@mastra/evals 0.14.0 → 0.14.1-alpha.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/CHANGELOG.md +18 -0
- package/dist/{dist-IEBPVOY3.cjs → dist-LDTK3TIP.cjs} +1207 -1199
- package/dist/dist-LDTK3TIP.cjs.map +1 -0
- package/dist/{dist-U76IQBNC.js → dist-OWYZEOJK.js} +1207 -1199
- package/dist/dist-OWYZEOJK.js.map +1 -0
- package/dist/index.cjs +1 -1
- package/dist/index.js +1 -1
- package/dist/{magic-string.es-WQRLTQPQ.js → magic-string.es-7ORA5OGR.js} +20 -5
- package/dist/magic-string.es-7ORA5OGR.js.map +1 -0
- package/dist/{magic-string.es-VZN2EYER.cjs → magic-string.es-NZ2XWFKN.cjs} +20 -5
- package/dist/magic-string.es-NZ2XWFKN.cjs.map +1 -0
- 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/context-precision/index.d.ts +2 -2
- package/dist/scorers/llm/context-precision/index.d.ts.map +1 -1
- package/dist/scorers/llm/context-relevance/index.d.ts +2 -2
- package/dist/scorers/llm/context-relevance/index.d.ts.map +1 -1
- package/dist/scorers/llm/index.cjs.map +1 -1
- package/dist/scorers/llm/index.js.map +1 -1
- package/dist/scorers/llm/noise-sensitivity/index.d.ts +2 -2
- 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/package.json +6 -6
- package/dist/dist-IEBPVOY3.cjs.map +0 -1
- package/dist/dist-U76IQBNC.js.map +0 -1
- package/dist/magic-string.es-VZN2EYER.cjs.map +0 -1
- package/dist/magic-string.es-WQRLTQPQ.js.map +0 -1
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import type {
|
|
1
|
+
import type { MastraModelConfig } from '@mastra/core/llm';
|
|
2
2
|
export interface NoiseSensitivityOptions {
|
|
3
3
|
baselineResponse: string;
|
|
4
4
|
noisyQuery: string;
|
|
@@ -19,7 +19,7 @@ export interface NoiseSensitivityOptions {
|
|
|
19
19
|
};
|
|
20
20
|
}
|
|
21
21
|
export declare function createNoiseSensitivityScorerLLM({ model, options, }: {
|
|
22
|
-
model:
|
|
22
|
+
model: MastraModelConfig;
|
|
23
23
|
options: NoiseSensitivityOptions;
|
|
24
24
|
}): import("@mastra/core/scores").MastraScorer<"Noise Sensitivity (LLM)", import("@mastra/core/scores").ScorerRunInputForAgent, import("@mastra/core/scores").ScorerRunOutputForAgent, Record<"analyzeStepResult", {
|
|
25
25
|
overallAssessment: string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/noise-sensitivity/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/noise-sensitivity/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAM1D,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;;;;;;;;;;6FAqHA"}
|
|
@@ -1,11 +1,11 @@
|
|
|
1
|
-
import type {
|
|
1
|
+
import type { MastraModelConfig } from '@mastra/core/llm';
|
|
2
2
|
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
3
3
|
export interface PromptAlignmentOptions {
|
|
4
4
|
scale?: number;
|
|
5
5
|
evaluationMode?: 'user' | 'system' | 'both';
|
|
6
6
|
}
|
|
7
7
|
export declare function createPromptAlignmentScorerLLM({ model, options, }: {
|
|
8
|
-
model:
|
|
8
|
+
model: MastraModelConfig;
|
|
9
9
|
options?: PromptAlignmentOptions;
|
|
10
10
|
}): import("@mastra/core/scores").MastraScorer<string, ScorerRunInputForAgent, ScorerRunOutputForAgent, Record<"analyzeStepResult", {
|
|
11
11
|
overallAssessment: 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,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAC1D,OAAO,KAAK,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAW3F,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;;;;;;;;;;;;;;;;;;;;;;;;;;;6FAoHA"}
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import type { Tool } from '@mastra/core';
|
|
2
|
-
import type {
|
|
2
|
+
import type { MastraModelConfig } from '@mastra/core/llm';
|
|
3
3
|
export interface ToolCallAccuracyOptions {
|
|
4
|
-
model:
|
|
4
|
+
model: MastraModelConfig;
|
|
5
5
|
availableTools: Tool[];
|
|
6
6
|
}
|
|
7
7
|
export declare function createToolCallAccuracyScorerLLM({ model, availableTools }: ToolCallAccuracyOptions): import("@mastra/core/scores").MastraScorer<"Tool Call Accuracy (LLM)", import("@mastra/core/scores").ScorerRunInputForAgent, import("@mastra/core/scores").ScorerRunOutputForAgent, Record<"preprocessStepResult", {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/tool-call-accuracy/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,cAAc,CAAC;AACzC,OAAO,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/tool-call-accuracy/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,cAAc,CAAC;AACzC,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAW1D,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;;;;;;;;;;;6FA2EjG"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@mastra/evals",
|
|
3
|
-
"version": "0.14.
|
|
3
|
+
"version": "0.14.1-alpha.1",
|
|
4
4
|
"description": "",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"files": [
|
|
@@ -103,7 +103,7 @@
|
|
|
103
103
|
"string-similarity": "^4.0.4"
|
|
104
104
|
},
|
|
105
105
|
"peerDependencies": {
|
|
106
|
-
"@mastra/core": ">=0.21.0-0 <0.
|
|
106
|
+
"@mastra/core": ">=0.21.0-0 <0.23.0-0",
|
|
107
107
|
"ai": "^4.0.0 || ^5.0.0",
|
|
108
108
|
"zod": "^3.25.0 || ^4.0.0"
|
|
109
109
|
},
|
|
@@ -116,14 +116,14 @@
|
|
|
116
116
|
"@types/string-similarity": "^4.0.2",
|
|
117
117
|
"ai": "^4.3.19",
|
|
118
118
|
"dotenv": "^17.0.0",
|
|
119
|
-
"eslint": "^9.
|
|
119
|
+
"eslint": "^9.37.0",
|
|
120
120
|
"tsup": "^8.5.0",
|
|
121
121
|
"typescript": "^5.8.3",
|
|
122
122
|
"vitest": "^3.2.4",
|
|
123
123
|
"zod": "^3.25.76",
|
|
124
|
-
"@internal/lint": "0.0.
|
|
125
|
-
"@
|
|
126
|
-
"@
|
|
124
|
+
"@internal/lint": "0.0.50",
|
|
125
|
+
"@mastra/core": "0.22.0-alpha.1",
|
|
126
|
+
"@internal/types-builder": "0.0.25"
|
|
127
127
|
},
|
|
128
128
|
"scripts": {
|
|
129
129
|
"check": "tsc --noEmit",
|