@mastra/evals 0.13.7 → 0.13.8-alpha.2
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 +24 -0
- package/dist/scorers/code/completeness/index.d.ts +1 -2
- package/dist/scorers/code/completeness/index.d.ts.map +1 -1
- package/dist/scorers/code/content-similarity/index.d.ts +1 -2
- package/dist/scorers/code/content-similarity/index.d.ts.map +1 -1
- package/dist/scorers/code/index.cjs +18 -12
- package/dist/scorers/code/index.cjs.map +1 -1
- package/dist/scorers/code/index.js +18 -12
- package/dist/scorers/code/index.js.map +1 -1
- package/dist/scorers/code/keyword-coverage/index.d.ts +1 -2
- package/dist/scorers/code/keyword-coverage/index.d.ts.map +1 -1
- package/dist/scorers/code/textual-difference/index.d.ts +1 -2
- package/dist/scorers/code/textual-difference/index.d.ts.map +1 -1
- package/dist/scorers/code/tone/index.d.ts +1 -2
- package/dist/scorers/code/tone/index.d.ts.map +1 -1
- package/dist/scorers/code/tool-call-accuracy/index.d.ts +1 -2
- package/dist/scorers/code/tool-call-accuracy/index.d.ts.map +1 -1
- package/dist/scorers/llm/answer-relevancy/index.d.ts +1 -2
- package/dist/scorers/llm/answer-relevancy/index.d.ts.map +1 -1
- package/dist/scorers/llm/answer-similarity/index.d.ts +2 -3
- package/dist/scorers/llm/answer-similarity/index.d.ts.map +1 -1
- package/dist/scorers/llm/bias/index.d.ts.map +1 -1
- package/dist/scorers/llm/context-precision/index.d.ts +1 -1
- package/dist/scorers/llm/context-precision/index.d.ts.map +1 -1
- package/dist/scorers/llm/context-relevance/index.d.ts +1 -1
- package/dist/scorers/llm/context-relevance/index.d.ts.map +1 -1
- package/dist/scorers/llm/faithfulness/index.d.ts +1 -2
- package/dist/scorers/llm/faithfulness/index.d.ts.map +1 -1
- package/dist/scorers/llm/hallucination/index.d.ts +1 -2
- package/dist/scorers/llm/hallucination/index.d.ts.map +1 -1
- package/dist/scorers/llm/index.cjs +20 -10
- package/dist/scorers/llm/index.cjs.map +1 -1
- package/dist/scorers/llm/index.js +20 -10
- package/dist/scorers/llm/index.js.map +1 -1
- package/dist/scorers/llm/noise-sensitivity/index.d.ts +1 -2
- package/dist/scorers/llm/noise-sensitivity/index.d.ts.map +1 -1
- package/dist/scorers/llm/tool-call-accuracy/index.d.ts +1 -2
- package/dist/scorers/llm/tool-call-accuracy/index.d.ts.map +1 -1
- package/dist/scorers/llm/toxicity/index.d.ts +1 -2
- package/dist/scorers/llm/toxicity/index.d.ts.map +1 -1
- package/package.json +6 -6
|
@@ -1,5 +1,4 @@
|
|
|
1
1
|
import type { MastraLanguageModel } from '@mastra/core/agent';
|
|
2
|
-
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
3
2
|
export interface NoiseSensitivityOptions {
|
|
4
3
|
baselineResponse: string;
|
|
5
4
|
noisyQuery: string;
|
|
@@ -22,7 +21,7 @@ export interface NoiseSensitivityOptions {
|
|
|
22
21
|
export declare function createNoiseSensitivityScorerLLM({ model, options, }: {
|
|
23
22
|
model: MastraLanguageModel;
|
|
24
23
|
options: NoiseSensitivityOptions;
|
|
25
|
-
}): import("@mastra/core/scores").MastraScorer<
|
|
24
|
+
}): import("@mastra/core/scores").MastraScorer<"Noise Sensitivity (LLM)", import("@mastra/core/scores").ScorerRunInputForAgent, import("@mastra/core/scores").ScorerRunOutputForAgent, Record<"analyzeStepResult", {
|
|
26
25
|
overallAssessment: string;
|
|
27
26
|
dimensions: {
|
|
28
27
|
dimension: 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,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/noise-sensitivity/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAM9D,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,mBAAmB,CAAC;IAC3B,OAAO,EAAE,uBAAuB,CAAC;CAClC;;;;;;;;;;6FAqHA"}
|
|
@@ -1,11 +1,10 @@
|
|
|
1
1
|
import type { Tool } from '@mastra/core';
|
|
2
2
|
import type { MastraLanguageModel } from '@mastra/core/agent';
|
|
3
|
-
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
4
3
|
export interface ToolCallAccuracyOptions {
|
|
5
4
|
model: MastraLanguageModel;
|
|
6
5
|
availableTools: Tool[];
|
|
7
6
|
}
|
|
8
|
-
export declare function createToolCallAccuracyScorerLLM({ model, availableTools }: ToolCallAccuracyOptions): import("@mastra/core/scores").MastraScorer<
|
|
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", {
|
|
9
8
|
actualTools: string[];
|
|
10
9
|
hasToolCalls: boolean;
|
|
11
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":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,cAAc,CAAC;AACzC,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;
|
|
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,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAW9D,MAAM,WAAW,uBAAuB;IACtC,KAAK,EAAE,mBAAmB,CAAC;IAC3B,cAAc,EAAE,IAAI,EAAE,CAAC;CACxB;AAaD,wBAAgB,+BAA+B,CAAC,EAAE,KAAK,EAAE,cAAc,EAAE,EAAE,uBAAuB;;;;;;;;;;;6FA2EjG"}
|
|
@@ -1,12 +1,11 @@
|
|
|
1
1
|
import type { LanguageModel } from '@mastra/core/llm';
|
|
2
|
-
import type { ScorerRunInputForAgent, ScorerRunOutputForAgent } from '@mastra/core/scores';
|
|
3
2
|
export interface ToxicityMetricOptions {
|
|
4
3
|
scale?: number;
|
|
5
4
|
}
|
|
6
5
|
export declare function createToxicityScorer({ model, options }: {
|
|
7
6
|
model: LanguageModel;
|
|
8
7
|
options?: ToxicityMetricOptions;
|
|
9
|
-
}): import("@mastra/core/scores").MastraScorer<
|
|
8
|
+
}): import("@mastra/core/scores").MastraScorer<"Toxicity Scorer", import("@mastra/core/scores").ScorerRunInputForAgent, import("@mastra/core/scores").ScorerRunOutputForAgent, Record<"analyzeStepResult", {
|
|
10
9
|
verdicts: {
|
|
11
10
|
verdict: string;
|
|
12
11
|
reason: string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/toxicity/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../../src/scorers/llm/toxicity/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAMtD,MAAM,WAAW,qBAAqB;IACpC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,wBAAgB,oBAAoB,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;IAAE,KAAK,EAAE,aAAa,CAAC;IAAC,OAAO,CAAC,EAAE,qBAAqB,CAAA;CAAE;;;;;6FA+CjH"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@mastra/evals",
|
|
3
|
-
"version": "0.13.
|
|
3
|
+
"version": "0.13.8-alpha.2",
|
|
4
4
|
"description": "",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"files": [
|
|
@@ -97,13 +97,13 @@
|
|
|
97
97
|
"dependencies": {
|
|
98
98
|
"compromise": "^14.14.4",
|
|
99
99
|
"difflib": "^0.2.4",
|
|
100
|
-
"fs-extra": "^11.3.
|
|
100
|
+
"fs-extra": "^11.3.2",
|
|
101
101
|
"keyword-extractor": "^0.0.28",
|
|
102
102
|
"sentiment": "^5.0.2",
|
|
103
103
|
"string-similarity": "^4.0.4"
|
|
104
104
|
},
|
|
105
105
|
"peerDependencies": {
|
|
106
|
-
"@mastra/core": ">=0.15.3-0 <0.
|
|
106
|
+
"@mastra/core": ">=0.15.3-0 <0.19.0-0",
|
|
107
107
|
"ai": "^4.0.0 || ^5.0.0",
|
|
108
108
|
"zod": "^3.25.0 || ^4.0.0"
|
|
109
109
|
},
|
|
@@ -121,9 +121,9 @@
|
|
|
121
121
|
"typescript": "^5.8.3",
|
|
122
122
|
"vitest": "^3.2.4",
|
|
123
123
|
"zod": "^3.25.76",
|
|
124
|
-
"@
|
|
125
|
-
"@internal/types-builder": "0.0.
|
|
126
|
-
"@
|
|
124
|
+
"@mastra/core": "0.18.0-alpha.3",
|
|
125
|
+
"@internal/types-builder": "0.0.16",
|
|
126
|
+
"@internal/lint": "0.0.41"
|
|
127
127
|
},
|
|
128
128
|
"scripts": {
|
|
129
129
|
"check": "tsc --noEmit",
|