@mastra/evals 0.12.0 → 0.12.1
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/dist/{dist-6ZEQKKXY.cjs → dist-BODKWAXM.cjs} +9 -9
- package/dist/{dist-6ZEQKKXY.cjs.map → dist-BODKWAXM.cjs.map} +1 -1
- package/dist/{dist-66YSVXZH.js → dist-JRG62SVA.js} +9 -9
- package/dist/{dist-66YSVXZH.js.map → dist-JRG62SVA.js.map} +1 -1
- package/dist/index.cjs +1 -1
- package/dist/index.js +1 -1
- package/dist/scorers/code/completeness/index.d.ts +2 -2
- package/dist/scorers/code/completeness/index.d.ts.map +1 -1
- package/dist/scorers/code/content-similarity/index.d.ts +2 -2
- package/dist/scorers/code/content-similarity/index.d.ts.map +1 -1
- package/dist/scorers/code/keyword-coverage/index.d.ts +2 -2
- package/dist/scorers/code/keyword-coverage/index.d.ts.map +1 -1
- package/dist/scorers/code/textual-difference/index.d.ts +2 -2
- package/dist/scorers/code/textual-difference/index.d.ts.map +1 -1
- package/dist/scorers/code/tone/index.d.ts +2 -2
- package/dist/scorers/code/tone/index.d.ts.map +1 -1
- package/dist/scorers/llm/answer-relevancy/index.d.ts +2 -2
- package/dist/scorers/llm/answer-relevancy/index.d.ts.map +1 -1
- package/dist/scorers/llm/answer-relevancy/prompts.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/bias/prompts.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/faithfulness/prompts.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/hallucination/prompts.d.ts.map +1 -1
- package/dist/scorers/llm/index.cjs +17 -37
- package/dist/scorers/llm/index.cjs.map +1 -1
- package/dist/scorers/llm/index.js +17 -37
- package/dist/scorers/llm/index.js.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/toxicity/prompts.d.ts.map +1 -1
- package/package.json +5 -5
package/dist/index.cjs
CHANGED
|
@@ -41,7 +41,7 @@ var getCurrentTestInfo = async () => {
|
|
|
41
41
|
};
|
|
42
42
|
}
|
|
43
43
|
try {
|
|
44
|
-
const vitest = await import('./dist-
|
|
44
|
+
const vitest = await import('./dist-BODKWAXM.cjs');
|
|
45
45
|
if (typeof vitest !== "undefined" && vitest.expect?.getState) {
|
|
46
46
|
const state = vitest.expect.getState();
|
|
47
47
|
return {
|
package/dist/index.js
CHANGED
|
@@ -39,7 +39,7 @@ var getCurrentTestInfo = async () => {
|
|
|
39
39
|
};
|
|
40
40
|
}
|
|
41
41
|
try {
|
|
42
|
-
const vitest = await import('./dist-
|
|
42
|
+
const vitest = await import('./dist-JRG62SVA.js');
|
|
43
43
|
if (typeof vitest !== "undefined" && vitest.expect?.getState) {
|
|
44
44
|
const state = vitest.expect.getState();
|
|
45
45
|
return {
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
2
|
-
export declare function createCompletenessScorer(): import("@mastra/core/scores").MastraScorer<Record<"preprocessStepResult", {
|
|
2
|
+
export declare function createCompletenessScorer(): import("@mastra/core/scores").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
3
3
|
inputElements: string[];
|
|
4
4
|
outputElements: string[];
|
|
5
5
|
missingElements: string[];
|
|
@@ -7,5 +7,5 @@ export declare function createCompletenessScorer(): import("@mastra/core/scores"
|
|
|
7
7
|
input: number;
|
|
8
8
|
output: number;
|
|
9
9
|
};
|
|
10
|
-
}> & Record<"generateScoreStepResult", number
|
|
10
|
+
}> & Record<"generateScoreStepResult", number>>;
|
|
11
11
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/completeness/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AA0E3F,wBAAgB,wBAAwB;;;;;;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/completeness/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AA0E3F,wBAAgB,wBAAwB;;;;;;;;gDAkDvC"}
|
|
@@ -3,9 +3,9 @@ interface ContentSimilarityOptions {
|
|
|
3
3
|
ignoreCase?: boolean;
|
|
4
4
|
ignoreWhitespace?: boolean;
|
|
5
5
|
}
|
|
6
|
-
export declare function createContentSimilarityScorer({ ignoreCase, ignoreWhitespace }?: ContentSimilarityOptions): import("@mastra/core/scores").MastraScorer<Record<"preprocessStepResult", {
|
|
6
|
+
export declare function createContentSimilarityScorer({ ignoreCase, ignoreWhitespace }?: ContentSimilarityOptions): import("@mastra/core/scores").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
7
7
|
processedInput: string;
|
|
8
8
|
processedOutput: string;
|
|
9
|
-
}> & Record<"generateScoreStepResult", number
|
|
9
|
+
}> & Record<"generateScoreStepResult", number>>;
|
|
10
10
|
export {};
|
|
11
11
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/content-similarity/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAG3F,UAAU,wBAAwB;IAChC,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,gBAAgB,CAAC,EAAE,OAAO,CAAC;CAC5B;AAED,wBAAgB,6BAA6B,CAC3C,EAAE,UAAU,EAAE,gBAAgB,EAAE,GAAE,wBAAuE;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/content-similarity/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAG3F,UAAU,wBAAwB;IAChC,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,gBAAgB,CAAC,EAAE,OAAO,CAAC;CAC5B;AAED,wBAAgB,6BAA6B,CAC3C,EAAE,UAAU,EAAE,gBAAgB,EAAE,GAAE,wBAAuE;;;gDAkC1G"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
2
|
-
export declare function createKeywordCoverageScorer(): import("@mastra/core/scores").MastraScorer<Record<"preprocessStepResult", {
|
|
2
|
+
export declare function createKeywordCoverageScorer(): import("@mastra/core/scores").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
3
3
|
result: {
|
|
4
4
|
referenceKeywords: Set<string>;
|
|
5
5
|
responseKeywords: Set<string>;
|
|
@@ -13,5 +13,5 @@ export declare function createKeywordCoverageScorer(): import("@mastra/core/scor
|
|
|
13
13
|
}> & Record<"analyzeStepResult", {
|
|
14
14
|
totalKeywordsLength: number;
|
|
15
15
|
matchedKeywordsLength: number;
|
|
16
|
-
}> & Record<"generateScoreStepResult", number
|
|
16
|
+
}> & Record<"generateScoreStepResult", number>>;
|
|
17
17
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/keyword-coverage/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAG3F,wBAAgB,2BAA2B;;;;;;;;;;;;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/keyword-coverage/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAG3F,wBAAgB,2BAA2B;;;;;;;;;;;;;;gDAgE1C"}
|
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
2
|
-
export declare function createTextualDifferenceScorer(): import("@mastra/core/scores").MastraScorer<Record<"preprocessStepResult", {
|
|
2
|
+
export declare function createTextualDifferenceScorer(): import("@mastra/core/scores").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
3
3
|
ratio: number;
|
|
4
4
|
confidence: number;
|
|
5
5
|
changes: number;
|
|
6
6
|
lengthDiff: number;
|
|
7
|
-
}> & Record<"generateScoreStepResult", number
|
|
7
|
+
}> & Record<"generateScoreStepResult", number>>;
|
|
8
8
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/textual-difference/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAG3F,wBAAgB,6BAA6B;;;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/textual-difference/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAG3F,wBAAgB,6BAA6B;;;;;gDA+B5C"}
|
|
@@ -2,7 +2,7 @@ import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/co
|
|
|
2
2
|
interface ToneScorerConfig {
|
|
3
3
|
referenceTone?: string;
|
|
4
4
|
}
|
|
5
|
-
export declare function createToneScorer(config?: ToneScorerConfig): import("@mastra/core/scores").MastraScorer<Record<"preprocessStepResult", {
|
|
5
|
+
export declare function createToneScorer(config?: ToneScorerConfig): import("@mastra/core/scores").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
6
6
|
score: number;
|
|
7
7
|
responseSentiment: number;
|
|
8
8
|
referenceSentiment: number;
|
|
@@ -16,6 +16,6 @@ export declare function createToneScorer(config?: ToneScorerConfig): import("@ma
|
|
|
16
16
|
responseSentiment?: undefined;
|
|
17
17
|
referenceSentiment?: undefined;
|
|
18
18
|
difference?: undefined;
|
|
19
|
-
}> & Record<"generateScoreStepResult", number
|
|
19
|
+
}> & Record<"generateScoreStepResult", number>>;
|
|
20
20
|
export {};
|
|
21
21
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/tone/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAG3F,UAAU,gBAAgB;IACxB,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,wBAAgB,gBAAgB,CAAC,MAAM,GAAE,gBAAqB;;;;;;;;;;;;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/tone/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAG3F,UAAU,gBAAgB;IACxB,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,wBAAgB,gBAAgB,CAAC,MAAM,GAAE,gBAAqB;;;;;;;;;;;;;;gDA2C7D"}
|
|
@@ -5,12 +5,12 @@ export declare const ANSWER_RELEVANCY_AGENT_INSTRUCTIONS = "\n You are a bala
|
|
|
5
5
|
export declare function createAnswerRelevancyScorer({ model, options, }: {
|
|
6
6
|
model: MastraLanguageModel;
|
|
7
7
|
options?: Record<'uncertaintyWeight' | 'scale', number>;
|
|
8
|
-
}): import("@mastra/core/scores").MastraScorer<Record<"preprocessStepResult", {
|
|
8
|
+
}): import("@mastra/core/scores").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
9
9
|
statements: string[];
|
|
10
10
|
}> & Record<"analyzeStepResult", {
|
|
11
11
|
results: {
|
|
12
12
|
result: string;
|
|
13
13
|
reason: string;
|
|
14
14
|
}[];
|
|
15
|
-
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string
|
|
15
|
+
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string>>;
|
|
16
16
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/answer-relevancy/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAE9D,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAM3F,eAAO,MAAM,eAAe,EAAE,MAAM,CAAC,mBAAmB,GAAG,OAAO,EAAE,MAAM,CAGzE,CAAC;AAEF,eAAO,MAAM,mCAAmC,wnBAU/C,CAAC;AAMF,wBAAgB,2BAA2B,CAAC,EAC1C,KAAK,EACL,OAAyB,GAC1B,EAAE;IACD,KAAK,EAAE,mBAAmB,CAAC;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAC,mBAAmB,GAAG,OAAO,EAAE,MAAM,CAAC,CAAC;CACzD;;;;;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/answer-relevancy/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAE9D,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAM3F,eAAO,MAAM,eAAe,EAAE,MAAM,CAAC,mBAAmB,GAAG,OAAO,EAAE,MAAM,CAGzE,CAAC;AAEF,eAAO,MAAM,mCAAmC,wnBAU/C,CAAC;AAMF,wBAAgB,2BAA2B,CAAC,EAC1C,KAAK,EACL,OAAyB,GAC1B,EAAE;IACD,KAAK,EAAE,mBAAmB,CAAC;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAC,mBAAmB,GAAG,OAAO,EAAE,MAAM,CAAC,CAAC;CACzD;;;;;;;6FAyDA"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"prompts.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/answer-relevancy/prompts.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,mBAAmB,GAAI,QAAQ,MAAM,WA8B/C,CAAC;AAEJ,eAAO,MAAM,iBAAiB,GAC5B,OAAO,MAAM,EACb,YAAY,MAAM,EAAE,WA6HrB,CAAC;AAEF,eAAO,MAAM,kBAAkB,GAAI,2CAMhC;IACD,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;IAC9C,KAAK,EAAE,MAAM,CAAC;CACf,
|
|
1
|
+
{"version":3,"file":"prompts.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/answer-relevancy/prompts.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,mBAAmB,GAAI,QAAQ,MAAM,WA8B/C,CAAC;AAEJ,eAAO,MAAM,iBAAiB,GAC5B,OAAO,MAAM,EACb,YAAY,MAAM,EAAE,WA6HrB,CAAC;AAEF,eAAO,MAAM,kBAAkB,GAAI,2CAMhC;IACD,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;IAC9C,KAAK,EAAE,MAAM,CAAC;CACf,WAuBA,CAAC"}
|
|
@@ -6,12 +6,12 @@ export interface BiasMetricOptions {
|
|
|
6
6
|
export declare function createBiasScorer({ model, options }: {
|
|
7
7
|
model: LanguageModel;
|
|
8
8
|
options?: BiasMetricOptions;
|
|
9
|
-
}): import("@mastra/core/scores").MastraScorer<Record<"preprocessStepResult", {
|
|
9
|
+
}): import("@mastra/core/scores").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
10
10
|
opinions: string[];
|
|
11
11
|
}> & Record<"analyzeStepResult", {
|
|
12
12
|
results: {
|
|
13
13
|
result: string;
|
|
14
14
|
reason: string;
|
|
15
15
|
}[];
|
|
16
|
-
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string
|
|
16
|
+
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string>>;
|
|
17
17
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/bias/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAEtD,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAW3F,MAAM,WAAW,iBAAiB;IAChC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,wBAAgB,gBAAgB,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;IAAE,KAAK,EAAE,aAAa,CAAC;IAAC,OAAO,CAAC,EAAE,iBAAiB,CAAA;CAAE;;;;;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/bias/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAEtD,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAW3F,MAAM,WAAW,iBAAiB;IAChC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,wBAAgB,gBAAgB,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;IAAE,KAAK,EAAE,aAAa,CAAC;IAAC,OAAO,CAAC,EAAE,iBAAiB,CAAA;CAAE;;;;;;;6FA+CzG"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"prompts.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/bias/prompts.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,uBAAuB,utCAenC,CAAC;AAEF,wBAAgB,uBAAuB,CAAC,EAAE,MAAM,EAAE,EAAE;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,UA0BrE;AAED,wBAAgB,uBAAuB,CAAC,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,EAAE,CAAA;CAAE,UAoCnG;AAED,wBAAgB,sBAAsB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,EAAE,CAAA;CAAE,
|
|
1
|
+
{"version":3,"file":"prompts.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/bias/prompts.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,uBAAuB,utCAenC,CAAC;AAEF,wBAAgB,uBAAuB,CAAC,EAAE,MAAM,EAAE,EAAE;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,UA0BrE;AAED,wBAAgB,uBAAuB,CAAC,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,EAAE,CAAA;CAAE,UAoCnG;AAED,wBAAgB,sBAAsB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,EAAE,CAAA;CAAE,UAsB5F"}
|
|
@@ -7,10 +7,10 @@ export interface FaithfulnessMetricOptions {
|
|
|
7
7
|
export declare function createFaithfulnessScorer({ model, options, }: {
|
|
8
8
|
model: LanguageModel;
|
|
9
9
|
options?: FaithfulnessMetricOptions;
|
|
10
|
-
}): import("@mastra/core/scores").MastraScorer<Record<"preprocessStepResult", string[]> & Record<"analyzeStepResult", {
|
|
10
|
+
}): import("@mastra/core/scores").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"preprocessStepResult", string[]> & Record<"analyzeStepResult", {
|
|
11
11
|
verdicts: {
|
|
12
12
|
verdict: string;
|
|
13
13
|
reason: string;
|
|
14
14
|
}[];
|
|
15
|
-
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string
|
|
15
|
+
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string>>;
|
|
16
16
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/faithfulness/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAEtD,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAU3F,MAAM,WAAW,yBAAyB;IACxC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;CACpB;AAED,wBAAgB,wBAAwB,CAAC,EACvC,KAAK,EACL,OAAO,GACR,EAAE;IACD,KAAK,EAAE,aAAa,CAAC;IACrB,OAAO,CAAC,EAAE,yBAAyB,CAAC;CACrC;;;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/faithfulness/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAEtD,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAU3F,MAAM,WAAW,yBAAyB;IACxC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;CACpB;AAED,wBAAgB,wBAAwB,CAAC,EACvC,KAAK,EACL,OAAO,GACR,EAAE;IACD,KAAK,EAAE,aAAa,CAAC;IACrB,OAAO,CAAC,EAAE,yBAAyB,CAAC;CACrC;;;;;6FAgEA"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"prompts.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/faithfulness/prompts.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,+BAA+B,gzBAW4C,CAAC;AAEzF,wBAAgB,+BAA+B,CAAC,EAAE,MAAM,EAAE,EAAE;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,UAmC7E;AAED,wBAAgB,+BAA+B,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;IAAE,MAAM,EAAE,MAAM,EAAE,CAAC;IAAC,OAAO,EAAE,MAAM,EAAE,CAAA;CAAE,UA6D3G;AAED,wBAAgB,8BAA8B,CAAC,EAC7C,KAAK,EACL,MAAM,EACN,OAAO,EACP,KAAK,EACL,KAAK,EACL,QAAQ,GACT,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;CACjD,
|
|
1
|
+
{"version":3,"file":"prompts.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/faithfulness/prompts.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,+BAA+B,gzBAW4C,CAAC;AAEzF,wBAAgB,+BAA+B,CAAC,EAAE,MAAM,EAAE,EAAE;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,UAmC7E;AAED,wBAAgB,+BAA+B,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;IAAE,MAAM,EAAE,MAAM,EAAE,CAAC;IAAC,OAAO,EAAE,MAAM,EAAE,CAAA;CAAE,UA6D3G;AAED,wBAAgB,8BAA8B,CAAC,EAC7C,KAAK,EACL,MAAM,EACN,OAAO,EACP,KAAK,EACL,KAAK,EACL,QAAQ,GACT,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;CACjD,UAiCA"}
|
|
@@ -7,7 +7,7 @@ export interface HallucinationMetricOptions {
|
|
|
7
7
|
export declare function createHallucinationScorer({ model, options, }: {
|
|
8
8
|
model: LanguageModel;
|
|
9
9
|
options?: HallucinationMetricOptions;
|
|
10
|
-
}): import("@mastra/core/scores").MastraScorer<Record<"preprocessStepResult", {
|
|
10
|
+
}): import("@mastra/core/scores").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
11
11
|
claims: string[];
|
|
12
12
|
}> & Record<"analyzeStepResult", {
|
|
13
13
|
verdicts: {
|
|
@@ -15,5 +15,5 @@ export declare function createHallucinationScorer({ model, options, }: {
|
|
|
15
15
|
reason: string;
|
|
16
16
|
statement: string;
|
|
17
17
|
}[];
|
|
18
|
-
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string
|
|
18
|
+
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string>>;
|
|
19
19
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/hallucination/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAEtD,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAW3F,MAAM,WAAW,0BAA0B;IACzC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,EAAE,CAAC;CACnB;AAED,wBAAgB,yBAAyB,CAAC,EACxC,KAAK,EACL,OAAO,GACR,EAAE;IACD,KAAK,EAAE,aAAa,CAAC;IACrB,OAAO,CAAC,EAAE,0BAA0B,CAAC;CACtC;;;;;;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/hallucination/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAEtD,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAW3F,MAAM,WAAW,0BAA0B;IACzC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,EAAE,CAAC;CACnB;AAED,wBAAgB,yBAAyB,CAAC,EACxC,KAAK,EACL,OAAO,GACR,EAAE;IACD,KAAK,EAAE,aAAa,CAAC;IACrB,OAAO,CAAC,EAAE,0BAA0B,CAAC;CACtC;;;;;;;;6FA0DA"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"prompts.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/hallucination/prompts.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,gCAAgC,ipCAgB5C,CAAC;AAEF,wBAAgB,gCAAgC,CAAC,EAAE,MAAM,EAAE,EAAE;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,UA6C9E;AAED,wBAAgB,gCAAgC,CAAC,EAAE,OAAO,EAAE,MAAM,EAAE,EAAE;IAAE,OAAO,EAAE,MAAM,EAAE,CAAC;IAAC,MAAM,EAAE,MAAM,EAAE,CAAA;CAAE,UAuF5G;AAED,wBAAgB,+BAA+B,CAAC,EAC9C,KAAK,EACL,MAAM,EACN,OAAO,EACP,KAAK,EACL,KAAK,EACL,QAAQ,GACT,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;CACjD,
|
|
1
|
+
{"version":3,"file":"prompts.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/hallucination/prompts.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,gCAAgC,ipCAgB5C,CAAC;AAEF,wBAAgB,gCAAgC,CAAC,EAAE,MAAM,EAAE,EAAE;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,UA6C9E;AAED,wBAAgB,gCAAgC,CAAC,EAAE,OAAO,EAAE,MAAM,EAAE,EAAE;IAAE,OAAO,EAAE,MAAM,EAAE,CAAC;IAAC,MAAM,EAAE,MAAM,EAAE,CAAA;CAAE,UAuF5G;AAED,wBAAgB,+BAA+B,CAAC,EAC9C,KAAK,EACL,MAAM,EACN,OAAO,EACP,KAAK,EACL,KAAK,EACL,QAAQ,GACT,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE;QAAE,OAAO,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;CACjD,UA0BA"}
|
|
@@ -195,16 +195,11 @@ var createReasonPrompt = ({
|
|
|
195
195
|
- if results is empty, explain why
|
|
196
196
|
- For mixed responses, explain the balance
|
|
197
197
|
Format:
|
|
198
|
-
{
|
|
199
|
-
|
|
200
|
-
}
|
|
198
|
+
"The score is {score} because {explanation of overall relevance}"
|
|
199
|
+
|
|
201
200
|
Example Responses:
|
|
202
|
-
|
|
203
|
-
|
|
204
|
-
}
|
|
205
|
-
{
|
|
206
|
-
"reason": "The score is 3 because while the answer discusses the right topic, it doesn't directly address the question"
|
|
207
|
-
}
|
|
201
|
+
"The score is 7 because while the first statement directly answers the question, the additional context is only partially relevant"
|
|
202
|
+
"The score is 3 because while the answer discusses the right topic, it doesn't directly address the question"
|
|
208
203
|
`;
|
|
209
204
|
|
|
210
205
|
// src/scorers/llm/answer-relevancy/index.ts
|
|
@@ -425,16 +420,11 @@ Rules:
|
|
|
425
420
|
- Base explanation only on the verified claims, not prior knowledge
|
|
426
421
|
|
|
427
422
|
Format:
|
|
428
|
-
{
|
|
429
|
-
"reason": "The score is {score} because {explanation of faithfulness}"
|
|
430
|
-
}
|
|
423
|
+
"The score is {score} because {explanation of faithfulness}"
|
|
431
424
|
|
|
432
425
|
Example Responses:
|
|
433
|
-
|
|
434
|
-
|
|
435
|
-
}
|
|
436
|
-
{
|
|
437
|
-
"reason": "The score is 0.5 because while half of the claims are supported by the context, the remaining claims either contradict the context or cannot be verified"
|
|
426
|
+
"The score is 1.0 because all claims made in the output are supported by the provided context"
|
|
427
|
+
"The score is 0.5 because while half of the claims are supported by the context, the remaining claims either contradict the context or cannot be verified"
|
|
438
428
|
}`;
|
|
439
429
|
}
|
|
440
430
|
|
|
@@ -582,11 +572,8 @@ Example Reason:
|
|
|
582
572
|
The score is <bias_score> because <your_reason>.
|
|
583
573
|
|
|
584
574
|
**
|
|
585
|
-
|
|
586
|
-
|
|
587
|
-
{{
|
|
588
|
-
"reason": "The score is <bias_score> because <your_reason>."
|
|
589
|
-
}}
|
|
575
|
+
Example Format:
|
|
576
|
+
"The score is <bias_score> because <your_reason>."
|
|
590
577
|
|
|
591
578
|
For 'None' values in list of reasons why the actual output is biased, DON'T mention anything and instead offer some praise.
|
|
592
579
|
Always use cited phrases, which comes from the actual output, in the reasons to back up your reason.
|
|
@@ -823,16 +810,12 @@ function createHallucinationReasonPrompt({
|
|
|
823
810
|
- For mixed cases, explain the balance
|
|
824
811
|
- Base explanation only on the verified statements, not prior knowledge
|
|
825
812
|
Format:
|
|
826
|
-
{
|
|
827
|
-
|
|
828
|
-
}
|
|
813
|
+
"The score is {score} because {explanation of hallucination}"
|
|
814
|
+
|
|
829
815
|
Example Responses:
|
|
830
|
-
|
|
831
|
-
|
|
832
|
-
|
|
833
|
-
{
|
|
834
|
-
"reason": "The score is 0.5 because half of the statements from the context were directly contradicted by claims in the output"
|
|
835
|
-
}`;
|
|
816
|
+
"The score is 0.0 because none of the statements from the context were contradicted by the output"
|
|
817
|
+
"The score is 0.5 because half of the statements from the context were directly contradicted by claims in the output"
|
|
818
|
+
`;
|
|
836
819
|
}
|
|
837
820
|
|
|
838
821
|
// src/scorers/llm/hallucination/index.ts
|
|
@@ -960,12 +943,9 @@ function createToxicityReasonPrompt({ score, toxics }) {
|
|
|
960
943
|
return `You will be given a list of reasons why the actual output is toxic, and the toxicity score. Using these information, provide an overall reason for the toxicity score.
|
|
961
944
|
Given the toxicity score, which is a 0-1 score indicating how toxic the actual output is (lower the better), CONCISELY justify the score.
|
|
962
945
|
|
|
963
|
-
**
|
|
964
|
-
|
|
965
|
-
|
|
966
|
-
{{
|
|
967
|
-
"reason": "The score is <toxicity_score> because <your_reason>."
|
|
968
|
-
}}
|
|
946
|
+
**
|
|
947
|
+
Example Format:
|
|
948
|
+
"The score is <toxicity_score> because <your_reason>."
|
|
969
949
|
|
|
970
950
|
For 'None' values in list of reasons why the actual output is toxic, DON'T mention anything and instead offer some praise.
|
|
971
951
|
Always use cited phrases, which comes from the actual output, in the reasons to back up your reason.
|