langsmith 0.3.9-rc.2 → 0.3.9-rc.4
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/utils/jestlike/index.cjs +2 -2
- package/dist/utils/jestlike/index.js +2 -2
- package/dist/utils/jestlike/types.d.ts +1 -1
- package/dist/utils/jestlike/vendor/evaluatedBy.cjs +17 -9
- package/dist/utils/jestlike/vendor/evaluatedBy.d.ts +3 -3
- package/dist/utils/jestlike/vendor/evaluatedBy.js +17 -9
- package/package.json +1 -1
|
@@ -372,7 +372,7 @@ function generateWrapperFromJestlikeMethods(methods, testRunnerName) {
|
|
|
372
372
|
}
|
|
373
373
|
const { dataset, createdAt, project, client, experimentUrl } = datasetSetupInfo.get(context.suiteUuid);
|
|
374
374
|
const testInput = inputs;
|
|
375
|
-
const testOutput = referenceOutputs;
|
|
375
|
+
const testOutput = referenceOutputs ?? {};
|
|
376
376
|
const testFeedback = [];
|
|
377
377
|
const onFeedbackLogged = (feedback) => testFeedback.push(feedback);
|
|
378
378
|
let loggedOutput;
|
|
@@ -459,7 +459,7 @@ function generateWrapperFromJestlikeMethods(methods, testRunnerName) {
|
|
|
459
459
|
exampleId,
|
|
460
460
|
datasetId: dataset.id,
|
|
461
461
|
inputs,
|
|
462
|
-
outputs: referenceOutputs,
|
|
462
|
+
outputs: referenceOutputs ?? {},
|
|
463
463
|
metadata: {},
|
|
464
464
|
createdAt,
|
|
465
465
|
}));
|
|
@@ -337,7 +337,7 @@ export function generateWrapperFromJestlikeMethods(methods, testRunnerName) {
|
|
|
337
337
|
}
|
|
338
338
|
const { dataset, createdAt, project, client, experimentUrl } = datasetSetupInfo.get(context.suiteUuid);
|
|
339
339
|
const testInput = inputs;
|
|
340
|
-
const testOutput = referenceOutputs;
|
|
340
|
+
const testOutput = referenceOutputs ?? {};
|
|
341
341
|
const testFeedback = [];
|
|
342
342
|
const onFeedbackLogged = (feedback) => testFeedback.push(feedback);
|
|
343
343
|
let loggedOutput;
|
|
@@ -424,7 +424,7 @@ export function generateWrapperFromJestlikeMethods(methods, testRunnerName) {
|
|
|
424
424
|
exampleId,
|
|
425
425
|
datasetId: dataset.id,
|
|
426
426
|
inputs,
|
|
427
|
-
outputs: referenceOutputs,
|
|
427
|
+
outputs: referenceOutputs ?? {},
|
|
428
428
|
metadata: {},
|
|
429
429
|
createdAt,
|
|
430
430
|
}));
|
|
@@ -10,7 +10,7 @@ export type LangSmithJestlikeWrapperConfig = Partial<Omit<RunTreeConfig, "client
|
|
|
10
10
|
};
|
|
11
11
|
export type LangSmithJestlikeWrapperParams<I, O> = {
|
|
12
12
|
inputs: I;
|
|
13
|
-
referenceOutputs
|
|
13
|
+
referenceOutputs?: O;
|
|
14
14
|
config?: LangSmithJestlikeWrapperConfig;
|
|
15
15
|
};
|
|
16
16
|
export type LangSmithJestlikeDescribeWrapperConfig = {
|
|
@@ -43,15 +43,20 @@ function wrapEvaluator(evaluator) {
|
|
|
43
43
|
else {
|
|
44
44
|
evalResult = await evaluator(input);
|
|
45
45
|
}
|
|
46
|
-
if (
|
|
47
|
-
|
|
48
|
-
|
|
49
|
-
|
|
50
|
-
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
|
|
54
|
-
|
|
46
|
+
if (!Array.isArray(evalResult)) {
|
|
47
|
+
evalResult = [evalResult];
|
|
48
|
+
}
|
|
49
|
+
for (const result of evalResult) {
|
|
50
|
+
if (isEvaluationResult(result)) {
|
|
51
|
+
(0, globals_js_1._logTestFeedback)({
|
|
52
|
+
exampleId: context?.currentExample?.id,
|
|
53
|
+
feedback: result,
|
|
54
|
+
context,
|
|
55
|
+
runTree: currentRunTree,
|
|
56
|
+
client: context.client,
|
|
57
|
+
sourceRunId: evalRunId,
|
|
58
|
+
});
|
|
59
|
+
}
|
|
55
60
|
}
|
|
56
61
|
return evalResult;
|
|
57
62
|
};
|
|
@@ -73,6 +78,9 @@ async function evaluatedBy(outputs, evaluator) {
|
|
|
73
78
|
referenceOutputs: context?.currentExample?.outputs ?? {},
|
|
74
79
|
outputs,
|
|
75
80
|
}, { runId: evalRunId });
|
|
81
|
+
if (Array.isArray(evalResult)) {
|
|
82
|
+
return evalResult.map((result) => result.score);
|
|
83
|
+
}
|
|
76
84
|
return evalResult.score;
|
|
77
85
|
}
|
|
78
86
|
exports.evaluatedBy = evaluatedBy;
|
|
@@ -6,7 +6,7 @@ export type SimpleEvaluatorParams = {
|
|
|
6
6
|
outputs: Record<string, any>;
|
|
7
7
|
};
|
|
8
8
|
export type SimpleEvaluator = (params: SimpleEvaluatorParams) => SimpleEvaluationResult | Promise<SimpleEvaluationResult>;
|
|
9
|
-
export declare function wrapEvaluator<I>(evaluator: (input: I) => SimpleEvaluationResult | Promise<SimpleEvaluationResult>): (input: I, config?: Partial<RunTreeConfig> & {
|
|
9
|
+
export declare function wrapEvaluator<I>(evaluator: (input: I) => SimpleEvaluationResult | SimpleEvaluationResult[] | Promise<SimpleEvaluationResult | SimpleEvaluationResult[]>): (input: I, config?: Partial<RunTreeConfig> & {
|
|
10
10
|
runId?: string;
|
|
11
|
-
}) => Promise<SimpleEvaluationResult>;
|
|
12
|
-
export declare function evaluatedBy(outputs: any, evaluator: SimpleEvaluator): Promise<NonNullable<import("../../../schemas.js").ScoreType | undefined
|
|
11
|
+
}) => Promise<SimpleEvaluationResult | SimpleEvaluationResult[]>;
|
|
12
|
+
export declare function evaluatedBy(outputs: any, evaluator: SimpleEvaluator): Promise<NonNullable<import("../../../schemas.js").ScoreType | undefined> | NonNullable<import("../../../schemas.js").ScoreType | undefined>[]>;
|
|
@@ -40,15 +40,20 @@ export function wrapEvaluator(evaluator) {
|
|
|
40
40
|
else {
|
|
41
41
|
evalResult = await evaluator(input);
|
|
42
42
|
}
|
|
43
|
-
if (
|
|
44
|
-
|
|
45
|
-
|
|
46
|
-
|
|
47
|
-
|
|
48
|
-
|
|
49
|
-
|
|
50
|
-
|
|
51
|
-
|
|
43
|
+
if (!Array.isArray(evalResult)) {
|
|
44
|
+
evalResult = [evalResult];
|
|
45
|
+
}
|
|
46
|
+
for (const result of evalResult) {
|
|
47
|
+
if (isEvaluationResult(result)) {
|
|
48
|
+
_logTestFeedback({
|
|
49
|
+
exampleId: context?.currentExample?.id,
|
|
50
|
+
feedback: result,
|
|
51
|
+
context,
|
|
52
|
+
runTree: currentRunTree,
|
|
53
|
+
client: context.client,
|
|
54
|
+
sourceRunId: evalRunId,
|
|
55
|
+
});
|
|
56
|
+
}
|
|
52
57
|
}
|
|
53
58
|
return evalResult;
|
|
54
59
|
};
|
|
@@ -69,5 +74,8 @@ export async function evaluatedBy(outputs, evaluator) {
|
|
|
69
74
|
referenceOutputs: context?.currentExample?.outputs ?? {},
|
|
70
75
|
outputs,
|
|
71
76
|
}, { runId: evalRunId });
|
|
77
|
+
if (Array.isArray(evalResult)) {
|
|
78
|
+
return evalResult.map((result) => result.score);
|
|
79
|
+
}
|
|
72
80
|
return evalResult.score;
|
|
73
81
|
}
|