@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
|
@@ -2,7 +2,7 @@ interface ContentSimilarityOptions {
|
|
|
2
2
|
ignoreCase?: boolean;
|
|
3
3
|
ignoreWhitespace?: boolean;
|
|
4
4
|
}
|
|
5
|
-
export declare function createContentSimilarityScorer({ ignoreCase, ignoreWhitespace }?: ContentSimilarityOptions): import("@mastra/core/
|
|
5
|
+
export declare function createContentSimilarityScorer({ ignoreCase, ignoreWhitespace }?: ContentSimilarityOptions): import("@mastra/core/evals").MastraScorer<"content-similarity-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
6
6
|
processedInput: string;
|
|
7
7
|
processedOutput: string;
|
|
8
8
|
}> & Record<"generateScoreStepResult", number>>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/content-similarity/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/content-similarity/index.ts"],"names":[],"mappings":"AAIA,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;;;gDAmC1G"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
export declare function createKeywordCoverageScorer(): import("@mastra/core/
|
|
1
|
+
export declare function createKeywordCoverageScorer(): import("@mastra/core/evals").MastraScorer<"keyword-coverage-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
2
2
|
result: {
|
|
3
3
|
referenceKeywords: Set<string>;
|
|
4
4
|
responseKeywords: Set<string>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/keyword-coverage/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/keyword-coverage/index.ts"],"names":[],"mappings":"AAIA,wBAAgB,2BAA2B;;;;;;;;;;;;;;gDAkE1C"}
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
export declare function createTextualDifferenceScorer(): import("@mastra/core/
|
|
1
|
+
export declare function createTextualDifferenceScorer(): import("@mastra/core/evals").MastraScorer<"textual-difference-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
2
2
|
ratio: number;
|
|
3
3
|
confidence: number;
|
|
4
4
|
changes: number;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/textual-difference/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/textual-difference/index.ts"],"names":[],"mappings":"AAoHA,wBAAgB,6BAA6B;;;;;gDAgC5C"}
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
interface ToneScorerConfig {
|
|
2
2
|
referenceTone?: string;
|
|
3
3
|
}
|
|
4
|
-
export declare function createToneScorer(config?: ToneScorerConfig): import("@mastra/core/
|
|
4
|
+
export declare function createToneScorer(config?: ToneScorerConfig): import("@mastra/core/evals").MastraScorer<"tone-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
5
5
|
score: number;
|
|
6
6
|
responseSentiment: number;
|
|
7
7
|
referenceSentiment: number;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/tone/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/tone/index.ts"],"names":[],"mappings":"AAIA,UAAU,gBAAgB;IACxB,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,wBAAgB,gBAAgB,CAAC,MAAM,GAAE,gBAAqB;;;;;;;;;;;;;;gDA6C7D"}
|
|
@@ -3,7 +3,7 @@ interface ToolCallAccuracyOptions {
|
|
|
3
3
|
strictMode?: boolean;
|
|
4
4
|
expectedToolOrder?: string[];
|
|
5
5
|
}
|
|
6
|
-
export declare function createToolCallAccuracyScorerCode(options: ToolCallAccuracyOptions): import("@mastra/core/
|
|
6
|
+
export declare function createToolCallAccuracyScorerCode(options: ToolCallAccuracyOptions): import("@mastra/core/evals").MastraScorer<"code-tool-call-accuracy-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
7
7
|
expectedTool: string | undefined;
|
|
8
8
|
actualTools: string[];
|
|
9
9
|
strictMode: boolean;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/tool-call-accuracy/index.ts"],"names":[],"mappings":"AAEA,UAAU,uBAAuB;IAC/B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;CAC9B;AAyDD,wBAAgB,gCAAgC,CAAC,OAAO,EAAE,uBAAuB;;;;;;;;;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/code/tool-call-accuracy/index.ts"],"names":[],"mappings":"AAEA,UAAU,uBAAuB;IAC/B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;CAC9B;AAyDD,wBAAgB,gCAAgC,CAAC,OAAO,EAAE,uBAAuB;;;;;;;;;gDA2DhF"}
|
|
@@ -4,7 +4,7 @@ export declare const ANSWER_RELEVANCY_AGENT_INSTRUCTIONS = "\n You are a bala
|
|
|
4
4
|
export declare function createAnswerRelevancyScorer({ model, options, }: {
|
|
5
5
|
model: MastraModelConfig;
|
|
6
6
|
options?: Record<'uncertaintyWeight' | 'scale', number>;
|
|
7
|
-
}): import("@mastra/core/
|
|
7
|
+
}): import("@mastra/core/evals").MastraScorer<"answer-relevancy-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
8
8
|
statements: string[];
|
|
9
9
|
}> & Record<"analyzeStepResult", {
|
|
10
10
|
results: {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/answer-relevancy/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/answer-relevancy/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAK1D,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,iBAAiB,CAAC;IACzB,OAAO,CAAC,EAAE,MAAM,CAAC,mBAAmB,GAAG,OAAO,EAAE,MAAM,CAAC,CAAC;CACzD;;;;;;;6FA2DA"}
|
|
@@ -13,14 +13,14 @@ export declare const ANSWER_SIMILARITY_INSTRUCTIONS = "\nYou are a precise answe
|
|
|
13
13
|
export declare function createAnswerSimilarityScorer({ model, options, }: {
|
|
14
14
|
model: MastraModelConfig;
|
|
15
15
|
options?: AnswerSimilarityOptions;
|
|
16
|
-
}): import("@mastra/core/
|
|
16
|
+
}): import("@mastra/core/evals").MastraScorer<"answer-similarity-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
17
17
|
outputUnits: string[];
|
|
18
18
|
groundTruthUnits: string[];
|
|
19
19
|
}> & Record<"analyzeStepResult", {
|
|
20
20
|
matches: {
|
|
21
21
|
groundTruthUnit: string;
|
|
22
22
|
outputUnit: string | null;
|
|
23
|
-
matchType: "exact" | "
|
|
23
|
+
matchType: "exact" | "semantic" | "partial" | "missing";
|
|
24
24
|
explanation: string;
|
|
25
25
|
}[];
|
|
26
26
|
extraInOutput: string[];
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/answer-similarity/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/answer-similarity/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAK1D,MAAM,WAAW,uBAAuB;IACtC,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,eAAO,MAAM,iCAAiC,EAAE,QAAQ,CAAC,uBAAuB,CAQ/E,CAAC;AAEF,eAAO,MAAM,8BAA8B,+kBAU1C,CAAC;AA0BF,wBAAgB,4BAA4B,CAAC,EAC3C,KAAK,EACL,OAA2C,GAC5C,EAAE;IACD,KAAK,EAAE,iBAAiB,CAAC;IACzB,OAAO,CAAC,EAAE,uBAAuB,CAAC;CACnC;;;;;;;;;;;;;;;;6FA8HA"}
|
|
@@ -1,12 +1,12 @@
|
|
|
1
|
+
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/evals';
|
|
1
2
|
import type { MastraModelConfig } from '@mastra/core/llm';
|
|
2
|
-
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
3
3
|
export interface BiasMetricOptions {
|
|
4
4
|
scale?: number;
|
|
5
5
|
}
|
|
6
6
|
export declare function createBiasScorer({ model, options }: {
|
|
7
7
|
model: MastraModelConfig;
|
|
8
8
|
options?: BiasMetricOptions;
|
|
9
|
-
}): import("@mastra/core/
|
|
9
|
+
}): import("@mastra/core/evals").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
10
10
|
opinions: string[];
|
|
11
11
|
}> & Record<"analyzeStepResult", {
|
|
12
12
|
results: {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/bias/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/bias/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,oBAAoB,CAAC;AAC1F,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAW1D,MAAM,WAAW,iBAAiB;IAChC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,wBAAgB,gBAAgB,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;IAAE,KAAK,EAAE,iBAAiB,CAAC;IAAC,OAAO,CAAC,EAAE,iBAAiB,CAAA;CAAE;;;;;;;6FAiD7G"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
+
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/evals';
|
|
1
2
|
import type { MastraModelConfig } from '@mastra/core/llm';
|
|
2
|
-
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
3
3
|
export interface ContextPrecisionMetricOptions {
|
|
4
4
|
scale?: number;
|
|
5
5
|
context?: string[];
|
|
@@ -8,10 +8,10 @@ export interface ContextPrecisionMetricOptions {
|
|
|
8
8
|
export declare function createContextPrecisionScorer({ model, options, }: {
|
|
9
9
|
model: MastraModelConfig;
|
|
10
10
|
options: ContextPrecisionMetricOptions;
|
|
11
|
-
}): import("@mastra/core/
|
|
11
|
+
}): import("@mastra/core/evals").MastraScorer<"context-precision-scorer", ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"analyzeStepResult", {
|
|
12
12
|
verdicts: {
|
|
13
|
-
verdict: string;
|
|
14
13
|
reason: string;
|
|
14
|
+
verdict: string;
|
|
15
15
|
context_index: number;
|
|
16
16
|
}[];
|
|
17
17
|
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string>>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/context-precision/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/context-precision/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,oBAAoB,CAAC;AAC1F,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAS1D,MAAM,WAAW,6BAA6B;IAC5C,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,gBAAgB,CAAC,EAAE,CAAC,KAAK,EAAE,sBAAsB,EAAE,MAAM,EAAE,uBAAuB,KAAK,MAAM,EAAE,CAAC;CACjG;AAYD,wBAAgB,4BAA4B,CAAC,EAC3C,KAAK,EACL,OAAO,GACR,EAAE;IACD,KAAK,EAAE,iBAAiB,CAAC;IACzB,OAAO,EAAE,6BAA6B,CAAC;CACxC;;;;;;6FAoGA"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
+
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/evals';
|
|
1
2
|
import type { MastraModelConfig } from '@mastra/core/llm';
|
|
2
|
-
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
3
3
|
export interface ContextRelevanceOptions {
|
|
4
4
|
scale?: number;
|
|
5
5
|
context?: string[];
|
|
@@ -13,12 +13,12 @@ export interface ContextRelevanceOptions {
|
|
|
13
13
|
export declare function createContextRelevanceScorerLLM({ model, options, }: {
|
|
14
14
|
model: MastraModelConfig;
|
|
15
15
|
options: ContextRelevanceOptions;
|
|
16
|
-
}): import("@mastra/core/
|
|
16
|
+
}): import("@mastra/core/evals").MastraScorer<"context-relevance-scorer", ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"analyzeStepResult", {
|
|
17
17
|
evaluations: {
|
|
18
18
|
reasoning: string;
|
|
19
19
|
context_index: number;
|
|
20
20
|
contextPiece: string;
|
|
21
|
-
relevanceLevel: "
|
|
21
|
+
relevanceLevel: "high" | "medium" | "low" | "none";
|
|
22
22
|
wasUsed: boolean;
|
|
23
23
|
}[];
|
|
24
24
|
overallAssessment: string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/context-relevance/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/context-relevance/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,oBAAoB,CAAC;AAE1F,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAK1D,MAAM,WAAW,uBAAuB;IACtC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,gBAAgB,CAAC,EAAE,CAAC,KAAK,EAAE,sBAAsB,EAAE,MAAM,EAAE,uBAAuB,KAAK,MAAM,EAAE,CAAC;IAChG,SAAS,CAAC,EAAE;QACV,0BAA0B,CAAC,EAAE,MAAM,CAAC;QACpC,qBAAqB,CAAC,EAAE,MAAM,CAAC;QAC/B,wBAAwB,CAAC,EAAE,MAAM,CAAC;KACnC,CAAC;CACH;AAuBD,wBAAgB,+BAA+B,CAAC,EAC9C,KAAK,EACL,OAAO,GACR,EAAE;IACD,KAAK,EAAE,iBAAiB,CAAC;IACzB,OAAO,EAAE,uBAAuB,CAAC;CAClC;;;;;;;;;;6FA8IA"}
|
|
@@ -6,10 +6,10 @@ export interface FaithfulnessMetricOptions {
|
|
|
6
6
|
export declare function createFaithfulnessScorer({ model, options, }: {
|
|
7
7
|
model: MastraModelConfig;
|
|
8
8
|
options?: FaithfulnessMetricOptions;
|
|
9
|
-
}): import("@mastra/core/
|
|
9
|
+
}): import("@mastra/core/evals").MastraScorer<"faithfulness-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"preprocessStepResult", string[]> & Record<"analyzeStepResult", {
|
|
10
10
|
verdicts: {
|
|
11
|
-
verdict: string;
|
|
12
11
|
reason: string;
|
|
12
|
+
verdict: string;
|
|
13
13
|
}[];
|
|
14
14
|
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string>>;
|
|
15
15
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/faithfulness/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/faithfulness/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAU1D,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,iBAAiB,CAAC;IACzB,OAAO,CAAC,EAAE,yBAAyB,CAAC;CACrC;;;;;6FAiEA"}
|
|
@@ -6,12 +6,12 @@ export interface HallucinationMetricOptions {
|
|
|
6
6
|
export declare function createHallucinationScorer({ model, options, }: {
|
|
7
7
|
model: MastraModelConfig;
|
|
8
8
|
options?: HallucinationMetricOptions;
|
|
9
|
-
}): import("@mastra/core/
|
|
9
|
+
}): import("@mastra/core/evals").MastraScorer<"hallucination-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
10
10
|
claims: string[];
|
|
11
11
|
}> & Record<"analyzeStepResult", {
|
|
12
12
|
verdicts: {
|
|
13
|
-
verdict: string;
|
|
14
13
|
reason: string;
|
|
14
|
+
verdict: string;
|
|
15
15
|
statement: string;
|
|
16
16
|
}[];
|
|
17
17
|
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string>>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/hallucination/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/hallucination/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAW1D,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,iBAAiB,CAAC;IACzB,OAAO,CAAC,EAAE,0BAA0B,CAAC;CACtC;;;;;;;;6FA4DA"}
|
|
@@ -21,7 +21,7 @@ export interface NoiseSensitivityOptions {
|
|
|
21
21
|
export declare function createNoiseSensitivityScorerLLM({ model, options, }: {
|
|
22
22
|
model: MastraModelConfig;
|
|
23
23
|
options: NoiseSensitivityOptions;
|
|
24
|
-
}): import("@mastra/core/
|
|
24
|
+
}): import("@mastra/core/evals").MastraScorer<"noise-sensitivity-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"analyzeStepResult", {
|
|
25
25
|
overallAssessment: string;
|
|
26
26
|
dimensions: {
|
|
27
27
|
dimension: string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/noise-sensitivity/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/noise-sensitivity/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAK1D,MAAM,WAAW,uBAAuB;IACtC,gBAAgB,EAAE,MAAM,CAAC;IACzB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE;QACR,aAAa,CAAC,EAAE;YACd,IAAI,CAAC,EAAE,MAAM,CAAC;YACd,OAAO,CAAC,EAAE,MAAM,CAAC;YACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;YAClB,WAAW,CAAC,EAAE,MAAM,CAAC;YACrB,MAAM,CAAC,EAAE,MAAM,CAAC;SACjB,CAAC;QACF,SAAS,CAAC,EAAE;YACV,iBAAiB,CAAC,EAAE,MAAM,CAAC;YAC3B,oBAAoB,CAAC,EAAE,MAAM,CAAC;SAC/B,CAAC;QACF,oBAAoB,CAAC,EAAE,MAAM,CAAC;KAC/B,CAAC;CACH;AA+BD,wBAAgB,+BAA+B,CAAC,EAC9C,KAAK,EACL,OAAO,GACR,EAAE;IACD,KAAK,EAAE,iBAAiB,CAAC;IACzB,OAAO,EAAE,uBAAuB,CAAC;CAClC;;;;;;;;;;6FAsHA"}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
+
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/evals';
|
|
1
2
|
import type { MastraModelConfig } from '@mastra/core/llm';
|
|
2
|
-
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
3
3
|
export interface PromptAlignmentOptions {
|
|
4
4
|
scale?: number;
|
|
5
5
|
evaluationMode?: 'user' | 'system' | 'both';
|
|
@@ -7,7 +7,7 @@ export interface PromptAlignmentOptions {
|
|
|
7
7
|
export declare function createPromptAlignmentScorerLLM({ model, options, }: {
|
|
8
8
|
model: MastraModelConfig;
|
|
9
9
|
options?: PromptAlignmentOptions;
|
|
10
|
-
}): import("@mastra/core/
|
|
10
|
+
}): import("@mastra/core/evals").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"analyzeStepResult", {
|
|
11
11
|
overallAssessment: string;
|
|
12
12
|
intentAlignment: {
|
|
13
13
|
reasoning: string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/prompt-alignment/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/prompt-alignment/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,oBAAoB,CAAC;AAE1F,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAU1D,MAAM,WAAW,sBAAsB;IACrC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,cAAc,CAAC,EAAE,MAAM,GAAG,QAAQ,GAAG,MAAM,CAAC;CAC7C;AAsDD,wBAAgB,8BAA8B,CAAC,EAC7C,KAAK,EACL,OAAO,GACR,EAAE;IACD,KAAK,EAAE,iBAAiB,CAAC;IACzB,OAAO,CAAC,EAAE,sBAAsB,CAAC;CAClC;;;;;;;;;;;;;;;;;;;;;;;;;;;6FAqHA"}
|
|
@@ -1,10 +1,10 @@
|
|
|
1
|
-
import type { Tool } from '@mastra/core';
|
|
2
1
|
import type { MastraModelConfig } from '@mastra/core/llm';
|
|
2
|
+
import type { Tool } from '@mastra/core/tools';
|
|
3
3
|
export interface ToolCallAccuracyOptions {
|
|
4
4
|
model: MastraModelConfig;
|
|
5
5
|
availableTools: Tool[];
|
|
6
6
|
}
|
|
7
|
-
export declare function createToolCallAccuracyScorerLLM({ model, availableTools }: ToolCallAccuracyOptions): import("@mastra/core/
|
|
7
|
+
export declare function createToolCallAccuracyScorerLLM({ model, availableTools }: ToolCallAccuracyOptions): import("@mastra/core/evals").MastraScorer<"llm-tool-call-accuracy-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
8
8
|
actualTools: string[];
|
|
9
9
|
hasToolCalls: boolean;
|
|
10
10
|
toolCallInfos: import("../../utils").ToolCallInfo[];
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/tool-call-accuracy/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/tool-call-accuracy/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAC1D,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,oBAAoB,CAAC;AAU/C,MAAM,WAAW,uBAAuB;IACtC,KAAK,EAAE,iBAAiB,CAAC;IACzB,cAAc,EAAE,IAAI,EAAE,CAAC;CACxB;AAaD,wBAAgB,+BAA+B,CAAC,EAAE,KAAK,EAAE,cAAc,EAAE,EAAE,uBAAuB;;;;;;;;;;;6FA4EjG"}
|
|
@@ -5,10 +5,10 @@ export interface ToxicityMetricOptions {
|
|
|
5
5
|
export declare function createToxicityScorer({ model, options, }: {
|
|
6
6
|
model: MastraModelConfig;
|
|
7
7
|
options?: ToxicityMetricOptions;
|
|
8
|
-
}): import("@mastra/core/
|
|
8
|
+
}): import("@mastra/core/evals").MastraScorer<"toxicity-scorer", import("@mastra/core/evals").ScorerRunInputForAgent, import("@mastra/core/evals").ScorerRunOutputForAgent, Record<"analyzeStepResult", {
|
|
9
9
|
verdicts: {
|
|
10
|
-
verdict: string;
|
|
11
10
|
reason: string;
|
|
11
|
+
verdict: string;
|
|
12
12
|
}[];
|
|
13
13
|
}> & Record<"generateScoreStepResult", number> & Record<"generateReasonStepResult", string>>;
|
|
14
14
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/toxicity/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/toxicity/index.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAK1D,MAAM,WAAW,qBAAqB;IACpC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,wBAAgB,oBAAoB,CAAC,EACnC,KAAK,EACL,OAAO,GACR,EAAE;IACD,KAAK,EAAE,iBAAiB,CAAC;IACzB,OAAO,CAAC,EAAE,qBAAqB,CAAC;CACjC;;;;;6FAgDA"}
|