: number): boolean;
|
|
6
12
|
export type TestCase = {
|
|
@@ -14,7 +20,7 @@ export type TestCase = {
|
|
|
14
20
|
export type TestCaseWithContext = TestCase & {
|
|
15
21
|
context: string[];
|
|
16
22
|
};
|
|
17
|
-
export declare const createTestRun: (input: string, output: string, additionalContext?: Record<string, any>,
|
|
23
|
+
export declare const createTestRun: (input: string, output: string, additionalContext?: Record<string, any>, requestContext?: Record<string, any>) => ScoringInput;
|
|
18
24
|
export declare const getUserMessageFromRunInput: (input?: ScorerRunInputForAgent) => string | undefined;
|
|
19
25
|
export declare const getSystemMessagesFromRunInput: (input?: ScorerRunInputForAgent) => string[];
|
|
20
26
|
export declare const getCombinedSystemPrompt: (input?: ScorerRunInputForAgent) => string;
|
|
@@ -32,10 +38,14 @@ export declare const createToolInvocation: ({ toolCallId, toolName, args, result
|
|
|
32
38
|
result: Record<string, any>;
|
|
33
39
|
state: string;
|
|
34
40
|
};
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
|
|
41
|
+
/**
|
|
42
|
+
* Helper function to create MastraDBMessage objects for tests
|
|
43
|
+
* Supports optional tool invocations for testing tool call scenarios
|
|
44
|
+
*/
|
|
45
|
+
export declare function createTestMessage({ content, role, id, toolInvocations, }: {
|
|
38
46
|
content: string;
|
|
47
|
+
role: 'user' | 'assistant' | 'system';
|
|
48
|
+
id?: string;
|
|
39
49
|
toolInvocations?: Array<{
|
|
40
50
|
toolCallId: string;
|
|
41
51
|
toolName: string;
|
|
@@ -43,19 +53,19 @@ export declare const createUIMessage: ({ content, role, id, toolInvocations, }:
|
|
|
43
53
|
result: Record<string, any>;
|
|
44
54
|
state: any;
|
|
45
55
|
}>;
|
|
46
|
-
})
|
|
47
|
-
export declare const createAgentTestRun: ({ inputMessages, output, rememberedMessages, systemMessages, taggedSystemMessages,
|
|
56
|
+
}): MastraDBMessage;
|
|
57
|
+
export declare const createAgentTestRun: ({ inputMessages, output, rememberedMessages, systemMessages, taggedSystemMessages, requestContext, runId, }: {
|
|
48
58
|
inputMessages?: ScorerRunInputForAgent["inputMessages"];
|
|
49
59
|
output: ScorerRunOutputForAgent;
|
|
50
60
|
rememberedMessages?: ScorerRunInputForAgent["rememberedMessages"];
|
|
51
61
|
systemMessages?: ScorerRunInputForAgent["systemMessages"];
|
|
52
62
|
taggedSystemMessages?: ScorerRunInputForAgent["taggedSystemMessages"];
|
|
53
|
-
|
|
63
|
+
requestContext?: RequestContext;
|
|
54
64
|
runId?: string;
|
|
55
65
|
}) => {
|
|
56
66
|
input: ScorerRunInputForAgent;
|
|
57
67
|
output: ScorerRunOutputForAgent;
|
|
58
|
-
|
|
68
|
+
requestContext: RequestContext;
|
|
59
69
|
runId: string;
|
|
60
70
|
};
|
|
61
71
|
export type ToolCallInfo = {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../src/scorers/utils.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,
|
|
1
|
+
{"version":3,"file":"utils.d.ts","sourceRoot":"","sources":["../../src/scorers/utils.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAC;AAC1D,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AACxG,OAAO,EAAE,cAAc,EAAE,MAAM,8BAA8B,CAAC;AAC9D,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,IAAI,CAAC;AAEzC;;;GAGG;AACH,wBAAgB,iCAAiC,CAAC,OAAO,EAAE,eAAe,GAAG,MAAM,CAUlF;AAED,eAAO,MAAM,kBAAkB,GAAI,KAAK,MAAM,WAE7C,CAAC;AAEF,wBAAgB,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAEnF;AAED,MAAM,MAAM,QAAQ,GAAG;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,cAAc,EAAE;QACd,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,CAAC,EAAE,MAAM,CAAC;KACjB,CAAC;CACH,CAAC;AAEF,MAAM,MAAM,mBAAmB,GAAG,QAAQ,GAAG;IAC3C,OAAO,EAAE,MAAM,EAAE,CAAC;CACnB,CAAC;AAEF,eAAO,MAAM,aAAa,GACxB,OAAO,MAAM,EACb,QAAQ,MAAM,EACd,oBAAoB,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,EACvC,iBAAiB,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,KACnC,YAOF,CAAC;AAEF,eAAO,MAAM,0BAA0B,GAAI,QAAQ,sBAAsB,KAAG,MAAM,GAAG,SAGpF,CAAC;AAEF,eAAO,MAAM,6BAA6B,GAAI,QAAQ,sBAAsB,KAAG,MAAM,EAoCpF,CAAC;AAEF,eAAO,MAAM,uBAAuB,GAAI,QAAQ,sBAAsB,KAAG,MAGxE,CAAC;AAEF,eAAO,MAAM,gCAAgC,GAAI,SAAS,uBAAuB,uBAGhF,CAAC;AAEF,eAAO,MAAM,oBAAoB,GAAI,gDAMlC;IACD,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC1B,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC5B,KAAK,CAAC,EAAE,cAAc,CAAC,OAAO,CAAC,CAAC;CACjC,KAAG;IAAE,UAAU,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAQhH,CAAC;AAEF;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,EAChC,OAAO,EACP,IAAI,EACJ,EAAmB,EACnB,eAAoB,GACrB,EAAE;IACD,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,GAAG,WAAW,GAAG,QAAQ,CAAC;IACtC,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,eAAe,CAAC,EAAE,KAAK,CAAC;QACtB,UAAU,EAAE,MAAM,CAAC;QACnB,QAAQ,EAAE,MAAM,CAAC;QACjB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAC1B,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QAC5B,KAAK,EAAE,GAAG,CAAC;KACZ,CAAC,CAAC;CACJ,GAAG,eAAe,CAoBlB;AAED,eAAO,MAAM,kBAAkB,GAAI,6GAQhC;IACD,aAAa,CAAC,EAAE,sBAAsB,CAAC,eAAe,CAAC,CAAC;IACxD,MAAM,EAAE,uBAAuB,CAAC;IAChC,kBAAkB,CAAC,EAAE,sBAAsB,CAAC,oBAAoB,CAAC,CAAC;IAClE,cAAc,CAAC,EAAE,sBAAsB,CAAC,gBAAgB,CAAC,CAAC;IAC1D,oBAAoB,CAAC,EAAE,sBAAsB,CAAC,sBAAsB,CAAC,CAAC;IACtE,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB,KAAG;IACF,KAAK,EAAE,sBAAsB,CAAC;IAC9B,MAAM,EAAE,uBAAuB,CAAC;IAChC,cAAc,EAAE,cAAc,CAAC;IAC/B,KAAK,EAAE,MAAM,CAAC;CAaf,CAAC;AAEF,MAAM,MAAM,YAAY,GAAG;IACzB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;CACzB,CAAC;AAEF,wBAAgB,gBAAgB,CAAC,MAAM,EAAE,uBAAuB,GAAG;IAAE,KAAK,EAAE,MAAM,EAAE,CAAC;IAAC,aAAa,EAAE,YAAY,EAAE,CAAA;CAAE,CAwBpH;AAED,eAAO,MAAM,oBAAoB,GAAI,UAAU,sBAAsB,GAAG,SAAS,KAAG,MAAM,EAEzF,CAAC;AAEF,eAAO,MAAM,4BAA4B,GAAI,WAAW,uBAAuB,KAAG,MAAM,EAEvF,CAAC"}
|
package/dist/scorers/utils.js
CHANGED
|
@@ -1,3 +1,3 @@
|
|
|
1
|
-
export { createAgentTestRun, createTestRun, createToolInvocation,
|
|
1
|
+
export { createAgentTestRun, createTestMessage, createTestRun, createToolInvocation, extractAgentResponseMessages, extractInputMessages, extractToolCalls, getAssistantMessageFromRunOutput, getCombinedSystemPrompt, getSystemMessagesFromRunInput, getTextContentFromMastraDBMessage, getUserMessageFromRunInput, isCloserTo, roundToTwoDecimals } from '../chunk-CCLM7KPF.js';
|
|
2
2
|
//# sourceMappingURL=utils.js.map
|
|
3
3
|
//# sourceMappingURL=utils.js.map
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@mastra/evals",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "1.0.0-beta.0",
|
|
4
4
|
"description": "",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"files": [
|
|
@@ -29,54 +29,14 @@
|
|
|
29
29
|
"default": "./dist/index.cjs"
|
|
30
30
|
}
|
|
31
31
|
},
|
|
32
|
-
"./
|
|
32
|
+
"./scorers/prebuilt": {
|
|
33
33
|
"import": {
|
|
34
|
-
"types": "./dist/
|
|
35
|
-
"default": "./dist/
|
|
34
|
+
"types": "./dist/scorers/prebuilt/index.d.ts",
|
|
35
|
+
"default": "./dist/scorers/prebuilt/index.js"
|
|
36
36
|
},
|
|
37
37
|
"require": {
|
|
38
|
-
"types": "./dist/
|
|
39
|
-
"default": "./dist/
|
|
40
|
-
}
|
|
41
|
-
},
|
|
42
|
-
"./nlp": {
|
|
43
|
-
"import": {
|
|
44
|
-
"types": "./dist/metrics/nlp/index.d.ts",
|
|
45
|
-
"default": "./dist/metrics/nlp/index.js"
|
|
46
|
-
},
|
|
47
|
-
"require": {
|
|
48
|
-
"types": "./dist/metrics/nlp/index.d.ts",
|
|
49
|
-
"default": "./dist/metrics/nlp/index.cjs"
|
|
50
|
-
}
|
|
51
|
-
},
|
|
52
|
-
"./llm": {
|
|
53
|
-
"import": {
|
|
54
|
-
"types": "./dist/metrics/llm/index.d.ts",
|
|
55
|
-
"default": "./dist/metrics/llm/index.js"
|
|
56
|
-
},
|
|
57
|
-
"require": {
|
|
58
|
-
"types": "./dist/metrics/llm/index.d.ts",
|
|
59
|
-
"default": "./dist/metrics/llm/index.cjs"
|
|
60
|
-
}
|
|
61
|
-
},
|
|
62
|
-
"./scorers/llm": {
|
|
63
|
-
"import": {
|
|
64
|
-
"types": "./dist/scorers/llm/index.d.ts",
|
|
65
|
-
"default": "./dist/scorers/llm/index.js"
|
|
66
|
-
},
|
|
67
|
-
"require": {
|
|
68
|
-
"types": "./dist/scorers/llm/index.d.ts",
|
|
69
|
-
"default": "./dist/scorers/llm/index.cjs"
|
|
70
|
-
}
|
|
71
|
-
},
|
|
72
|
-
"./scorers/code": {
|
|
73
|
-
"import": {
|
|
74
|
-
"types": "./dist/scorers/code/index.d.ts",
|
|
75
|
-
"default": "./dist/scorers/code/index.js"
|
|
76
|
-
},
|
|
77
|
-
"require": {
|
|
78
|
-
"types": "./dist/scorers/code/index.d.ts",
|
|
79
|
-
"default": "./dist/scorers/code/index.cjs"
|
|
38
|
+
"types": "./dist/scorers/prebuilt/index.d.ts",
|
|
39
|
+
"default": "./dist/scorers/prebuilt/index.cjs"
|
|
80
40
|
}
|
|
81
41
|
},
|
|
82
42
|
"./scorers/utils": {
|
|
@@ -96,21 +56,19 @@
|
|
|
96
56
|
"license": "Apache-2.0",
|
|
97
57
|
"dependencies": {
|
|
98
58
|
"compromise": "^14.14.4",
|
|
99
|
-
"difflib": "^0.2.4",
|
|
100
59
|
"fs-extra": "^11.3.2",
|
|
101
60
|
"keyword-extractor": "^0.0.28",
|
|
102
61
|
"sentiment": "^5.0.2",
|
|
103
62
|
"string-similarity": "^4.0.4"
|
|
104
63
|
},
|
|
105
64
|
"peerDependencies": {
|
|
106
|
-
"@mastra/core": ">=0.
|
|
65
|
+
"@mastra/core": ">=1.0.0-0 <2.0.0-0",
|
|
107
66
|
"ai": "^4.0.0 || ^5.0.0",
|
|
108
67
|
"zod": "^3.25.0 || ^4.0.0"
|
|
109
68
|
},
|
|
110
69
|
"devDependencies": {
|
|
111
70
|
"@ai-sdk/openai": "^1.3.24",
|
|
112
71
|
"@microsoft/api-extractor": "^7.52.8",
|
|
113
|
-
"@types/difflib": "^0.2.7",
|
|
114
72
|
"@types/fs-extra": "^11.0.4",
|
|
115
73
|
"@types/sentiment": "^5.0.4",
|
|
116
74
|
"@types/string-similarity": "^4.0.2",
|
|
@@ -121,16 +79,12 @@
|
|
|
121
79
|
"typescript": "^5.8.3",
|
|
122
80
|
"vitest": "^3.2.4",
|
|
123
81
|
"zod": "^3.25.76",
|
|
124
|
-
"@
|
|
125
|
-
"@
|
|
126
|
-
"@internal/types-builder": "0.0.
|
|
82
|
+
"@internal/lint": "0.0.53",
|
|
83
|
+
"@mastra/core": "1.0.0-beta.0",
|
|
84
|
+
"@internal/types-builder": "0.0.28"
|
|
127
85
|
},
|
|
128
|
-
"
|
|
129
|
-
"
|
|
130
|
-
"publish-branch": [
|
|
131
|
-
"main",
|
|
132
|
-
"0.x"
|
|
133
|
-
]
|
|
86
|
+
"engines": {
|
|
87
|
+
"node": ">=22.13.0"
|
|
134
88
|
},
|
|
135
89
|
"scripts": {
|
|
136
90
|
"check": "tsc --noEmit",
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"attachListeners.d.ts","sourceRoot":"","sources":["../src/attachListeners.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,cAAc,CAAC;AAO3C,wBAAsB,eAAe,CAAC,MAAM,CAAC,EAAE,MAAM,iBA0BpD;AAED,wBAAsB,WAAW,kBAOhC"}
|
package/dist/chunk-7QAUEU4L.cjs
DELETED
|
@@ -1,10 +0,0 @@
|
|
|
1
|
-
'use strict';
|
|
2
|
-
|
|
3
|
-
// src/metrics/llm/utils.ts
|
|
4
|
-
var roundToTwoDecimals = (num) => {
|
|
5
|
-
return Math.round((num + Number.EPSILON) * 100) / 100;
|
|
6
|
-
};
|
|
7
|
-
|
|
8
|
-
exports.roundToTwoDecimals = roundToTwoDecimals;
|
|
9
|
-
//# sourceMappingURL=chunk-7QAUEU4L.cjs.map
|
|
10
|
-
//# sourceMappingURL=chunk-7QAUEU4L.cjs.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/metrics/llm/utils.ts"],"names":[],"mappings":";;;AAAO,IAAM,kBAAA,GAAqB,CAAC,GAAA,KAAgB;AACjD,EAAA,OAAO,KAAK,KAAA,CAAA,CAAO,GAAA,GAAM,MAAA,CAAO,OAAA,IAAW,GAAG,CAAA,GAAI,GAAA;AACpD","file":"chunk-7QAUEU4L.cjs","sourcesContent":["export const roundToTwoDecimals = (num: number) => {\n return Math.round((num + Number.EPSILON) * 100) / 100;\n};\n\nexport function isCloserTo(value: number, target1: number, target2: number): boolean {\n return Math.abs(value - target1) < Math.abs(value - target2);\n}\n\nexport type TestCase = {\n input: string;\n output: string;\n expectedResult: {\n score: number;\n reason?: string;\n };\n};\n\nexport type TestCaseWithContext = TestCase & {\n context: string[];\n};\n"]}
|
package/dist/chunk-EMMSS5I5.cjs
DELETED
|
@@ -1,37 +0,0 @@
|
|
|
1
|
-
'use strict';
|
|
2
|
-
|
|
3
|
-
var __create = Object.create;
|
|
4
|
-
var __defProp = Object.defineProperty;
|
|
5
|
-
var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
|
|
6
|
-
var __getOwnPropNames = Object.getOwnPropertyNames;
|
|
7
|
-
var __getProtoOf = Object.getPrototypeOf;
|
|
8
|
-
var __hasOwnProp = Object.prototype.hasOwnProperty;
|
|
9
|
-
var __commonJS = (cb, mod) => function __require() {
|
|
10
|
-
return mod || (0, cb[__getOwnPropNames(cb)[0]])((mod = { exports: {} }).exports, mod), mod.exports;
|
|
11
|
-
};
|
|
12
|
-
var __export = (target, all) => {
|
|
13
|
-
for (var name in all)
|
|
14
|
-
__defProp(target, name, { get: all[name], enumerable: true });
|
|
15
|
-
};
|
|
16
|
-
var __copyProps = (to, from, except, desc) => {
|
|
17
|
-
if (from && typeof from === "object" || typeof from === "function") {
|
|
18
|
-
for (let key of __getOwnPropNames(from))
|
|
19
|
-
if (!__hasOwnProp.call(to, key) && key !== except)
|
|
20
|
-
__defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
|
|
21
|
-
}
|
|
22
|
-
return to;
|
|
23
|
-
};
|
|
24
|
-
var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(
|
|
25
|
-
// If the importer is in node compatibility mode or this is not an ESM
|
|
26
|
-
// file that has been converted to a CommonJS file using a Babel-
|
|
27
|
-
// compatible transform (i.e. "__esModule" has not been set), then set
|
|
28
|
-
// "default" to the CommonJS "module.exports" for node compatibility.
|
|
29
|
-
isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target,
|
|
30
|
-
mod
|
|
31
|
-
));
|
|
32
|
-
|
|
33
|
-
exports.__commonJS = __commonJS;
|
|
34
|
-
exports.__export = __export;
|
|
35
|
-
exports.__toESM = __toESM;
|
|
36
|
-
//# sourceMappingURL=chunk-EMMSS5I5.cjs.map
|
|
37
|
-
//# sourceMappingURL=chunk-EMMSS5I5.cjs.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"sources":[],"names":[],"mappings":"","file":"chunk-EMMSS5I5.cjs"}
|
package/dist/chunk-G3PMV62Z.js
DELETED
|
@@ -1,33 +0,0 @@
|
|
|
1
|
-
var __create = Object.create;
|
|
2
|
-
var __defProp = Object.defineProperty;
|
|
3
|
-
var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
|
|
4
|
-
var __getOwnPropNames = Object.getOwnPropertyNames;
|
|
5
|
-
var __getProtoOf = Object.getPrototypeOf;
|
|
6
|
-
var __hasOwnProp = Object.prototype.hasOwnProperty;
|
|
7
|
-
var __commonJS = (cb, mod) => function __require() {
|
|
8
|
-
return mod || (0, cb[__getOwnPropNames(cb)[0]])((mod = { exports: {} }).exports, mod), mod.exports;
|
|
9
|
-
};
|
|
10
|
-
var __export = (target, all) => {
|
|
11
|
-
for (var name in all)
|
|
12
|
-
__defProp(target, name, { get: all[name], enumerable: true });
|
|
13
|
-
};
|
|
14
|
-
var __copyProps = (to, from, except, desc) => {
|
|
15
|
-
if (from && typeof from === "object" || typeof from === "function") {
|
|
16
|
-
for (let key of __getOwnPropNames(from))
|
|
17
|
-
if (!__hasOwnProp.call(to, key) && key !== except)
|
|
18
|
-
__defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
|
|
19
|
-
}
|
|
20
|
-
return to;
|
|
21
|
-
};
|
|
22
|
-
var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(
|
|
23
|
-
// If the importer is in node compatibility mode or this is not an ESM
|
|
24
|
-
// file that has been converted to a CommonJS file using a Babel-
|
|
25
|
-
// compatible transform (i.e. "__esModule" has not been set), then set
|
|
26
|
-
// "default" to the CommonJS "module.exports" for node compatibility.
|
|
27
|
-
isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target,
|
|
28
|
-
mod
|
|
29
|
-
));
|
|
30
|
-
|
|
31
|
-
export { __commonJS, __export, __toESM };
|
|
32
|
-
//# sourceMappingURL=chunk-G3PMV62Z.js.map
|
|
33
|
-
//# sourceMappingURL=chunk-G3PMV62Z.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"sources":[],"names":[],"mappings":"","file":"chunk-G3PMV62Z.js"}
|
package/dist/chunk-IUSAD2BW.cjs
DELETED
|
@@ -1,19 +0,0 @@
|
|
|
1
|
-
'use strict';
|
|
2
|
-
|
|
3
|
-
var agent = require('@mastra/core/agent');
|
|
4
|
-
|
|
5
|
-
// src/metrics/judge/index.ts
|
|
6
|
-
var MastraAgentJudge = class {
|
|
7
|
-
agent;
|
|
8
|
-
constructor(name, instructions, model) {
|
|
9
|
-
this.agent = new agent.Agent({
|
|
10
|
-
name: `Mastra Eval Judge ${name}`,
|
|
11
|
-
instructions,
|
|
12
|
-
model
|
|
13
|
-
});
|
|
14
|
-
}
|
|
15
|
-
};
|
|
16
|
-
|
|
17
|
-
exports.MastraAgentJudge = MastraAgentJudge;
|
|
18
|
-
//# sourceMappingURL=chunk-IUSAD2BW.cjs.map
|
|
19
|
-
//# sourceMappingURL=chunk-IUSAD2BW.cjs.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/metrics/judge/index.ts"],"names":["Agent"],"mappings":";;;;;AAGO,IAAe,mBAAf,MAAgC;AAAA,EAClB,KAAA;AAAA,EAEnB,WAAA,CAAY,IAAA,EAAc,YAAA,EAAsB,KAAA,EAAsB;AACpE,IAAA,IAAA,CAAK,KAAA,GAAQ,IAAIA,WAAA,CAAM;AAAA,MACrB,IAAA,EAAM,qBAAqB,IAAI,CAAA,CAAA;AAAA,MAC/B,YAAA;AAAA,MACA;AAAA,KACD,CAAA;AAAA,EACH;AACF","file":"chunk-IUSAD2BW.cjs","sourcesContent":["import { Agent } from '@mastra/core/agent';\nimport type { LanguageModel } from '@mastra/core/llm';\n\nexport abstract class MastraAgentJudge {\n protected readonly agent: Agent;\n\n constructor(name: string, instructions: string, model: LanguageModel) {\n this.agent = new Agent({\n name: `Mastra Eval Judge ${name}`,\n instructions: instructions,\n model,\n });\n }\n}\n"]}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/scorers/utils.ts"],"names":[],"mappings":";;;AAIO,IAAM,kBAAA,GAAqB,CAAC,GAAA,KAAgB;AACjD,EAAA,OAAO,KAAK,KAAA,CAAA,CAAO,GAAA,GAAM,MAAA,CAAO,OAAA,IAAW,GAAG,CAAA,GAAI,GAAA;AACpD;AAEO,SAAS,UAAA,CAAW,KAAA,EAAe,OAAA,EAAiB,OAAA,EAA0B;AACnF,EAAA,OAAO,IAAA,CAAK,IAAI,KAAA,GAAQ,OAAO,IAAI,IAAA,CAAK,GAAA,CAAI,QAAQ,OAAO,CAAA;AAC7D;AAeO,IAAM,aAAA,GAAgB,CAC3B,KAAA,EACA,MAAA,EACA,mBACA,cAAA,KACiB;AACjB,EAAA,OAAO;AAAA,IACL,OAAO,CAAC,EAAE,MAAM,MAAA,EAAQ,OAAA,EAAS,OAAO,CAAA;AAAA,IACxC,MAAA,EAAQ,EAAE,IAAA,EAAM,WAAA,EAAa,MAAM,MAAA,EAAO;AAAA,IAC1C,iBAAA,EAAmB,qBAAqB,EAAC;AAAA,IACzC,cAAA,EAAgB,kBAAkB;AAAC,GACrC;AACF;AAEO,IAAM,0BAAA,GAA6B,CAAC,KAAA,KAAmC;AAC5E,EAAA,OAAO,KAAA,EAAO,cAAc,IAAA,CAAK,CAAC,EAAE,IAAA,EAAK,KAAM,IAAA,KAAS,MAAM,CAAA,EAAG,OAAA;AACnE;AAEO,IAAM,6BAAA,GAAgC,CAAC,KAAA,KAA6C;AACzF,EAAA,MAAM,iBAA2B,EAAC;AAGlC,EAAA,IAAI,OAAO,cAAA,EAAgB;AACzB,IAAA,cAAA,CAAe,IAAA;AAAA,MACb,GAAG,KAAA,CAAM,cAAA,CACN,GAAA,CAAI,CAAA,GAAA,KAAO;AAEV,QAAA,IAAI,OAAO,GAAA,CAAI,OAAA,KAAY,QAAA,EAAU;AACnC,UAAA,OAAO,GAAA,CAAI,OAAA;AAAA,QACb,CAAA,MAAA,IAAW,KAAA,CAAM,OAAA,CAAQ,GAAA,CAAI,OAAO,CAAA,EAAG;AAErC,UAAA,OAAO,GAAA,CAAI,OAAA,CACR,MAAA,CAAO,CAAA,IAAA,KAAQ,KAAK,IAAA,KAAS,MAAM,CAAA,CACnC,GAAA,CAAI,UAAQ,IAAA,CAAK,IAAA,IAAQ,EAAE,CAAA,CAC3B,KAAK,GAAG,CAAA;AAAA,QACb;AACA,QAAA,OAAO,EAAA;AAAA,MACT,CAAC,CAAA,CACA,MAAA,CAAO,CAAA,OAAA,KAAW,OAAO;AAAA,KAC9B;AAAA,EACF;AAGA,EAAA,IAAI,OAAO,oBAAA,EAAsB;AAC/B,IAAA,MAAA,CAAO,MAAA,CAAO,KAAA,CAAM,oBAAoB,CAAA,CAAE,QAAQ,CAAA,QAAA,KAAY;AAC5D,MAAA,QAAA,CAAS,QAAQ,CAAA,GAAA,KAAO;AACtB,QAAA,IAAI,OAAO,GAAA,CAAI,OAAA,KAAY,QAAA,EAAU;AACnC,UAAA,cAAA,CAAe,IAAA,CAAK,IAAI,OAAO,CAAA;AAAA,QACjC;AAAA,MACF,CAAC,CAAA;AAAA,IACH,CAAC,CAAA;AAAA,EACH;AAEA,EAAA,OAAO,cAAA;AACT;AAEO,IAAM,uBAAA,GAA0B,CAAC,KAAA,KAA2C;AACjF,EAAA,MAAM,cAAA,GAAiB,8BAA8B,KAAK,CAAA;AAC1D,EAAA,OAAO,cAAA,CAAe,KAAK,MAAM,CAAA;AACnC;AAEO,IAAM,gCAAA,GAAmC,CAAC,MAAA,KAAqC;AACpF,EAAA,OAAO,MAAA,EAAQ,KAAK,CAAC,EAAE,MAAK,KAAM,IAAA,KAAS,WAAW,CAAA,EAAG,OAAA;AAC3D;AAEO,IAAM,uBAAuB,CAAC;AAAA,EACnC,UAAA;AAAA,EACA,QAAA;AAAA,EACA,IAAA;AAAA,EACA,MAAA;AAAA,EACA,KAAA,GAAQ;AACV,CAAA,KAMuH;AACrH,EAAA,OAAO;AAAA,IACL,UAAA;AAAA,IACA,QAAA;AAAA,IACA,IAAA;AAAA,IACA,MAAA;AAAA,IACA;AAAA,GACF;AACF;AAEO,IAAM,kBAAkB,CAAC;AAAA,EAC9B,OAAA;AAAA,EACA,IAAA;AAAA,EACA,EAAA,GAAK,cAAA;AAAA,EACL,kBAAkB;AACpB,CAAA,KAWiB;AACf,EAAA,OAAO;AAAA,IACL,EAAA;AAAA,IACA,IAAA;AAAA,IACA,OAAA;AAAA,IACA,OAAO,CAAC,EAAE,MAAM,MAAA,EAAQ,IAAA,EAAM,SAAS,CAAA;AAAA,IACvC;AAAA,GACF;AACF;AAEO,IAAM,qBAAqB,CAAC;AAAA,EACjC,gBAAgB,EAAC;AAAA,EACjB,MAAA;AAAA,EACA,qBAAqB,EAAC;AAAA,EACtB,iBAAiB,EAAC;AAAA,EAClB,uBAAuB,EAAC;AAAA,EACxB,cAAA,GAAiB,IAAI,cAAA,EAAe;AAAA,EACpC,KAAA,GAAQ,OAAO,UAAA;AACjB,CAAA,KAaK;AACH,EAAA,OAAO;AAAA,IACL,KAAA,EAAO;AAAA,MACL,aAAA;AAAA,MACA,kBAAA;AAAA,MACA,cAAA;AAAA,MACA;AAAA,KACF;AAAA,IACA,MAAA;AAAA,IACA,cAAA;AAAA,IACA;AAAA,GACF;AACF;AASO,SAAS,iBAAiB,MAAA,EAAqF;AACpH,EAAA,MAAM,YAAsB,EAAC;AAC7B,EAAA,MAAM,gBAAgC,EAAC;AAEvC,EAAA,KAAA,IAAS,YAAA,GAAe,CAAA,EAAG,YAAA,GAAe,MAAA,CAAO,QAAQ,YAAA,EAAA,EAAgB;AACvE,IAAA,MAAM,OAAA,GAAU,OAAO,YAAY,CAAA;AACnC,IAAA,IAAI,SAAS,eAAA,EAAiB;AAC5B,MAAA,KAAA,IAAS,kBAAkB,CAAA,EAAG,eAAA,GAAkB,OAAA,CAAQ,eAAA,CAAgB,QAAQ,eAAA,EAAA,EAAmB;AACjG,QAAA,MAAM,UAAA,GAAa,OAAA,CAAQ,eAAA,CAAgB,eAAe,CAAA;AAC1D,QAAA,IAAI,UAAA,IAAc,WAAW,QAAA,KAAa,UAAA,CAAW,UAAU,QAAA,IAAY,UAAA,CAAW,UAAU,MAAA,CAAA,EAAS;AACvG,UAAA,SAAA,CAAU,IAAA,CAAK,WAAW,QAAQ,CAAA;AAClC,UAAA,aAAA,CAAc,IAAA,CAAK;AAAA,YACjB,UAAU,UAAA,CAAW,QAAA;AAAA,YACrB,YAAY,UAAA,CAAW,UAAA,IAAc,CAAA,EAAG,YAAY,IAAI,eAAe,CAAA,CAAA;AAAA,YACvE,YAAA;AAAA,YACA;AAAA,WACD,CAAA;AAAA,QACH;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,EAAA,OAAO,EAAE,KAAA,EAAO,SAAA,EAAW,aAAA,EAAc;AAC3C;AAEO,IAAM,oBAAA,GAAuB,CAAC,QAAA,KAA2D;AAC9F,EAAA,OAAO,UAAU,aAAA,EAAe,GAAA,CAAI,SAAO,GAAA,CAAI,OAAO,KAAK,EAAC;AAC9D;AAEO,IAAM,4BAAA,GAA+B,CAAC,SAAA,KAAiD;AAC5F,EAAA,OAAO,SAAA,CAAU,MAAA,CAAO,CAAA,GAAA,KAAO,GAAA,CAAI,IAAA,KAAS,WAAW,CAAA,CAAE,GAAA,CAAI,CAAA,GAAA,KAAO,GAAA,CAAI,OAAO,CAAA;AACjF","file":"chunk-KHEXN75Q.js","sourcesContent":["import { RuntimeContext } from '@mastra/core/runtime-context';\nimport type { ScorerRunInputForAgent, ScorerRunOutputForAgent, ScoringInput } from '@mastra/core/scores';\nimport type { ToolInvocation, UIMessage } from 'ai';\n\nexport const roundToTwoDecimals = (num: number) => {\n return Math.round((num + Number.EPSILON) * 100) / 100;\n};\n\nexport function isCloserTo(value: number, target1: number, target2: number): boolean {\n return Math.abs(value - target1) < Math.abs(value - target2);\n}\n\nexport type TestCase = {\n input: string;\n output: string;\n expectedResult: {\n score: number;\n reason?: string;\n };\n};\n\nexport type TestCaseWithContext = TestCase & {\n context: string[];\n};\n\nexport const createTestRun = (\n input: string,\n output: string,\n additionalContext?: Record<string, any>,\n runtimeContext?: Record<string, any>,\n): ScoringInput => {\n return {\n input: [{ role: 'user', content: input }],\n output: { role: 'assistant', text: output },\n additionalContext: additionalContext ?? {},\n runtimeContext: runtimeContext ?? {},\n };\n};\n\nexport const getUserMessageFromRunInput = (input?: ScorerRunInputForAgent) => {\n return input?.inputMessages.find(({ role }) => role === 'user')?.content;\n};\n\nexport const getSystemMessagesFromRunInput = (input?: ScorerRunInputForAgent): string[] => {\n const systemMessages: string[] = [];\n\n // Add standard system messages\n if (input?.systemMessages) {\n systemMessages.push(\n ...input.systemMessages\n .map(msg => {\n // Handle different content types - extract text if it's an array of parts\n if (typeof msg.content === 'string') {\n return msg.content;\n } else if (Array.isArray(msg.content)) {\n // Extract text from parts array\n return msg.content\n .filter(part => part.type === 'text')\n .map(part => part.text || '')\n .join(' ');\n }\n return '';\n })\n .filter(content => content),\n );\n }\n\n // Add tagged system messages (these are specialized system prompts)\n if (input?.taggedSystemMessages) {\n Object.values(input.taggedSystemMessages).forEach(messages => {\n messages.forEach(msg => {\n if (typeof msg.content === 'string') {\n systemMessages.push(msg.content);\n }\n });\n });\n }\n\n return systemMessages;\n};\n\nexport const getCombinedSystemPrompt = (input?: ScorerRunInputForAgent): string => {\n const systemMessages = getSystemMessagesFromRunInput(input);\n return systemMessages.join('\\n\\n');\n};\n\nexport const getAssistantMessageFromRunOutput = (output?: ScorerRunOutputForAgent) => {\n return output?.find(({ role }) => role === 'assistant')?.content;\n};\n\nexport const createToolInvocation = ({\n toolCallId,\n toolName,\n args,\n result,\n state = 'result',\n}: {\n toolCallId: string;\n toolName: string;\n args: Record<string, any>;\n result: Record<string, any>;\n state?: ToolInvocation['state'];\n}): { toolCallId: string; toolName: string; args: Record<string, any>; result: Record<string, any>; state: string } => {\n return {\n toolCallId,\n toolName,\n args,\n result,\n state,\n };\n};\n\nexport const createUIMessage = ({\n content,\n role,\n id = 'test-message',\n toolInvocations = [],\n}: {\n id: string;\n role: 'user' | 'assistant' | 'system';\n content: string;\n toolInvocations?: Array<{\n toolCallId: string;\n toolName: string;\n args: Record<string, any>;\n result: Record<string, any>;\n state: any;\n }>;\n}): UIMessage => {\n return {\n id,\n role,\n content,\n parts: [{ type: 'text', text: content }],\n toolInvocations,\n };\n};\n\nexport const createAgentTestRun = ({\n inputMessages = [],\n output,\n rememberedMessages = [],\n systemMessages = [],\n taggedSystemMessages = {},\n runtimeContext = new RuntimeContext(),\n runId = crypto.randomUUID(),\n}: {\n inputMessages?: ScorerRunInputForAgent['inputMessages'];\n output: ScorerRunOutputForAgent;\n rememberedMessages?: ScorerRunInputForAgent['rememberedMessages'];\n systemMessages?: ScorerRunInputForAgent['systemMessages'];\n taggedSystemMessages?: ScorerRunInputForAgent['taggedSystemMessages'];\n runtimeContext?: RuntimeContext;\n runId?: string;\n}): {\n input: ScorerRunInputForAgent;\n output: ScorerRunOutputForAgent;\n runtimeContext: RuntimeContext;\n runId: string;\n} => {\n return {\n input: {\n inputMessages,\n rememberedMessages,\n systemMessages,\n taggedSystemMessages,\n },\n output,\n runtimeContext,\n runId,\n };\n};\n\nexport type ToolCallInfo = {\n toolName: string;\n toolCallId: string;\n messageIndex: number;\n invocationIndex: number;\n};\n\nexport function extractToolCalls(output: ScorerRunOutputForAgent): { tools: string[]; toolCallInfos: ToolCallInfo[] } {\n const toolCalls: string[] = [];\n const toolCallInfos: ToolCallInfo[] = [];\n\n for (let messageIndex = 0; messageIndex < output.length; messageIndex++) {\n const message = output[messageIndex];\n if (message?.toolInvocations) {\n for (let invocationIndex = 0; invocationIndex < message.toolInvocations.length; invocationIndex++) {\n const invocation = message.toolInvocations[invocationIndex];\n if (invocation && invocation.toolName && (invocation.state === 'result' || invocation.state === 'call')) {\n toolCalls.push(invocation.toolName);\n toolCallInfos.push({\n toolName: invocation.toolName,\n toolCallId: invocation.toolCallId || `${messageIndex}-${invocationIndex}`,\n messageIndex,\n invocationIndex,\n });\n }\n }\n }\n }\n\n return { tools: toolCalls, toolCallInfos };\n}\n\nexport const extractInputMessages = (runInput: ScorerRunInputForAgent | undefined): string[] => {\n return runInput?.inputMessages?.map(msg => msg.content) || [];\n};\n\nexport const extractAgentResponseMessages = (runOutput: ScorerRunOutputForAgent): string[] => {\n return runOutput.filter(msg => msg.role === 'assistant').map(msg => msg.content);\n};\n"]}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/scorers/utils.ts"],"names":["runtimeContext","RuntimeContext"],"mappings":";;;;;AAIO,IAAM,kBAAA,GAAqB,CAAC,GAAA,KAAgB;AACjD,EAAA,OAAO,KAAK,KAAA,CAAA,CAAO,GAAA,GAAM,MAAA,CAAO,OAAA,IAAW,GAAG,CAAA,GAAI,GAAA;AACpD;AAEO,SAAS,UAAA,CAAW,KAAA,EAAe,OAAA,EAAiB,OAAA,EAA0B;AACnF,EAAA,OAAO,IAAA,CAAK,IAAI,KAAA,GAAQ,OAAO,IAAI,IAAA,CAAK,GAAA,CAAI,QAAQ,OAAO,CAAA;AAC7D;AAeO,IAAM,aAAA,GAAgB,CAC3B,KAAA,EACA,MAAA,EACA,mBACA,cAAA,KACiB;AACjB,EAAA,OAAO;AAAA,IACL,OAAO,CAAC,EAAE,MAAM,MAAA,EAAQ,OAAA,EAAS,OAAO,CAAA;AAAA,IACxC,MAAA,EAAQ,EAAE,IAAA,EAAM,WAAA,EAAa,MAAM,MAAA,EAAO;AAAA,IAC1C,iBAAA,EAAmB,qBAAqB,EAAC;AAAA,IACzC,cAAA,EAAgB,kBAAkB;AAAC,GACrC;AACF;AAEO,IAAM,0BAAA,GAA6B,CAAC,KAAA,KAAmC;AAC5E,EAAA,OAAO,KAAA,EAAO,cAAc,IAAA,CAAK,CAAC,EAAE,IAAA,EAAK,KAAM,IAAA,KAAS,MAAM,CAAA,EAAG,OAAA;AACnE;AAEO,IAAM,6BAAA,GAAgC,CAAC,KAAA,KAA6C;AACzF,EAAA,MAAM,iBAA2B,EAAC;AAGlC,EAAA,IAAI,OAAO,cAAA,EAAgB;AACzB,IAAA,cAAA,CAAe,IAAA;AAAA,MACb,GAAG,KAAA,CAAM,cAAA,CACN,GAAA,CAAI,CAAA,GAAA,KAAO;AAEV,QAAA,IAAI,OAAO,GAAA,CAAI,OAAA,KAAY,QAAA,EAAU;AACnC,UAAA,OAAO,GAAA,CAAI,OAAA;AAAA,QACb,CAAA,MAAA,IAAW,KAAA,CAAM,OAAA,CAAQ,GAAA,CAAI,OAAO,CAAA,EAAG;AAErC,UAAA,OAAO,GAAA,CAAI,OAAA,CACR,MAAA,CAAO,CAAA,IAAA,KAAQ,KAAK,IAAA,KAAS,MAAM,CAAA,CACnC,GAAA,CAAI,UAAQ,IAAA,CAAK,IAAA,IAAQ,EAAE,CAAA,CAC3B,KAAK,GAAG,CAAA;AAAA,QACb;AACA,QAAA,OAAO,EAAA;AAAA,MACT,CAAC,CAAA,CACA,MAAA,CAAO,CAAA,OAAA,KAAW,OAAO;AAAA,KAC9B;AAAA,EACF;AAGA,EAAA,IAAI,OAAO,oBAAA,EAAsB;AAC/B,IAAA,MAAA,CAAO,MAAA,CAAO,KAAA,CAAM,oBAAoB,CAAA,CAAE,QAAQ,CAAA,QAAA,KAAY;AAC5D,MAAA,QAAA,CAAS,QAAQ,CAAA,GAAA,KAAO;AACtB,QAAA,IAAI,OAAO,GAAA,CAAI,OAAA,KAAY,QAAA,EAAU;AACnC,UAAA,cAAA,CAAe,IAAA,CAAK,IAAI,OAAO,CAAA;AAAA,QACjC;AAAA,MACF,CAAC,CAAA;AAAA,IACH,CAAC,CAAA;AAAA,EACH;AAEA,EAAA,OAAO,cAAA;AACT;AAEO,IAAM,uBAAA,GAA0B,CAAC,KAAA,KAA2C;AACjF,EAAA,MAAM,cAAA,GAAiB,8BAA8B,KAAK,CAAA;AAC1D,EAAA,OAAO,cAAA,CAAe,KAAK,MAAM,CAAA;AACnC;AAEO,IAAM,gCAAA,GAAmC,CAAC,MAAA,KAAqC;AACpF,EAAA,OAAO,MAAA,EAAQ,KAAK,CAAC,EAAE,MAAK,KAAM,IAAA,KAAS,WAAW,CAAA,EAAG,OAAA;AAC3D;AAEO,IAAM,uBAAuB,CAAC;AAAA,EACnC,UAAA;AAAA,EACA,QAAA;AAAA,EACA,IAAA;AAAA,EACA,MAAA;AAAA,EACA,KAAA,GAAQ;AACV,CAAA,KAMuH;AACrH,EAAA,OAAO;AAAA,IACL,UAAA;AAAA,IACA,QAAA;AAAA,IACA,IAAA;AAAA,IACA,MAAA;AAAA,IACA;AAAA,GACF;AACF;AAEO,IAAM,kBAAkB,CAAC;AAAA,EAC9B,OAAA;AAAA,EACA,IAAA;AAAA,EACA,EAAA,GAAK,cAAA;AAAA,EACL,kBAAkB;AACpB,CAAA,KAWiB;AACf,EAAA,OAAO;AAAA,IACL,EAAA;AAAA,IACA,IAAA;AAAA,IACA,OAAA;AAAA,IACA,OAAO,CAAC,EAAE,MAAM,MAAA,EAAQ,IAAA,EAAM,SAAS,CAAA;AAAA,IACvC;AAAA,GACF;AACF;AAEO,IAAM,qBAAqB,CAAC;AAAA,EACjC,gBAAgB,EAAC;AAAA,EACjB,MAAA;AAAA,EACA,qBAAqB,EAAC;AAAA,EACtB,iBAAiB,EAAC;AAAA,EAClB,uBAAuB,EAAC;AAAA,kBACxBA,gBAAA,GAAiB,IAAIC,6BAAA,EAAe;AAAA,EACpC,KAAA,GAAQ,OAAO,UAAA;AACjB,CAAA,KAaK;AACH,EAAA,OAAO;AAAA,IACL,KAAA,EAAO;AAAA,MACL,aAAA;AAAA,MACA,kBAAA;AAAA,MACA,cAAA;AAAA,MACA;AAAA,KACF;AAAA,IACA,MAAA;AAAA,oBACAD,gBAAA;AAAA,IACA;AAAA,GACF;AACF;AASO,SAAS,iBAAiB,MAAA,EAAqF;AACpH,EAAA,MAAM,YAAsB,EAAC;AAC7B,EAAA,MAAM,gBAAgC,EAAC;AAEvC,EAAA,KAAA,IAAS,YAAA,GAAe,CAAA,EAAG,YAAA,GAAe,MAAA,CAAO,QAAQ,YAAA,EAAA,EAAgB;AACvE,IAAA,MAAM,OAAA,GAAU,OAAO,YAAY,CAAA;AACnC,IAAA,IAAI,SAAS,eAAA,EAAiB;AAC5B,MAAA,KAAA,IAAS,kBAAkB,CAAA,EAAG,eAAA,GAAkB,OAAA,CAAQ,eAAA,CAAgB,QAAQ,eAAA,EAAA,EAAmB;AACjG,QAAA,MAAM,UAAA,GAAa,OAAA,CAAQ,eAAA,CAAgB,eAAe,CAAA;AAC1D,QAAA,IAAI,UAAA,IAAc,WAAW,QAAA,KAAa,UAAA,CAAW,UAAU,QAAA,IAAY,UAAA,CAAW,UAAU,MAAA,CAAA,EAAS;AACvG,UAAA,SAAA,CAAU,IAAA,CAAK,WAAW,QAAQ,CAAA;AAClC,UAAA,aAAA,CAAc,IAAA,CAAK;AAAA,YACjB,UAAU,UAAA,CAAW,QAAA;AAAA,YACrB,YAAY,UAAA,CAAW,UAAA,IAAc,CAAA,EAAG,YAAY,IAAI,eAAe,CAAA,CAAA;AAAA,YACvE,YAAA;AAAA,YACA;AAAA,WACD,CAAA;AAAA,QACH;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,EAAA,OAAO,EAAE,KAAA,EAAO,SAAA,EAAW,aAAA,EAAc;AAC3C;AAEO,IAAM,oBAAA,GAAuB,CAAC,QAAA,KAA2D;AAC9F,EAAA,OAAO,UAAU,aAAA,EAAe,GAAA,CAAI,SAAO,GAAA,CAAI,OAAO,KAAK,EAAC;AAC9D;AAEO,IAAM,4BAAA,GAA+B,CAAC,SAAA,KAAiD;AAC5F,EAAA,OAAO,SAAA,CAAU,MAAA,CAAO,CAAA,GAAA,KAAO,GAAA,CAAI,IAAA,KAAS,WAAW,CAAA,CAAE,GAAA,CAAI,CAAA,GAAA,KAAO,GAAA,CAAI,OAAO,CAAA;AACjF","file":"chunk-QKR2PMLZ.cjs","sourcesContent":["import { RuntimeContext } from '@mastra/core/runtime-context';\nimport type { ScorerRunInputForAgent, ScorerRunOutputForAgent, ScoringInput } from '@mastra/core/scores';\nimport type { ToolInvocation, UIMessage } from 'ai';\n\nexport const roundToTwoDecimals = (num: number) => {\n return Math.round((num + Number.EPSILON) * 100) / 100;\n};\n\nexport function isCloserTo(value: number, target1: number, target2: number): boolean {\n return Math.abs(value - target1) < Math.abs(value - target2);\n}\n\nexport type TestCase = {\n input: string;\n output: string;\n expectedResult: {\n score: number;\n reason?: string;\n };\n};\n\nexport type TestCaseWithContext = TestCase & {\n context: string[];\n};\n\nexport const createTestRun = (\n input: string,\n output: string,\n additionalContext?: Record<string, any>,\n runtimeContext?: Record<string, any>,\n): ScoringInput => {\n return {\n input: [{ role: 'user', content: input }],\n output: { role: 'assistant', text: output },\n additionalContext: additionalContext ?? {},\n runtimeContext: runtimeContext ?? {},\n };\n};\n\nexport const getUserMessageFromRunInput = (input?: ScorerRunInputForAgent) => {\n return input?.inputMessages.find(({ role }) => role === 'user')?.content;\n};\n\nexport const getSystemMessagesFromRunInput = (input?: ScorerRunInputForAgent): string[] => {\n const systemMessages: string[] = [];\n\n // Add standard system messages\n if (input?.systemMessages) {\n systemMessages.push(\n ...input.systemMessages\n .map(msg => {\n // Handle different content types - extract text if it's an array of parts\n if (typeof msg.content === 'string') {\n return msg.content;\n } else if (Array.isArray(msg.content)) {\n // Extract text from parts array\n return msg.content\n .filter(part => part.type === 'text')\n .map(part => part.text || '')\n .join(' ');\n }\n return '';\n })\n .filter(content => content),\n );\n }\n\n // Add tagged system messages (these are specialized system prompts)\n if (input?.taggedSystemMessages) {\n Object.values(input.taggedSystemMessages).forEach(messages => {\n messages.forEach(msg => {\n if (typeof msg.content === 'string') {\n systemMessages.push(msg.content);\n }\n });\n });\n }\n\n return systemMessages;\n};\n\nexport const getCombinedSystemPrompt = (input?: ScorerRunInputForAgent): string => {\n const systemMessages = getSystemMessagesFromRunInput(input);\n return systemMessages.join('\\n\\n');\n};\n\nexport const getAssistantMessageFromRunOutput = (output?: ScorerRunOutputForAgent) => {\n return output?.find(({ role }) => role === 'assistant')?.content;\n};\n\nexport const createToolInvocation = ({\n toolCallId,\n toolName,\n args,\n result,\n state = 'result',\n}: {\n toolCallId: string;\n toolName: string;\n args: Record<string, any>;\n result: Record<string, any>;\n state?: ToolInvocation['state'];\n}): { toolCallId: string; toolName: string; args: Record<string, any>; result: Record<string, any>; state: string } => {\n return {\n toolCallId,\n toolName,\n args,\n result,\n state,\n };\n};\n\nexport const createUIMessage = ({\n content,\n role,\n id = 'test-message',\n toolInvocations = [],\n}: {\n id: string;\n role: 'user' | 'assistant' | 'system';\n content: string;\n toolInvocations?: Array<{\n toolCallId: string;\n toolName: string;\n args: Record<string, any>;\n result: Record<string, any>;\n state: any;\n }>;\n}): UIMessage => {\n return {\n id,\n role,\n content,\n parts: [{ type: 'text', text: content }],\n toolInvocations,\n };\n};\n\nexport const createAgentTestRun = ({\n inputMessages = [],\n output,\n rememberedMessages = [],\n systemMessages = [],\n taggedSystemMessages = {},\n runtimeContext = new RuntimeContext(),\n runId = crypto.randomUUID(),\n}: {\n inputMessages?: ScorerRunInputForAgent['inputMessages'];\n output: ScorerRunOutputForAgent;\n rememberedMessages?: ScorerRunInputForAgent['rememberedMessages'];\n systemMessages?: ScorerRunInputForAgent['systemMessages'];\n taggedSystemMessages?: ScorerRunInputForAgent['taggedSystemMessages'];\n runtimeContext?: RuntimeContext;\n runId?: string;\n}): {\n input: ScorerRunInputForAgent;\n output: ScorerRunOutputForAgent;\n runtimeContext: RuntimeContext;\n runId: string;\n} => {\n return {\n input: {\n inputMessages,\n rememberedMessages,\n systemMessages,\n taggedSystemMessages,\n },\n output,\n runtimeContext,\n runId,\n };\n};\n\nexport type ToolCallInfo = {\n toolName: string;\n toolCallId: string;\n messageIndex: number;\n invocationIndex: number;\n};\n\nexport function extractToolCalls(output: ScorerRunOutputForAgent): { tools: string[]; toolCallInfos: ToolCallInfo[] } {\n const toolCalls: string[] = [];\n const toolCallInfos: ToolCallInfo[] = [];\n\n for (let messageIndex = 0; messageIndex < output.length; messageIndex++) {\n const message = output[messageIndex];\n if (message?.toolInvocations) {\n for (let invocationIndex = 0; invocationIndex < message.toolInvocations.length; invocationIndex++) {\n const invocation = message.toolInvocations[invocationIndex];\n if (invocation && invocation.toolName && (invocation.state === 'result' || invocation.state === 'call')) {\n toolCalls.push(invocation.toolName);\n toolCallInfos.push({\n toolName: invocation.toolName,\n toolCallId: invocation.toolCallId || `${messageIndex}-${invocationIndex}`,\n messageIndex,\n invocationIndex,\n });\n }\n }\n }\n }\n\n return { tools: toolCalls, toolCallInfos };\n}\n\nexport const extractInputMessages = (runInput: ScorerRunInputForAgent | undefined): string[] => {\n return runInput?.inputMessages?.map(msg => msg.content) || [];\n};\n\nexport const extractAgentResponseMessages = (runOutput: ScorerRunOutputForAgent): string[] => {\n return runOutput.filter(msg => msg.role === 'assistant').map(msg => msg.content);\n};\n"]}
|
package/dist/chunk-QTWX6TKR.js
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/metrics/llm/utils.ts"],"names":[],"mappings":";AAAO,IAAM,kBAAA,GAAqB,CAAC,GAAA,KAAgB;AACjD,EAAA,OAAO,KAAK,KAAA,CAAA,CAAO,GAAA,GAAM,MAAA,CAAO,OAAA,IAAW,GAAG,CAAA,GAAI,GAAA;AACpD","file":"chunk-QTWX6TKR.js","sourcesContent":["export const roundToTwoDecimals = (num: number) => {\n return Math.round((num + Number.EPSILON) * 100) / 100;\n};\n\nexport function isCloserTo(value: number, target1: number, target2: number): boolean {\n return Math.abs(value - target1) < Math.abs(value - target2);\n}\n\nexport type TestCase = {\n input: string;\n output: string;\n expectedResult: {\n score: number;\n reason?: string;\n };\n};\n\nexport type TestCaseWithContext = TestCase & {\n context: string[];\n};\n"]}
|
package/dist/chunk-YGTIO3J5.js
DELETED
|
@@ -1,17 +0,0 @@
|
|
|
1
|
-
import { Agent } from '@mastra/core/agent';
|
|
2
|
-
|
|
3
|
-
// src/metrics/judge/index.ts
|
|
4
|
-
var MastraAgentJudge = class {
|
|
5
|
-
agent;
|
|
6
|
-
constructor(name, instructions, model) {
|
|
7
|
-
this.agent = new Agent({
|
|
8
|
-
name: `Mastra Eval Judge ${name}`,
|
|
9
|
-
instructions,
|
|
10
|
-
model
|
|
11
|
-
});
|
|
12
|
-
}
|
|
13
|
-
};
|
|
14
|
-
|
|
15
|
-
export { MastraAgentJudge };
|
|
16
|
-
//# sourceMappingURL=chunk-YGTIO3J5.js.map
|
|
17
|
-
//# sourceMappingURL=chunk-YGTIO3J5.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/metrics/judge/index.ts"],"names":[],"mappings":";;;AAGO,IAAe,mBAAf,MAAgC;AAAA,EAClB,KAAA;AAAA,EAEnB,WAAA,CAAY,IAAA,EAAc,YAAA,EAAsB,KAAA,EAAsB;AACpE,IAAA,IAAA,CAAK,KAAA,GAAQ,IAAI,KAAA,CAAM;AAAA,MACrB,IAAA,EAAM,qBAAqB,IAAI,CAAA,CAAA;AAAA,MAC/B,YAAA;AAAA,MACA;AAAA,KACD,CAAA;AAAA,EACH;AACF","file":"chunk-YGTIO3J5.js","sourcesContent":["import { Agent } from '@mastra/core/agent';\nimport type { LanguageModel } from '@mastra/core/llm';\n\nexport abstract class MastraAgentJudge {\n protected readonly agent: Agent;\n\n constructor(name: string, instructions: string, model: LanguageModel) {\n this.agent = new Agent({\n name: `Mastra Eval Judge ${name}`,\n instructions: instructions,\n model,\n });\n }\n}\n"]}
|