judgeval 0.3.0 → 0.3.2
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/dist/constants.d.ts +1 -1
- package/dist/constants.d.ts.map +1 -1
- package/dist/env.d.ts +2 -0
- package/dist/env.d.ts.map +1 -1
- package/dist/index.d.ts +2 -0
- package/dist/index.d.ts.map +1 -1
- package/dist/index.mjs +5 -2
- package/dist/index.mjs.map +30 -1
- package/dist/index.umd.js +5 -2
- package/dist/index.umd.js.map +30 -1
- package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.d.ts +1 -1
- package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.d.ts.map +1 -1
- package/dist/scorers/base-scorer.d.ts +1 -1
- package/dist/scorers/base-scorer.d.ts.map +1 -1
- package/dist/tracer/BrowserTracer.d.ts +11 -0
- package/dist/tracer/BrowserTracer.d.ts.map +1 -0
- package/dist/tracer/NodeTracer.d.ts +16 -0
- package/dist/tracer/NodeTracer.d.ts.map +1 -0
- package/dist/tracer/OpenTelemetryKeys.d.ts +28 -1
- package/dist/tracer/OpenTelemetryKeys.d.ts.map +1 -1
- package/dist/tracer/Tracer.d.ts +22 -23
- package/dist/tracer/Tracer.d.ts.map +1 -1
- package/dist/tracer/TracerConfiguration.d.ts.map +1 -1
- package/dist/tracer/exporters/JudgmentSpanExporter.d.ts +49 -2
- package/dist/tracer/exporters/JudgmentSpanExporter.d.ts.map +1 -1
- package/dist/utils/guards.d.ts +4 -0
- package/dist/utils/guards.d.ts.map +1 -0
- package/dist/utils/index.d.ts +1 -0
- package/dist/utils/index.d.ts.map +1 -1
- package/dist/version.d.ts +1 -1
- package/package.json +19 -20
- package/dist/constants.js +0 -2
- package/dist/constants.js.map +0 -1
- package/dist/data/example-params.js +0 -12
- package/dist/data/example-params.js.map +0 -1
- package/dist/data/example.js +0 -10
- package/dist/data/example.js.map +0 -1
- package/dist/data/index.js +0 -3
- package/dist/data/index.js.map +0 -1
- package/dist/env.js +0 -13
- package/dist/env.js.map +0 -1
- package/dist/index.js +0 -2
- package/dist/index.js.map +0 -1
- package/dist/internal/api/index.js +0 -132
- package/dist/internal/api/index.js.map +0 -1
- package/dist/internal/api/models/BaseScorer.js +0 -6
- package/dist/internal/api/models/BaseScorer.js.map +0 -1
- package/dist/internal/api/models/EvalResults.js +0 -6
- package/dist/internal/api/models/EvalResults.js.map +0 -1
- package/dist/internal/api/models/EvalResultsFetch.js +0 -6
- package/dist/internal/api/models/EvalResultsFetch.js.map +0 -1
- package/dist/internal/api/models/Example.js +0 -6
- package/dist/internal/api/models/Example.js.map +0 -1
- package/dist/internal/api/models/ExampleEvaluationRun.js +0 -6
- package/dist/internal/api/models/ExampleEvaluationRun.js.map +0 -1
- package/dist/internal/api/models/FetchPromptScorersRequest.js +0 -6
- package/dist/internal/api/models/FetchPromptScorersRequest.js.map +0 -1
- package/dist/internal/api/models/FetchPromptScorersResponse.js +0 -6
- package/dist/internal/api/models/FetchPromptScorersResponse.js.map +0 -1
- package/dist/internal/api/models/OtelTraceSpan.js +0 -6
- package/dist/internal/api/models/OtelTraceSpan.js.map +0 -1
- package/dist/internal/api/models/PromptScorer.js +0 -6
- package/dist/internal/api/models/PromptScorer.js.map +0 -1
- package/dist/internal/api/models/ResolveProjectNameRequest.js +0 -6
- package/dist/internal/api/models/ResolveProjectNameRequest.js.map +0 -1
- package/dist/internal/api/models/ResolveProjectNameResponse.js +0 -6
- package/dist/internal/api/models/ResolveProjectNameResponse.js.map +0 -1
- package/dist/internal/api/models/SavePromptScorerRequest.js +0 -6
- package/dist/internal/api/models/SavePromptScorerRequest.js.map +0 -1
- package/dist/internal/api/models/SavePromptScorerResponse.js +0 -6
- package/dist/internal/api/models/SavePromptScorerResponse.js.map +0 -1
- package/dist/internal/api/models/ScorerConfig.js +0 -6
- package/dist/internal/api/models/ScorerConfig.js.map +0 -1
- package/dist/internal/api/models/ScorerData.js +0 -6
- package/dist/internal/api/models/ScorerData.js.map +0 -1
- package/dist/internal/api/models/ScorerExistsRequest.js +0 -6
- package/dist/internal/api/models/ScorerExistsRequest.js.map +0 -1
- package/dist/internal/api/models/ScorerExistsResponse.js +0 -6
- package/dist/internal/api/models/ScorerExistsResponse.js.map +0 -1
- package/dist/internal/api/models/ScoringResult.js +0 -6
- package/dist/internal/api/models/ScoringResult.js.map +0 -1
- package/dist/internal/api/models/TraceEvaluationRun.js +0 -6
- package/dist/internal/api/models/TraceEvaluationRun.js.map +0 -1
- package/dist/internal/api/models.js +0 -24
- package/dist/internal/api/models.js.map +0 -1
- package/dist/scorers/api-scorer.js +0 -36
- package/dist/scorers/api-scorer.js.map +0 -1
- package/dist/scorers/api_scorers/answer-correctness-scorer.js +0 -20
- package/dist/scorers/api_scorers/answer-correctness-scorer.js.map +0 -1
- package/dist/scorers/api_scorers/answer-relevancy-scorer.js +0 -19
- package/dist/scorers/api_scorers/answer-relevancy-scorer.js.map +0 -1
- package/dist/scorers/api_scorers/faithfulness-scorer.js +0 -20
- package/dist/scorers/api_scorers/faithfulness-scorer.js.map +0 -1
- package/dist/scorers/api_scorers/index.js +0 -2
- package/dist/scorers/api_scorers/index.js.map +0 -1
- package/dist/scorers/api_scorers/instruction-adherence-scorer.js +0 -20
- package/dist/scorers/api_scorers/instruction-adherence-scorer.js.map +0 -1
- package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.js +0 -133
- package/dist/scorers/api_scorers/prompt_scorer/base-prompt-scorer.js.map +0 -1
- package/dist/scorers/api_scorers/prompt_scorer/index.js +0 -5
- package/dist/scorers/api_scorers/prompt_scorer/index.js.map +0 -1
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.js +0 -44
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.js.map +0 -1
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer.js +0 -9
- package/dist/scorers/api_scorers/prompt_scorer/prompt-scorer.js.map +0 -1
- package/dist/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.js +0 -8
- package/dist/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.js.map +0 -1
- package/dist/scorers/base-scorer.js +0 -41
- package/dist/scorers/base-scorer.js.map +0 -1
- package/dist/scorers/index.js +0 -4
- package/dist/scorers/index.js.map +0 -1
- package/dist/tracer/OpenTelemetryKeys.js +0 -12
- package/dist/tracer/OpenTelemetryKeys.js.map +0 -1
- package/dist/tracer/Tracer.js +0 -235
- package/dist/tracer/Tracer.js.map +0 -1
- package/dist/tracer/TracerConfiguration.js +0 -111
- package/dist/tracer/TracerConfiguration.js.map +0 -1
- package/dist/tracer/exporters/JudgmentSpanExporter.js +0 -72
- package/dist/tracer/exporters/JudgmentSpanExporter.js.map +0 -1
- package/dist/tracer/exporters/NoOpSpanExporter.js +0 -16
- package/dist/tracer/exporters/NoOpSpanExporter.js.map +0 -1
- package/dist/tracer/exporters/index.js +0 -3
- package/dist/tracer/exporters/index.js.map +0 -1
- package/dist/tracer/index.js +0 -5
- package/dist/tracer/index.js.map +0 -1
- package/dist/umd.js +0 -2
- package/dist/umd.js.map +0 -1
- package/dist/utils/annotate.js +0 -27
- package/dist/utils/annotate.js.map +0 -1
- package/dist/utils/index.js +0 -5
- package/dist/utils/index.js.map +0 -1
- package/dist/utils/logger.js +0 -70
- package/dist/utils/logger.js.map +0 -1
- package/dist/utils/types.js +0 -2
- package/dist/utils/types.js.map +0 -1
- package/dist/utils/validation.js +0 -16
- package/dist/utils/validation.js.map +0 -1
- package/dist/version.js +0 -2
- package/dist/version.js.map +0 -1
|
@@ -1,20 +0,0 @@
|
|
|
1
|
-
import { ExampleParams } from "../../data";
|
|
2
|
-
import { APIScorerType, createAPIScorer } from "../api-scorer";
|
|
3
|
-
const ANSWER_CORRECTNESS_REQUIRED_PARAMS = [
|
|
4
|
-
ExampleParams.INPUT,
|
|
5
|
-
ExampleParams.ACTUAL_OUTPUT,
|
|
6
|
-
ExampleParams.EXPECTED_OUTPUT,
|
|
7
|
-
];
|
|
8
|
-
export function createAnswerCorrectnessScorer(scorerArgs) {
|
|
9
|
-
const scorer = createAPIScorer(APIScorerType.ANSWER_CORRECTNESS, ANSWER_CORRECTNESS_REQUIRED_PARAMS);
|
|
10
|
-
if (scorerArgs) {
|
|
11
|
-
if (scorerArgs.threshold !== undefined) {
|
|
12
|
-
scorer.setThreshold(scorerArgs.threshold);
|
|
13
|
-
}
|
|
14
|
-
if (scorerArgs.model) {
|
|
15
|
-
scorer.addModel(scorerArgs.model);
|
|
16
|
-
}
|
|
17
|
-
}
|
|
18
|
-
return scorer;
|
|
19
|
-
}
|
|
20
|
-
//# sourceMappingURL=answer-correctness-scorer.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"answer-correctness-scorer.js","sourceRoot":"","sources":["../../../src/scorers/api_scorers/answer-correctness-scorer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAC3C,OAAO,EAAa,aAAa,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAE1E,MAAM,kCAAkC,GAAG;IACzC,aAAa,CAAC,KAAK;IACnB,aAAa,CAAC,aAAa;IAC3B,aAAa,CAAC,eAAe;CACrB,CAAC;AAYX,MAAM,UAAU,6BAA6B,CAC3C,UAAwC;IAExC,MAAM,MAAM,GAAG,eAAe,CAC5B,aAAa,CAAC,kBAAkB,EAChC,kCAAkC,CACnC,CAAC;IAEF,IAAI,UAAU,EAAE,CAAC;QACf,IAAI,UAAU,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;YACvC,MAAM,CAAC,YAAY,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;QAC5C,CAAC;QACD,IAAI,UAAU,CAAC,KAAK,EAAE,CAAC;YACrB,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC;QACpC,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC"}
|
|
@@ -1,19 +0,0 @@
|
|
|
1
|
-
import { ExampleParams } from "../../data";
|
|
2
|
-
import { APIScorerType, createAPIScorer } from "../api-scorer";
|
|
3
|
-
const ANSWER_RELEVANCY_REQUIRED_PARAMS = [
|
|
4
|
-
ExampleParams.INPUT,
|
|
5
|
-
ExampleParams.ACTUAL_OUTPUT,
|
|
6
|
-
];
|
|
7
|
-
export function createAnswerRelevancyScorer(scorerArgs) {
|
|
8
|
-
const scorer = createAPIScorer(APIScorerType.ANSWER_RELEVANCY, ANSWER_RELEVANCY_REQUIRED_PARAMS);
|
|
9
|
-
if (scorerArgs) {
|
|
10
|
-
if (scorerArgs.threshold !== undefined) {
|
|
11
|
-
scorer.setThreshold(scorerArgs.threshold);
|
|
12
|
-
}
|
|
13
|
-
if (scorerArgs.model) {
|
|
14
|
-
scorer.addModel(scorerArgs.model);
|
|
15
|
-
}
|
|
16
|
-
}
|
|
17
|
-
return scorer;
|
|
18
|
-
}
|
|
19
|
-
//# sourceMappingURL=answer-relevancy-scorer.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"answer-relevancy-scorer.js","sourceRoot":"","sources":["../../../src/scorers/api_scorers/answer-relevancy-scorer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAC3C,OAAO,EAAa,aAAa,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAE1E,MAAM,gCAAgC,GAAG;IACvC,aAAa,CAAC,KAAK;IACnB,aAAa,CAAC,aAAa;CACnB,CAAC;AAYX,MAAM,UAAU,2BAA2B,CACzC,UAAsC;IAEtC,MAAM,MAAM,GAAG,eAAe,CAC5B,aAAa,CAAC,gBAAgB,EAC9B,gCAAgC,CACjC,CAAC;IAEF,IAAI,UAAU,EAAE,CAAC;QACf,IAAI,UAAU,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;YACvC,MAAM,CAAC,YAAY,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;QAC5C,CAAC;QACD,IAAI,UAAU,CAAC,KAAK,EAAE,CAAC;YACrB,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC;QACpC,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC"}
|
|
@@ -1,20 +0,0 @@
|
|
|
1
|
-
import { ExampleParams } from "../../data";
|
|
2
|
-
import { APIScorerType, createAPIScorer } from "../api-scorer";
|
|
3
|
-
const FAITHFULNESS_REQUIRED_PARAMS = [
|
|
4
|
-
ExampleParams.INPUT,
|
|
5
|
-
ExampleParams.ACTUAL_OUTPUT,
|
|
6
|
-
ExampleParams.RETRIEVAL_CONTEXT,
|
|
7
|
-
];
|
|
8
|
-
export function createFaithfulnessScorer(scorerArgs) {
|
|
9
|
-
const scorer = createAPIScorer(APIScorerType.FAITHFULNESS, FAITHFULNESS_REQUIRED_PARAMS);
|
|
10
|
-
if (scorerArgs) {
|
|
11
|
-
if (scorerArgs.threshold !== undefined) {
|
|
12
|
-
scorer.setThreshold(scorerArgs.threshold);
|
|
13
|
-
}
|
|
14
|
-
if (scorerArgs.model) {
|
|
15
|
-
scorer.addModel(scorerArgs.model);
|
|
16
|
-
}
|
|
17
|
-
}
|
|
18
|
-
return scorer;
|
|
19
|
-
}
|
|
20
|
-
//# sourceMappingURL=faithfulness-scorer.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"faithfulness-scorer.js","sourceRoot":"","sources":["../../../src/scorers/api_scorers/faithfulness-scorer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAC3C,OAAO,EAAa,aAAa,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAE1E,MAAM,4BAA4B,GAAG;IACnC,aAAa,CAAC,KAAK;IACnB,aAAa,CAAC,aAAa;IAC3B,aAAa,CAAC,iBAAiB;CACvB,CAAC;AAYX,MAAM,UAAU,wBAAwB,CACtC,UAAmC;IAEnC,MAAM,MAAM,GAAG,eAAe,CAC5B,aAAa,CAAC,YAAY,EAC1B,4BAA4B,CAC7B,CAAC;IAEF,IAAI,UAAU,EAAE,CAAC;QACf,IAAI,UAAU,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;YACvC,MAAM,CAAC,YAAY,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;QAC5C,CAAC;QACD,IAAI,UAAU,CAAC,KAAK,EAAE,CAAC;YACrB,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC;QACpC,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/scorers/api_scorers/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,gBAAgB,EAChB,iBAAiB,EACjB,gBAAgB,EAChB,YAAY,EACZ,gBAAgB,EAChB,YAAY,EACZ,iBAAiB,GAClB,MAAM,iBAAiB,CAAC"}
|
|
@@ -1,20 +0,0 @@
|
|
|
1
|
-
import { ExampleParams } from "../../data";
|
|
2
|
-
import { APIScorerType, createAPIScorer } from "../api-scorer";
|
|
3
|
-
const INSTRUCTION_ADHERENCE_REQUIRED_PARAMS = [
|
|
4
|
-
ExampleParams.INPUT,
|
|
5
|
-
ExampleParams.ACTUAL_OUTPUT,
|
|
6
|
-
];
|
|
7
|
-
export function createInstructionAdherenceScorer(scorerArgs) {
|
|
8
|
-
const scorer = createAPIScorer(APIScorerType.INSTRUCTION_ADHERENCE, INSTRUCTION_ADHERENCE_REQUIRED_PARAMS);
|
|
9
|
-
scorer.name = "Instruction Adherence";
|
|
10
|
-
if (scorerArgs) {
|
|
11
|
-
if (scorerArgs.threshold !== undefined) {
|
|
12
|
-
scorer.setThreshold(scorerArgs.threshold);
|
|
13
|
-
}
|
|
14
|
-
if (scorerArgs.model) {
|
|
15
|
-
scorer.addModel(scorerArgs.model);
|
|
16
|
-
}
|
|
17
|
-
}
|
|
18
|
-
return scorer;
|
|
19
|
-
}
|
|
20
|
-
//# sourceMappingURL=instruction-adherence-scorer.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"instruction-adherence-scorer.js","sourceRoot":"","sources":["../../../src/scorers/api_scorers/instruction-adherence-scorer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAC3C,OAAO,EAAa,aAAa,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAE1E,MAAM,qCAAqC,GAAG;IAC5C,aAAa,CAAC,KAAK;IACnB,aAAa,CAAC,aAAa;CACnB,CAAC;AAYX,MAAM,UAAU,gCAAgC,CAC9C,UAA2C;IAE3C,MAAM,MAAM,GAAG,eAAe,CAC5B,aAAa,CAAC,qBAAqB,EACnC,qCAAqC,CACtC,CAAC;IAEF,MAAM,CAAC,IAAI,GAAG,uBAAuB,CAAC;IAEtC,IAAI,UAAU,EAAE,CAAC;QACf,IAAI,UAAU,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;YACvC,MAAM,CAAC,YAAY,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;QAC5C,CAAC;QACD,IAAI,UAAU,CAAC,KAAK,EAAE,CAAC;YACrB,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC;QACpC,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC"}
|
|
@@ -1,133 +0,0 @@
|
|
|
1
|
-
import { JUDGMENT_API_KEY, JUDGMENT_ORG_ID } from "../../../env";
|
|
2
|
-
import { APIScorerType } from "../../api-scorer";
|
|
3
|
-
import { fetchPromptScorer, JudgmentAPIError, pushPromptScorer, scorerExists, } from "./prompt-scorer-utils";
|
|
4
|
-
export class BasePromptScorer {
|
|
5
|
-
constructor(scoreType, name, prompt, threshold, requiredParams, options, judgmentApiKey = JUDGMENT_API_KEY || "", organizationId = JUDGMENT_ORG_ID || "") {
|
|
6
|
-
this.scoreType = scoreType;
|
|
7
|
-
this.name = name;
|
|
8
|
-
this.prompt = prompt;
|
|
9
|
-
this.threshold = threshold;
|
|
10
|
-
this.requiredParams = requiredParams;
|
|
11
|
-
this.options = options;
|
|
12
|
-
this.judgmentApiKey = judgmentApiKey;
|
|
13
|
-
this.organizationId = organizationId;
|
|
14
|
-
this.score_type = scoreType;
|
|
15
|
-
this.class_name = "BasePromptScorer";
|
|
16
|
-
this.model = undefined;
|
|
17
|
-
this.score = undefined;
|
|
18
|
-
this.error = null;
|
|
19
|
-
this.strict_mode = false;
|
|
20
|
-
}
|
|
21
|
-
static async get(name, judgmentApiKey = JUDGMENT_API_KEY || "", organizationId = JUDGMENT_ORG_ID || "") {
|
|
22
|
-
const config = await fetchPromptScorer(name, judgmentApiKey, organizationId);
|
|
23
|
-
const isTrace = config.is_trace === true;
|
|
24
|
-
const expectedIsTrace = this.prototype.scoreType === APIScorerType.TRACE_PROMPT_SCORER;
|
|
25
|
-
if (isTrace !== expectedIsTrace) {
|
|
26
|
-
throw new JudgmentAPIError(400, `Scorer with name ${name} is not a ${this.name}`);
|
|
27
|
-
}
|
|
28
|
-
const scoreType = isTrace
|
|
29
|
-
? APIScorerType.TRACE_PROMPT_SCORER
|
|
30
|
-
: APIScorerType.PROMPT_SCORER;
|
|
31
|
-
return new this(scoreType, config.name, // Use config.name instead of name parameter
|
|
32
|
-
config.prompt, config.threshold, [], config.options, judgmentApiKey, organizationId);
|
|
33
|
-
}
|
|
34
|
-
static async create(name, prompt, threshold = 0.5, options, judgmentApiKey = JUDGMENT_API_KEY || "", organizationId = JUDGMENT_ORG_ID || "") {
|
|
35
|
-
if (await scorerExists(name, judgmentApiKey, organizationId)) {
|
|
36
|
-
throw new JudgmentAPIError(400, `Scorer with name ${name} already exists. Either use the existing scorer with the get() method or use a new name.`);
|
|
37
|
-
}
|
|
38
|
-
const isTrace = this.prototype.scoreType === APIScorerType.TRACE_PROMPT_SCORER;
|
|
39
|
-
const scoreType = isTrace
|
|
40
|
-
? APIScorerType.TRACE_PROMPT_SCORER
|
|
41
|
-
: APIScorerType.PROMPT_SCORER;
|
|
42
|
-
await pushPromptScorer(name, prompt, threshold, options, judgmentApiKey, organizationId, isTrace);
|
|
43
|
-
return new this(scoreType, name, prompt, threshold, [], options, judgmentApiKey, organizationId);
|
|
44
|
-
}
|
|
45
|
-
async updateThreshold(threshold) {
|
|
46
|
-
this.setThreshold(threshold);
|
|
47
|
-
await this.pushPromptScorer();
|
|
48
|
-
}
|
|
49
|
-
async setPrompt(prompt) {
|
|
50
|
-
this.prompt = prompt;
|
|
51
|
-
await this.pushPromptScorer();
|
|
52
|
-
}
|
|
53
|
-
async setOptions(options) {
|
|
54
|
-
this.options = options;
|
|
55
|
-
await this.pushPromptScorer();
|
|
56
|
-
}
|
|
57
|
-
async appendToPrompt(promptAddition) {
|
|
58
|
-
this.prompt += promptAddition;
|
|
59
|
-
await this.pushPromptScorer();
|
|
60
|
-
}
|
|
61
|
-
getThreshold() {
|
|
62
|
-
return this.threshold;
|
|
63
|
-
}
|
|
64
|
-
getPrompt() {
|
|
65
|
-
return this.prompt;
|
|
66
|
-
}
|
|
67
|
-
getOptions() {
|
|
68
|
-
return this.options ? { ...this.options } : null;
|
|
69
|
-
}
|
|
70
|
-
getName() {
|
|
71
|
-
return this.name;
|
|
72
|
-
}
|
|
73
|
-
getConfig() {
|
|
74
|
-
return {
|
|
75
|
-
name: this.name,
|
|
76
|
-
prompt: this.prompt,
|
|
77
|
-
threshold: this.threshold,
|
|
78
|
-
options: this.options,
|
|
79
|
-
};
|
|
80
|
-
}
|
|
81
|
-
async pushPromptScorer() {
|
|
82
|
-
await pushPromptScorer(this.name, this.prompt, this.threshold, this.options, this.judgmentApiKey, this.organizationId);
|
|
83
|
-
}
|
|
84
|
-
toString() {
|
|
85
|
-
return `${this.constructor.name}(name=${this.name}, prompt=${this.prompt}, threshold=${this.threshold}, options=${JSON.stringify(this.options)})`;
|
|
86
|
-
}
|
|
87
|
-
addModel(model) {
|
|
88
|
-
this.model = model;
|
|
89
|
-
}
|
|
90
|
-
successCheck() {
|
|
91
|
-
if (this.error != null) {
|
|
92
|
-
return false;
|
|
93
|
-
}
|
|
94
|
-
if (this.score == null) {
|
|
95
|
-
return false;
|
|
96
|
-
}
|
|
97
|
-
const threshold = this.threshold ?? 0.5;
|
|
98
|
-
const score = this.score;
|
|
99
|
-
return threshold != null && score != null && score >= threshold;
|
|
100
|
-
}
|
|
101
|
-
getRequiredParams() {
|
|
102
|
-
if (Array.isArray(this.requiredParams)) {
|
|
103
|
-
return [...this.requiredParams];
|
|
104
|
-
}
|
|
105
|
-
return [];
|
|
106
|
-
}
|
|
107
|
-
setThreshold(threshold) {
|
|
108
|
-
if (threshold < 0 || threshold > 1) {
|
|
109
|
-
throw new Error(`Threshold must be between 0 and 1, got: ${threshold}`);
|
|
110
|
-
}
|
|
111
|
-
this.threshold = threshold;
|
|
112
|
-
}
|
|
113
|
-
getScoreType() {
|
|
114
|
-
return this.scoreType;
|
|
115
|
-
}
|
|
116
|
-
setRequiredParams(params) {
|
|
117
|
-
this.requiredParams = params;
|
|
118
|
-
}
|
|
119
|
-
toTransport() {
|
|
120
|
-
return {
|
|
121
|
-
score_type: this.getScoreType(),
|
|
122
|
-
name: this.getName(),
|
|
123
|
-
threshold: this.getThreshold(),
|
|
124
|
-
strict_mode: this.strict_mode ?? false,
|
|
125
|
-
required_params: this.getRequiredParams(),
|
|
126
|
-
kwargs: {
|
|
127
|
-
prompt: this.getPrompt(),
|
|
128
|
-
...(this.getOptions() ? { options: this.getOptions() } : {}),
|
|
129
|
-
},
|
|
130
|
-
};
|
|
131
|
-
}
|
|
132
|
-
}
|
|
133
|
-
//# sourceMappingURL=base-prompt-scorer.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"base-prompt-scorer.js","sourceRoot":"","sources":["../../../../src/scorers/api_scorers/prompt_scorer/base-prompt-scorer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,cAAc,CAAC;AAEjE,OAAO,EAAa,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAC5D,OAAO,EACL,iBAAiB,EACjB,gBAAgB,EAChB,gBAAgB,EAChB,YAAY,GACb,MAAM,uBAAuB,CAAC;AAE/B,MAAM,OAAgB,gBAAgB;IAkBpC,YACE,SAAwB,EACxB,IAAY,EACZ,MAAc,EACd,SAAiB,EACjB,cAAiC,EACjC,OAAuC,EACvC,iBAAyB,gBAAgB,IAAI,EAAE,EAC/C,iBAAyB,eAAe,IAAI,EAAE;QAE9C,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;QACjB,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,cAAc,GAAG,cAAc,CAAC;QACrC,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;QACvB,IAAI,CAAC,cAAc,GAAG,cAAc,CAAC;QACrC,IAAI,CAAC,cAAc,GAAG,cAAc,CAAC;QACrC,IAAI,CAAC,UAAU,GAAG,SAAS,CAAC;QAC5B,IAAI,CAAC,UAAU,GAAG,kBAAkB,CAAC;QACrC,IAAI,CAAC,KAAK,GAAG,SAAS,CAAC;QACvB,IAAI,CAAC,KAAK,GAAG,SAAS,CAAC;QACvB,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;QAClB,IAAI,CAAC,WAAW,GAAG,KAAK,CAAC;IAC3B,CAAC;IAUD,MAAM,CAAC,KAAK,CAAC,GAAG,CAEd,IAAY,EACZ,iBAAyB,gBAAgB,IAAI,EAAE,EAC/C,iBAAyB,eAAe,IAAI,EAAE;QAE9C,MAAM,MAAM,GAAG,MAAM,iBAAiB,CACpC,IAAI,EACJ,cAAc,EACd,cAAc,CACf,CAAC;QAEF,MAAM,OAAO,GAAG,MAAM,CAAC,QAAQ,KAAK,IAAI,CAAC;QACzC,MAAM,eAAe,GACnB,IAAI,CAAC,SAAS,CAAC,SAAS,KAAK,aAAa,CAAC,mBAAmB,CAAC;QAEjE,IAAI,OAAO,KAAK,eAAe,EAAE,CAAC;YAChC,MAAM,IAAI,gBAAgB,CACxB,GAAG,EACH,oBAAoB,IAAI,aAAa,IAAI,CAAC,IAAI,EAAE,CACjD,CAAC;QACJ,CAAC;QAED,MAAM,SAAS,GAAG,OAAO;YACvB,CAAC,CAAC,aAAa,CAAC,mBAAmB;YACnC,CAAC,CAAC,aAAa,CAAC,aAAa,CAAC;QAEhC,OAAO,IAAI,IAAI,CACb,SAAS,EACT,MAAM,CAAC,IAAI,EAAE,4CAA4C;QACzD,MAAM,CAAC,MAAM,EACb,MAAM,CAAC,SAAS,EAChB,EAAE,EACF,MAAM,CAAC,OAAO,EACd,cAAc,EACd,cAAc,CACf,CAAC;IACJ,CAAC;IAED,MAAM,CAAC,KAAK,CAAC,MAAM,CAEjB,IAAY,EACZ,MAAc,EACd,YAAoB,GAAG,EACvB,OAAuC,EACvC,iBAAyB,gBAAgB,IAAI,EAAE,EAC/C,iBAAyB,eAAe,IAAI,EAAE;QAE9C,IAAI,MAAM,YAAY,CAAC,IAAI,EAAE,cAAc,EAAE,cAAc,CAAC,EAAE,CAAC;YAC7D,MAAM,IAAI,gBAAgB,CACxB,GAAG,EACH,oBAAoB,IAAI,0FAA0F,CACnH,CAAC;QACJ,CAAC;QAED,MAAM,OAAO,GACX,IAAI,CAAC,SAAS,CAAC,SAAS,KAAK,aAAa,CAAC,mBAAmB,CAAC;QACjE,MAAM,SAAS,GAAG,OAAO;YACvB,CAAC,CAAC,aAAa,CAAC,mBAAmB;YACnC,CAAC,CAAC,aAAa,CAAC,aAAa,CAAC;QAEhC,MAAM,gBAAgB,CACpB,IAAI,EACJ,MAAM,EACN,SAAS,EACT,OAAO,EACP,cAAc,EACd,cAAc,EACd,OAAO,CACR,CAAC;QAEF,OAAO,IAAI,IAAI,CACb,SAAS,EACT,IAAI,EACJ,MAAM,EACN,SAAS,EACT,EAAE,EACF,OAAO,EACP,cAAc,EACd,cAAc,CACf,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,eAAe,CAAC,SAAiB;QACrC,IAAI,CAAC,YAAY,CAAC,SAAS,CAAC,CAAC;QAC7B,MAAM,IAAI,CAAC,gBAAgB,EAAE,CAAC;IAChC,CAAC;IAED,KAAK,CAAC,SAAS,CAAC,MAAc;QAC5B,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,MAAM,IAAI,CAAC,gBAAgB,EAAE,CAAC;IAChC,CAAC;IAED,KAAK,CAAC,UAAU,CAAC,OAAsC;QACrD,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;QACvB,MAAM,IAAI,CAAC,gBAAgB,EAAE,CAAC;IAChC,CAAC;IAED,KAAK,CAAC,cAAc,CAAC,cAAsB;QACzC,IAAI,CAAC,MAAM,IAAI,cAAc,CAAC;QAC9B,MAAM,IAAI,CAAC,gBAAgB,EAAE,CAAC;IAChC,CAAC;IAED,YAAY;QACV,OAAO,IAAI,CAAC,SAAS,CAAC;IACxB,CAAC;IAED,SAAS;QACP,OAAO,IAAI,CAAC,MAAM,CAAC;IACrB,CAAC;IAED,UAAU;QACR,OAAO,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;IACnD,CAAC;IAED,OAAO;QACL,OAAO,IAAI,CAAC,IAAI,CAAC;IACnB,CAAC;IAED,SAAS;QACP,OAAO;YACL,IAAI,EAAE,IAAI,CAAC,IAAI;YACf,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,OAAO,EAAE,IAAI,CAAC,OAAO;SACtB,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,gBAAgB;QACpB,MAAM,gBAAgB,CACpB,IAAI,CAAC,IAAI,EACT,IAAI,CAAC,MAAM,EACX,IAAI,CAAC,SAAS,EACd,IAAI,CAAC,OAAO,EACZ,IAAI,CAAC,cAAc,EACnB,IAAI,CAAC,cAAc,CACpB,CAAC;IACJ,CAAC;IAED,QAAQ;QACN,OAAO,GAAG,IAAI,CAAC,WAAW,CAAC,IAAI,SAAS,IAAI,CAAC,IAAI,YAC/C,IAAI,CAAC,MACP,eAAe,IAAI,CAAC,SAAS,aAAa,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC;IAC5E,CAAC;IAED,QAAQ,CAAC,KAAa;QACpB,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;IACrB,CAAC;IAED,YAAY;QACV,IAAI,IAAI,CAAC,KAAK,IAAI,IAAI,EAAE,CAAC;YACvB,OAAO,KAAK,CAAC;QACf,CAAC;QACD,IAAI,IAAI,CAAC,KAAK,IAAI,IAAI,EAAE,CAAC;YACvB,OAAO,KAAK,CAAC;QACf,CAAC;QACD,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,GAAG,CAAC;QACxC,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACzB,OAAO,SAAS,IAAI,IAAI,IAAI,KAAK,IAAI,IAAI,IAAI,KAAK,IAAI,SAAS,CAAC;IAClE,CAAC;IAED,iBAAiB;QACf,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,cAAc,CAAC,EAAE,CAAC;YACvC,OAAO,CAAC,GAAG,IAAI,CAAC,cAAc,CAAC,CAAC;QAClC,CAAC;QACD,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,YAAY,CAAC,SAAiB;QAC5B,IAAI,SAAS,GAAG,CAAC,IAAI,SAAS,GAAG,CAAC,EAAE,CAAC;YACnC,MAAM,IAAI,KAAK,CAAC,2CAA2C,SAAS,EAAE,CAAC,CAAC;QAC1E,CAAC;QACD,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;IAC7B,CAAC;IAED,YAAY;QACV,OAAO,IAAI,CAAC,SAAS,CAAC;IACxB,CAAC;IAED,iBAAiB,CAAC,MAAyB;QACzC,IAAI,CAAC,cAAc,GAAG,MAAM,CAAC;IAC/B,CAAC;IAED,WAAW;QACT,OAAO;YACL,UAAU,EAAE,IAAI,CAAC,YAAY,EAAE;YAC/B,IAAI,EAAE,IAAI,CAAC,OAAO,EAAE;YACpB,SAAS,EAAE,IAAI,CAAC,YAAY,EAAE;YAC9B,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,KAAK;YACtC,eAAe,EAAE,IAAI,CAAC,iBAAiB,EAAE;YACzC,MAAM,EAAE;gBACN,MAAM,EAAE,IAAI,CAAC,SAAS,EAAE;gBACxB,GAAG,CAAC,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC7D;SACF,CAAC;IACJ,CAAC;CACF"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../../src/scorers/api_scorers/prompt_scorer/index.ts"],"names":[],"mappings":"AAAA,cAAc,sBAAsB,CAAC;AACrC,cAAc,iBAAiB,CAAC;AAChC,cAAc,uBAAuB,CAAC;AACtC,cAAc,uBAAuB,CAAC"}
|
|
@@ -1,44 +0,0 @@
|
|
|
1
|
-
import { JUDGMENT_API_KEY, JUDGMENT_API_URL, JUDGMENT_ORG_ID, } from "../../../env";
|
|
2
|
-
import { JudgmentApiClient } from "../../../internal/api";
|
|
3
|
-
export class JudgmentAPIError extends Error {
|
|
4
|
-
constructor(statusCode, message) {
|
|
5
|
-
super(message);
|
|
6
|
-
this.statusCode = statusCode;
|
|
7
|
-
this.name = "JudgmentAPIError";
|
|
8
|
-
}
|
|
9
|
-
}
|
|
10
|
-
export async function pushPromptScorer(name, prompt, threshold, options, judgmentApiKey = JUDGMENT_API_KEY || "", organizationId = JUDGMENT_ORG_ID || "", isTrace) {
|
|
11
|
-
if (!JUDGMENT_API_URL || !judgmentApiKey || !organizationId) {
|
|
12
|
-
throw new Error("Missing required API credentials");
|
|
13
|
-
}
|
|
14
|
-
const client = new JudgmentApiClient(JUDGMENT_API_URL, judgmentApiKey, organizationId);
|
|
15
|
-
const response = await client.saveScorer({
|
|
16
|
-
name,
|
|
17
|
-
prompt,
|
|
18
|
-
threshold,
|
|
19
|
-
options,
|
|
20
|
-
is_trace: isTrace,
|
|
21
|
-
});
|
|
22
|
-
return response.name;
|
|
23
|
-
}
|
|
24
|
-
export async function fetchPromptScorer(name, judgmentApiKey = JUDGMENT_API_KEY || "", organizationId = JUDGMENT_ORG_ID || "") {
|
|
25
|
-
if (!JUDGMENT_API_URL || !judgmentApiKey || !organizationId) {
|
|
26
|
-
throw new Error("Missing required API credentials");
|
|
27
|
-
}
|
|
28
|
-
const client = new JudgmentApiClient(JUDGMENT_API_URL, judgmentApiKey, organizationId);
|
|
29
|
-
const response = await client.fetchScorers({ names: [name] });
|
|
30
|
-
if (response.scorers.length === 0) {
|
|
31
|
-
throw new JudgmentAPIError(404, `Scorer with name ${name} not found`);
|
|
32
|
-
}
|
|
33
|
-
const { created_at, updated_at, ...config } = response.scorers[0];
|
|
34
|
-
return config;
|
|
35
|
-
}
|
|
36
|
-
export async function scorerExists(name, judgmentApiKey = JUDGMENT_API_KEY || "", organizationId = JUDGMENT_ORG_ID || "") {
|
|
37
|
-
if (!JUDGMENT_API_URL || !judgmentApiKey || !organizationId) {
|
|
38
|
-
throw new Error("Missing required API credentials");
|
|
39
|
-
}
|
|
40
|
-
const client = new JudgmentApiClient(JUDGMENT_API_URL, judgmentApiKey, organizationId);
|
|
41
|
-
const response = await client.scorerExists({ name });
|
|
42
|
-
return response.exists;
|
|
43
|
-
}
|
|
44
|
-
//# sourceMappingURL=prompt-scorer-utils.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"prompt-scorer-utils.js","sourceRoot":"","sources":["../../../../src/scorers/api_scorers/prompt_scorer/prompt-scorer-utils.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,gBAAgB,EAChB,gBAAgB,EAChB,eAAe,GAChB,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,iBAAiB,EAAE,MAAM,uBAAuB,CAAC;AAE1D,MAAM,OAAO,gBAAiB,SAAQ,KAAK;IACzC,YACS,UAAkB,EACzB,OAAe;QAEf,KAAK,CAAC,OAAO,CAAC,CAAC;QAHR,eAAU,GAAV,UAAU,CAAQ;QAIzB,IAAI,CAAC,IAAI,GAAG,kBAAkB,CAAC;IACjC,CAAC;CACF;AAED,MAAM,CAAC,KAAK,UAAU,gBAAgB,CACpC,IAAY,EACZ,MAAc,EACd,SAAiB,EACjB,OAAuC,EACvC,iBAAyB,gBAAgB,IAAI,EAAE,EAC/C,iBAAyB,eAAe,IAAI,EAAE,EAC9C,OAAiB;IAEjB,IAAI,CAAC,gBAAgB,IAAI,CAAC,cAAc,IAAI,CAAC,cAAc,EAAE,CAAC;QAC5D,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC,CAAC;IACtD,CAAC;IAED,MAAM,MAAM,GAAG,IAAI,iBAAiB,CAClC,gBAAgB,EAChB,cAAc,EACd,cAAc,CACf,CAAC;IACF,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,UAAU,CAAC;QACvC,IAAI;QACJ,MAAM;QACN,SAAS;QACT,OAAO;QACP,QAAQ,EAAE,OAAO;KAClB,CAAC,CAAC;IACH,OAAO,QAAQ,CAAC,IAAI,CAAC;AACvB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,IAAY,EACZ,iBAAyB,gBAAgB,IAAI,EAAE,EAC/C,iBAAyB,eAAe,IAAI,EAAE;IAE9C,IAAI,CAAC,gBAAgB,IAAI,CAAC,cAAc,IAAI,CAAC,cAAc,EAAE,CAAC;QAC5D,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC,CAAC;IACtD,CAAC;IAED,MAAM,MAAM,GAAG,IAAI,iBAAiB,CAClC,gBAAgB,EAChB,cAAc,EACd,cAAc,CACf,CAAC;IACF,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,YAAY,CAAC,EAAE,KAAK,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAC9D,IAAI,QAAQ,CAAC,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAClC,MAAM,IAAI,gBAAgB,CAAC,GAAG,EAAE,oBAAoB,IAAI,YAAY,CAAC,CAAC;IACxE,CAAC;IACD,MAAM,EAAE,UAAU,EAAE,UAAU,EAAE,GAAG,MAAM,EAAE,GAAG,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IAClE,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,IAAY,EACZ,iBAAyB,gBAAgB,IAAI,EAAE,EAC/C,iBAAyB,eAAe,IAAI,EAAE;IAE9C,IAAI,CAAC,gBAAgB,IAAI,CAAC,cAAc,IAAI,CAAC,cAAc,EAAE,CAAC;QAC5D,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC,CAAC;IACtD,CAAC;IAED,MAAM,MAAM,GAAG,IAAI,iBAAiB,CAClC,gBAAgB,EAChB,cAAc,EACd,cAAc,CACf,CAAC;IACF,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,YAAY,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC;IACrD,OAAO,QAAQ,CAAC,MAAM,CAAC;AACzB,CAAC"}
|
|
@@ -1,9 +0,0 @@
|
|
|
1
|
-
import { JUDGMENT_API_KEY, JUDGMENT_ORG_ID } from "../../../env";
|
|
2
|
-
import { BasePromptScorer } from "./base-prompt-scorer";
|
|
3
|
-
export { fetchPromptScorer, JudgmentAPIError, pushPromptScorer, scorerExists, } from "./prompt-scorer-utils";
|
|
4
|
-
export class PromptScorer extends BasePromptScorer {
|
|
5
|
-
constructor(scoreType, name, prompt, threshold, requiredParams = [], options, judgmentApiKey = JUDGMENT_API_KEY || "", organizationId = JUDGMENT_ORG_ID || "") {
|
|
6
|
-
super(scoreType, name, prompt, threshold, requiredParams, options, judgmentApiKey, organizationId);
|
|
7
|
-
}
|
|
8
|
-
}
|
|
9
|
-
//# sourceMappingURL=prompt-scorer.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"prompt-scorer.js","sourceRoot":"","sources":["../../../../src/scorers/api_scorers/prompt_scorer/prompt-scorer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,cAAc,CAAC;AAEjE,OAAO,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AACxD,OAAO,EACL,iBAAiB,EACjB,gBAAgB,EAChB,gBAAgB,EAChB,YAAY,GACb,MAAM,uBAAuB,CAAC;AAE/B,MAAM,OAAO,YAAa,SAAQ,gBAAgB;IAChD,YACE,SAAwB,EACxB,IAAY,EACZ,MAAc,EACd,SAAiB,EACjB,iBAAoC,EAAE,EACtC,OAAuC,EACvC,iBAAyB,gBAAgB,IAAI,EAAE,EAC/C,iBAAyB,eAAe,IAAI,EAAE;QAE9C,KAAK,CACH,SAAS,EACT,IAAI,EACJ,MAAM,EACN,SAAS,EACT,cAAc,EACd,OAAO,EACP,cAAc,EACd,cAAc,CACf,CAAC;IACJ,CAAC;CACF"}
|
|
@@ -1,8 +0,0 @@
|
|
|
1
|
-
import { JUDGMENT_API_KEY, JUDGMENT_ORG_ID } from "../../../env";
|
|
2
|
-
import { BasePromptScorer } from "./base-prompt-scorer";
|
|
3
|
-
export class TracePromptScorer extends BasePromptScorer {
|
|
4
|
-
constructor(scoreType, name, prompt, threshold, requiredParams = [], options, judgmentApiKey = JUDGMENT_API_KEY || "", organizationId = JUDGMENT_ORG_ID || "") {
|
|
5
|
-
super(scoreType, name, prompt, threshold, requiredParams, options, judgmentApiKey, organizationId);
|
|
6
|
-
}
|
|
7
|
-
}
|
|
8
|
-
//# sourceMappingURL=trace-prompt-scorer.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"trace-prompt-scorer.js","sourceRoot":"","sources":["../../../../src/scorers/api_scorers/prompt_scorer/trace-prompt-scorer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,cAAc,CAAC;AAEjE,OAAO,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAExD,MAAM,OAAO,iBAAkB,SAAQ,gBAAgB;IACrD,YACE,SAAwB,EACxB,IAAY,EACZ,MAAc,EACd,SAAiB,EACjB,iBAAoC,EAAE,EACtC,OAAuC,EACvC,iBAAyB,gBAAgB,IAAI,EAAE,EAC/C,iBAAyB,eAAe,IAAI,EAAE;QAE9C,KAAK,CACH,SAAS,EACT,IAAI,EACJ,MAAM,EACN,SAAS,EACT,cAAc,EACd,OAAO,EACP,cAAc,EACd,cAAc,CACf,CAAC;IACJ,CAAC;CACF"}
|
|
@@ -1,41 +0,0 @@
|
|
|
1
|
-
export function createBaseScorer() {
|
|
2
|
-
const scorer = {
|
|
3
|
-
score_type: "",
|
|
4
|
-
class_name: "",
|
|
5
|
-
name: "",
|
|
6
|
-
addModel: (model) => {
|
|
7
|
-
scorer.model = model;
|
|
8
|
-
},
|
|
9
|
-
successCheck: () => {
|
|
10
|
-
if (scorer.error != null) {
|
|
11
|
-
return false;
|
|
12
|
-
}
|
|
13
|
-
if (scorer.score == null) {
|
|
14
|
-
return false;
|
|
15
|
-
}
|
|
16
|
-
const threshold = scorer.threshold ?? 0.5;
|
|
17
|
-
const score = scorer.score;
|
|
18
|
-
return threshold != null && score != null && score >= threshold;
|
|
19
|
-
},
|
|
20
|
-
getRequiredParams: () => {
|
|
21
|
-
return [];
|
|
22
|
-
},
|
|
23
|
-
toTransport: () => {
|
|
24
|
-
return {
|
|
25
|
-
score_type: scorer.score_type,
|
|
26
|
-
name: scorer.name,
|
|
27
|
-
threshold: scorer.threshold ?? 0.5,
|
|
28
|
-
strict_mode: scorer.strict_mode ?? false,
|
|
29
|
-
required_params: [],
|
|
30
|
-
kwargs: {},
|
|
31
|
-
};
|
|
32
|
-
},
|
|
33
|
-
};
|
|
34
|
-
scorer.class_name = "BaseScorer";
|
|
35
|
-
scorer.name = scorer.class_name;
|
|
36
|
-
if (scorer.strict_mode === true) {
|
|
37
|
-
scorer.threshold = 1.0;
|
|
38
|
-
}
|
|
39
|
-
return scorer;
|
|
40
|
-
}
|
|
41
|
-
//# sourceMappingURL=base-scorer.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"base-scorer.js","sourceRoot":"","sources":["../../src/scorers/base-scorer.ts"],"names":[],"mappings":"AAUA,MAAM,UAAU,gBAAgB;IAC9B,MAAM,MAAM,GAAe;QACzB,UAAU,EAAE,EAAE;QACd,UAAU,EAAE,EAAE;QACd,IAAI,EAAE,EAAE;QACR,QAAQ,EAAE,CAAC,KAAa,EAAE,EAAE;YAC1B,MAAM,CAAC,KAAK,GAAG,KAAK,CAAC;QACvB,CAAC;QACD,YAAY,EAAE,GAAG,EAAE;YACjB,IAAI,MAAM,CAAC,KAAK,IAAI,IAAI,EAAE,CAAC;gBACzB,OAAO,KAAK,CAAC;YACf,CAAC;YACD,IAAI,MAAM,CAAC,KAAK,IAAI,IAAI,EAAE,CAAC;gBACzB,OAAO,KAAK,CAAC;YACf,CAAC;YACD,MAAM,SAAS,GAAG,MAAM,CAAC,SAAS,IAAI,GAAG,CAAC;YAC1C,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;YAC3B,OAAO,SAAS,IAAI,IAAI,IAAI,KAAK,IAAI,IAAI,IAAI,KAAK,IAAI,SAAS,CAAC;QAClE,CAAC;QACD,iBAAiB,EAAE,GAAG,EAAE;YACtB,OAAO,EAAE,CAAC;QACZ,CAAC;QACD,WAAW,EAAE,GAAG,EAAE;YAChB,OAAO;gBACL,UAAU,EAAE,MAAM,CAAC,UAAU;gBAC7B,IAAI,EAAE,MAAM,CAAC,IAAI;gBACjB,SAAS,EAAE,MAAM,CAAC,SAAS,IAAI,GAAG;gBAClC,WAAW,EAAE,MAAM,CAAC,WAAW,IAAI,KAAK;gBACxC,eAAe,EAAE,EAAE;gBACnB,MAAM,EAAE,EAAE;aACX,CAAC;QACJ,CAAC;KACF,CAAC;IAEF,MAAM,CAAC,UAAU,GAAG,YAAY,CAAC;IACjC,MAAM,CAAC,IAAI,GAAG,MAAM,CAAC,UAAU,CAAC;IAChC,IAAI,MAAM,CAAC,WAAW,KAAK,IAAI,EAAE,CAAC;QAChC,MAAM,CAAC,SAAS,GAAG,GAAG,CAAC;IACzB,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC"}
|
package/dist/scorers/index.js
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/scorers/index.ts"],"names":[],"mappings":"AAAA,cAAc,cAAc,CAAC;AAC7B,cAAc,eAAe,CAAC;AAC9B,cAAc,eAAe,CAAC"}
|
|
@@ -1,12 +0,0 @@
|
|
|
1
|
-
export class OpenTelemetryKeys {
|
|
2
|
-
}
|
|
3
|
-
OpenTelemetryKeys.AttributeKeys = {
|
|
4
|
-
JUDGMENT_SPAN_KIND: "judgment.span_kind",
|
|
5
|
-
JUDGMENT_INPUT: "judgment.input",
|
|
6
|
-
JUDGMENT_OUTPUT: "judgment.output",
|
|
7
|
-
PENDING_TRACE_EVAL: "judgment.pending_trace_eval",
|
|
8
|
-
};
|
|
9
|
-
OpenTelemetryKeys.ResourceKeys = {
|
|
10
|
-
JUDGMENT_PROJECT_ID: "judgment.project_id",
|
|
11
|
-
};
|
|
12
|
-
//# sourceMappingURL=OpenTelemetryKeys.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"OpenTelemetryKeys.js","sourceRoot":"","sources":["../../src/tracer/OpenTelemetryKeys.ts"],"names":[],"mappings":"AAAA,MAAM,OAAO,iBAAiB;;AACL,+BAAa,GAAG;IACrC,kBAAkB,EAAE,oBAAoB;IACxC,cAAc,EAAE,gBAAgB;IAChC,eAAe,EAAE,iBAAiB;IAClC,kBAAkB,EAAE,6BAA6B;CACzC,CAAC;AAEY,8BAAY,GAAG;IACpC,mBAAmB,EAAE,qBAAqB;CAClC,CAAC"}
|
package/dist/tracer/Tracer.js
DELETED
|
@@ -1,235 +0,0 @@
|
|
|
1
|
-
import { trace } from "@opentelemetry/api";
|
|
2
|
-
import { JUDGMENT_DEFAULT_GPT_MODEL } from "../env";
|
|
3
|
-
import { JudgmentApiClient } from "../internal/api";
|
|
4
|
-
import { Logger } from "../utils/logger";
|
|
5
|
-
import { JudgmentSpanExporter, NoOpSpanExporter } from "./exporters";
|
|
6
|
-
import { OpenTelemetryKeys } from "./OpenTelemetryKeys";
|
|
7
|
-
import { TracerConfiguration } from "./TracerConfiguration";
|
|
8
|
-
export class Tracer {
|
|
9
|
-
getConfiguration() {
|
|
10
|
-
return this.configuration;
|
|
11
|
-
}
|
|
12
|
-
getProjectId() {
|
|
13
|
-
return this.projectId;
|
|
14
|
-
}
|
|
15
|
-
getSerializer() {
|
|
16
|
-
return this.serializer;
|
|
17
|
-
}
|
|
18
|
-
constructor(configuration, apiClient, serializer) {
|
|
19
|
-
this.projectId = null;
|
|
20
|
-
this.configuration = configuration;
|
|
21
|
-
this.apiClient = apiClient;
|
|
22
|
-
this.serializer = serializer;
|
|
23
|
-
this.projectIdPromise = this.resolveProjectId();
|
|
24
|
-
}
|
|
25
|
-
async resolveProjectId() {
|
|
26
|
-
try {
|
|
27
|
-
Logger.info(`Resolving project ID for project: ${this.configuration.projectName}`);
|
|
28
|
-
const request = {
|
|
29
|
-
project_name: this.configuration.projectName,
|
|
30
|
-
};
|
|
31
|
-
const response = await this.apiClient.projectsResolve(request);
|
|
32
|
-
this.projectId = response.project_id?.toString() || null;
|
|
33
|
-
if (this.projectId) {
|
|
34
|
-
Logger.info(`Successfully resolved project ID: ${this.projectId}`);
|
|
35
|
-
}
|
|
36
|
-
else {
|
|
37
|
-
Logger.warn(`Project ID not found for project: ${this.configuration.projectName}`);
|
|
38
|
-
}
|
|
39
|
-
return this.projectId;
|
|
40
|
-
}
|
|
41
|
-
catch (error) {
|
|
42
|
-
this.projectId = null;
|
|
43
|
-
return null;
|
|
44
|
-
}
|
|
45
|
-
}
|
|
46
|
-
static builder() {
|
|
47
|
-
return new TracerBuilder();
|
|
48
|
-
}
|
|
49
|
-
static createDefault(projectName) {
|
|
50
|
-
return TracerBuilder.builder()
|
|
51
|
-
.configuration(TracerConfiguration.createDefault(projectName))
|
|
52
|
-
.build();
|
|
53
|
-
}
|
|
54
|
-
static createWithConfiguration(configuration) {
|
|
55
|
-
return TracerBuilder.builder().configuration(configuration).build();
|
|
56
|
-
}
|
|
57
|
-
async getSpanExporter() {
|
|
58
|
-
const projectId = await this.projectIdPromise;
|
|
59
|
-
if (!projectId) {
|
|
60
|
-
Logger.error("Failed to resolve project " +
|
|
61
|
-
this.configuration.projectName +
|
|
62
|
-
", please create it first at https://app.judgmentlabs.ai/org/" +
|
|
63
|
-
this.configuration.organizationId +
|
|
64
|
-
"/projects. Skipping Judgment export.");
|
|
65
|
-
return new NoOpSpanExporter();
|
|
66
|
-
}
|
|
67
|
-
return this.createJudgmentSpanExporter(projectId);
|
|
68
|
-
}
|
|
69
|
-
setSpanKind(kind) {
|
|
70
|
-
const currentSpan = trace.getActiveSpan();
|
|
71
|
-
if (!currentSpan) {
|
|
72
|
-
Logger.warn("No active span found, skipping setSpanKind");
|
|
73
|
-
return;
|
|
74
|
-
}
|
|
75
|
-
if (kind !== null) {
|
|
76
|
-
currentSpan.setAttribute(OpenTelemetryKeys.AttributeKeys.JUDGMENT_SPAN_KIND, kind);
|
|
77
|
-
}
|
|
78
|
-
}
|
|
79
|
-
setAttribute(key, value, type) {
|
|
80
|
-
const currentSpan = trace.getActiveSpan();
|
|
81
|
-
if (!currentSpan) {
|
|
82
|
-
Logger.warn("No active span found, skipping setAttribute");
|
|
83
|
-
return;
|
|
84
|
-
}
|
|
85
|
-
currentSpan.setAttribute(key, this.serializer(value));
|
|
86
|
-
}
|
|
87
|
-
setLLMSpan() {
|
|
88
|
-
this.setSpanKind("llm");
|
|
89
|
-
}
|
|
90
|
-
setToolSpan() {
|
|
91
|
-
this.setSpanKind("tool");
|
|
92
|
-
}
|
|
93
|
-
setGeneralSpan() {
|
|
94
|
-
this.setSpanKind("span");
|
|
95
|
-
}
|
|
96
|
-
setAttributes(attributes) {
|
|
97
|
-
if (!attributes) {
|
|
98
|
-
return;
|
|
99
|
-
}
|
|
100
|
-
const currentSpan = trace.getActiveSpan();
|
|
101
|
-
if (!currentSpan) {
|
|
102
|
-
Logger.warn("No active span found, skipping setAttributes");
|
|
103
|
-
return;
|
|
104
|
-
}
|
|
105
|
-
for (const [key, value] of Object.entries(attributes)) {
|
|
106
|
-
currentSpan.setAttribute(key, this.serializer(value));
|
|
107
|
-
}
|
|
108
|
-
}
|
|
109
|
-
setInput(input, type) {
|
|
110
|
-
this.setAttribute(OpenTelemetryKeys.AttributeKeys.JUDGMENT_INPUT, input);
|
|
111
|
-
}
|
|
112
|
-
setOutput(output, type) {
|
|
113
|
-
this.setAttribute(OpenTelemetryKeys.AttributeKeys.JUDGMENT_OUTPUT, output);
|
|
114
|
-
}
|
|
115
|
-
asyncEvaluate(scorer, example, model) {
|
|
116
|
-
if (!this.configuration.enableEvaluation) {
|
|
117
|
-
return;
|
|
118
|
-
}
|
|
119
|
-
const currentSpan = trace.getActiveSpan();
|
|
120
|
-
if (!currentSpan) {
|
|
121
|
-
Logger.warn("No active span found, skipping asyncEvaluate");
|
|
122
|
-
return;
|
|
123
|
-
}
|
|
124
|
-
if (!currentSpan.isRecording()) {
|
|
125
|
-
Logger.warn("Active span is not recording, skipping asyncEvaluate");
|
|
126
|
-
return;
|
|
127
|
-
}
|
|
128
|
-
const spanContext = currentSpan.spanContext();
|
|
129
|
-
const traceId = spanContext.traceId;
|
|
130
|
-
const spanId = spanContext.spanId;
|
|
131
|
-
Logger.info(`asyncEvaluate: project=${this.configuration.projectName}, traceId=${traceId}, spanId=${spanId}, scorer=${scorer.name}`);
|
|
132
|
-
const evaluationRun = this.createEvaluationRun(scorer, example, model, traceId, spanId);
|
|
133
|
-
this.enqueueEvaluation(evaluationRun);
|
|
134
|
-
}
|
|
135
|
-
asyncTraceEvaluate(scorer, model) {
|
|
136
|
-
if (!this.configuration.enableEvaluation) {
|
|
137
|
-
return;
|
|
138
|
-
}
|
|
139
|
-
const currentSpan = trace.getActiveSpan();
|
|
140
|
-
if (!currentSpan) {
|
|
141
|
-
Logger.warn("No active span found, skipping asyncTraceEvaluate");
|
|
142
|
-
return;
|
|
143
|
-
}
|
|
144
|
-
if (!currentSpan.isRecording()) {
|
|
145
|
-
Logger.warn("Active span is not recording, skipping asyncTraceEvaluate");
|
|
146
|
-
return;
|
|
147
|
-
}
|
|
148
|
-
const spanContext = currentSpan.spanContext();
|
|
149
|
-
const traceId = spanContext.traceId;
|
|
150
|
-
const spanId = spanContext.spanId;
|
|
151
|
-
Logger.info(`asyncTraceEvaluate: project=${this.configuration.projectName}, traceId=${traceId}, spanId=${spanId}, scorer=${scorer.name}`);
|
|
152
|
-
try {
|
|
153
|
-
const traceEvaluationRun = this.createTraceEvaluationRun(scorer, model, traceId, spanId);
|
|
154
|
-
const traceEvalJson = this.serializer(traceEvaluationRun);
|
|
155
|
-
currentSpan.setAttribute(OpenTelemetryKeys.AttributeKeys.PENDING_TRACE_EVAL, traceEvalJson);
|
|
156
|
-
}
|
|
157
|
-
catch (error) {
|
|
158
|
-
Logger.error(`Failed to serialize trace evaluation: ${error instanceof Error ? error.message : String(error)}`);
|
|
159
|
-
}
|
|
160
|
-
}
|
|
161
|
-
createTraceEvaluationRun(scorer, model, traceId, spanId) {
|
|
162
|
-
const evalName = `async_trace_evaluate_${spanId || Date.now()}`;
|
|
163
|
-
const modelName = model || JUDGMENT_DEFAULT_GPT_MODEL;
|
|
164
|
-
const scorerConfig = scorer.toTransport();
|
|
165
|
-
return {
|
|
166
|
-
project_name: this.configuration.projectName,
|
|
167
|
-
eval_name: evalName,
|
|
168
|
-
scorer: scorerConfig,
|
|
169
|
-
model: modelName,
|
|
170
|
-
organization_id: this.configuration.organizationId,
|
|
171
|
-
trace_id: traceId,
|
|
172
|
-
trace_span_id: spanId,
|
|
173
|
-
};
|
|
174
|
-
}
|
|
175
|
-
createJudgmentSpanExporter(projectId) {
|
|
176
|
-
const endpoint = this.configuration.apiUrl.endsWith("/")
|
|
177
|
-
? `${this.configuration.apiUrl}otel/v1/traces`
|
|
178
|
-
: `${this.configuration.apiUrl}/otel/v1/traces`;
|
|
179
|
-
return new JudgmentSpanExporter(endpoint, this.configuration.apiKey, this.configuration.organizationId, projectId);
|
|
180
|
-
}
|
|
181
|
-
createEvaluationRun(scorer, example, model, traceId, spanId) {
|
|
182
|
-
const runId = `async_evaluate_${spanId || Date.now()}`;
|
|
183
|
-
const modelName = model || JUDGMENT_DEFAULT_GPT_MODEL;
|
|
184
|
-
const scorerConfig = scorer.toTransport();
|
|
185
|
-
const evaluationRun = {
|
|
186
|
-
project_name: this.configuration.projectName,
|
|
187
|
-
eval_name: runId,
|
|
188
|
-
examples: [example],
|
|
189
|
-
custom_scorers: [],
|
|
190
|
-
judgment_scorers: [scorerConfig],
|
|
191
|
-
model: modelName,
|
|
192
|
-
trace_id: traceId,
|
|
193
|
-
trace_span_id: spanId,
|
|
194
|
-
};
|
|
195
|
-
return evaluationRun;
|
|
196
|
-
}
|
|
197
|
-
async enqueueEvaluation(evaluationRun) {
|
|
198
|
-
try {
|
|
199
|
-
await this.apiClient.addToRunEvalQueueExamples(evaluationRun);
|
|
200
|
-
Logger.info(`Enqueuing evaluation run: ${evaluationRun.eval_name}`);
|
|
201
|
-
}
|
|
202
|
-
catch (error) {
|
|
203
|
-
Logger.error(`Failed to enqueue evaluation run: ${error instanceof Error ? error.message : String(error)}`);
|
|
204
|
-
}
|
|
205
|
-
}
|
|
206
|
-
}
|
|
207
|
-
export class TracerBuilder {
|
|
208
|
-
constructor() {
|
|
209
|
-
this._serializer = JSON.stringify;
|
|
210
|
-
}
|
|
211
|
-
static builder() {
|
|
212
|
-
return new TracerBuilder();
|
|
213
|
-
}
|
|
214
|
-
configuration(configuration) {
|
|
215
|
-
this.config = configuration;
|
|
216
|
-
return this;
|
|
217
|
-
}
|
|
218
|
-
apiClient(apiClient) {
|
|
219
|
-
this._apiClient = apiClient;
|
|
220
|
-
return this;
|
|
221
|
-
}
|
|
222
|
-
serializer(serializer) {
|
|
223
|
-
this._serializer = serializer;
|
|
224
|
-
return this;
|
|
225
|
-
}
|
|
226
|
-
build() {
|
|
227
|
-
if (!this.config) {
|
|
228
|
-
throw new Error("Configuration is required");
|
|
229
|
-
}
|
|
230
|
-
const client = this._apiClient ||
|
|
231
|
-
new JudgmentApiClient(this.config.apiUrl, this.config.apiKey, this.config.organizationId);
|
|
232
|
-
return new Tracer(this.config, client, this._serializer);
|
|
233
|
-
}
|
|
234
|
-
}
|
|
235
|
-
//# sourceMappingURL=Tracer.js.map
|