@arizeai/phoenix-evals 2.4.0 → 2.6.0
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 +262 -0
- package/README.md +6 -5
- package/dist/esm/__generated__/default_templates/COMPLETENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts +3 -0
- package/dist/esm/__generated__/default_templates/COMPLETENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -0
- package/dist/esm/__generated__/default_templates/COMPLETENESS_CLASSIFICATION_EVALUATOR_CONFIG.js +93 -0
- package/dist/esm/__generated__/default_templates/COMPLETENESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -0
- package/dist/esm/__generated__/default_templates/index.d.ts +1 -0
- package/dist/esm/__generated__/default_templates/index.d.ts.map +1 -1
- package/dist/esm/__generated__/default_templates/index.js +1 -0
- package/dist/esm/__generated__/default_templates/index.js.map +1 -1
- package/dist/esm/llm/ClassificationEvaluator.d.ts +2 -1
- package/dist/esm/llm/ClassificationEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/ClassificationEvaluator.js.map +1 -1
- package/dist/esm/llm/createCompletenessEvaluator.d.ts +46 -0
- package/dist/esm/llm/createCompletenessEvaluator.d.ts.map +1 -0
- package/dist/esm/llm/createCompletenessEvaluator.js +38 -0
- package/dist/esm/llm/createCompletenessEvaluator.js.map +1 -0
- package/dist/esm/llm/createDocumentRelevanceEvaluator.d.ts +16 -11
- package/dist/esm/llm/createDocumentRelevanceEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createDocumentRelevanceEvaluator.js +7 -11
- package/dist/esm/llm/createDocumentRelevanceEvaluator.js.map +1 -1
- package/dist/esm/llm/createRetrievalRelevanceEvaluator.d.ts +4 -5
- package/dist/esm/llm/createRetrievalRelevanceEvaluator.d.ts.map +1 -1
- package/dist/esm/llm/createRetrievalRelevanceEvaluator.js +4 -5
- package/dist/esm/llm/createRetrievalRelevanceEvaluator.js.map +1 -1
- package/dist/esm/llm/generateClassification.d.ts +6 -0
- package/dist/esm/llm/generateClassification.d.ts.map +1 -1
- package/dist/esm/llm/generateClassification.js +66 -1
- package/dist/esm/llm/generateClassification.js.map +1 -1
- package/dist/esm/llm/index.d.ts +1 -0
- package/dist/esm/llm/index.d.ts.map +1 -1
- package/dist/esm/llm/index.js +1 -0
- package/dist/esm/llm/index.js.map +1 -1
- package/dist/esm/tsconfig.esm.tsbuildinfo +1 -1
- package/dist/esm/types/evals.d.ts +14 -2
- package/dist/esm/types/evals.d.ts.map +1 -1
- package/dist/esm/utils/bindEvaluator.d.ts +4 -5
- package/dist/esm/utils/bindEvaluator.d.ts.map +1 -1
- package/dist/esm/utils/bindEvaluator.js +4 -5
- package/dist/esm/utils/bindEvaluator.js.map +1 -1
- package/dist/esm/utils/isEvaluationModel.d.ts +13 -0
- package/dist/esm/utils/isEvaluationModel.d.ts.map +1 -0
- package/dist/esm/utils/isEvaluationModel.js +11 -0
- package/dist/esm/utils/isEvaluationModel.js.map +1 -0
- package/dist/src/__generated__/default_templates/COMPLETENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts +3 -0
- package/dist/src/__generated__/default_templates/COMPLETENESS_CLASSIFICATION_EVALUATOR_CONFIG.d.ts.map +1 -0
- package/dist/src/__generated__/default_templates/COMPLETENESS_CLASSIFICATION_EVALUATOR_CONFIG.js +96 -0
- package/dist/src/__generated__/default_templates/COMPLETENESS_CLASSIFICATION_EVALUATOR_CONFIG.js.map +1 -0
- package/dist/src/__generated__/default_templates/index.d.ts +1 -0
- package/dist/src/__generated__/default_templates/index.d.ts.map +1 -1
- package/dist/src/__generated__/default_templates/index.js +3 -1
- package/dist/src/__generated__/default_templates/index.js.map +1 -1
- package/dist/src/llm/ClassificationEvaluator.d.ts +2 -1
- package/dist/src/llm/ClassificationEvaluator.d.ts.map +1 -1
- package/dist/src/llm/ClassificationEvaluator.js.map +1 -1
- package/dist/src/llm/createCompletenessEvaluator.d.ts +46 -0
- package/dist/src/llm/createCompletenessEvaluator.d.ts.map +1 -0
- package/dist/src/llm/createCompletenessEvaluator.js +49 -0
- package/dist/src/llm/createCompletenessEvaluator.js.map +1 -0
- package/dist/src/llm/createDocumentRelevanceEvaluator.d.ts +16 -11
- package/dist/src/llm/createDocumentRelevanceEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createDocumentRelevanceEvaluator.js +7 -11
- package/dist/src/llm/createDocumentRelevanceEvaluator.js.map +1 -1
- package/dist/src/llm/createRetrievalRelevanceEvaluator.d.ts +4 -5
- package/dist/src/llm/createRetrievalRelevanceEvaluator.d.ts.map +1 -1
- package/dist/src/llm/createRetrievalRelevanceEvaluator.js +4 -5
- package/dist/src/llm/createRetrievalRelevanceEvaluator.js.map +1 -1
- package/dist/src/llm/generateClassification.d.ts +6 -0
- package/dist/src/llm/generateClassification.d.ts.map +1 -1
- package/dist/src/llm/generateClassification.js +65 -0
- package/dist/src/llm/generateClassification.js.map +1 -1
- package/dist/src/llm/index.d.ts +1 -0
- package/dist/src/llm/index.d.ts.map +1 -1
- package/dist/src/llm/index.js +1 -0
- package/dist/src/llm/index.js.map +1 -1
- package/dist/src/types/evals.d.ts +14 -2
- package/dist/src/types/evals.d.ts.map +1 -1
- package/dist/src/utils/bindEvaluator.d.ts +4 -5
- package/dist/src/utils/bindEvaluator.d.ts.map +1 -1
- package/dist/src/utils/bindEvaluator.js +4 -5
- package/dist/src/utils/bindEvaluator.js.map +1 -1
- package/dist/src/utils/isEvaluationModel.d.ts +13 -0
- package/dist/src/utils/isEvaluationModel.d.ts.map +1 -0
- package/dist/src/utils/isEvaluationModel.js +14 -0
- package/dist/src/utils/isEvaluationModel.js.map +1 -0
- package/dist/tsconfig.tsbuildinfo +1 -1
- package/package.json +13 -12
- package/src/__generated__/default_templates/COMPLETENESS_CLASSIFICATION_EVALUATOR_CONFIG.ts +95 -0
- package/src/__generated__/default_templates/index.ts +1 -0
- package/src/llm/ClassificationEvaluator.ts +2 -1
- package/src/llm/createCompletenessEvaluator.ts +74 -0
- package/src/llm/createDocumentRelevanceEvaluator.ts +16 -11
- package/src/llm/createRetrievalRelevanceEvaluator.ts +4 -5
- package/src/llm/generateClassification.ts +78 -1
- package/src/llm/index.ts +1 -0
- package/src/types/evals.ts +15 -3
- package/src/utils/bindEvaluator.ts +4 -5
- package/src/utils/isEvaluationModel.ts +23 -0
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"generateClassification.d.ts","sourceRoot":"","sources":["../../../src/llm/generateClassification.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,oBAAoB,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AACpE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AACnD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;
|
|
1
|
+
{"version":3,"file":"generateClassification.d.ts","sourceRoot":"","sources":["../../../src/llm/generateClassification.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,oBAAoB,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AACpE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AACnD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAMnD,MAAM,MAAM,YAAY,GAAG,OAAO,GAChC,aAAa,GACb,UAAU,GAAG;IACX;;OAEG;IACH,MAAM,EAAE,CAAC,MAAM,EAAE,GAAG,MAAM,EAAE,CAAC,CAAC;IAC9B;;OAEG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;;OAEG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC5B,CAAC;AACJ;;;;;;;;GAQG;AACH,wBAAsB,sBAAsB,CAC1C,IAAI,EAAE,YAAY,GACjB,OAAO,CAAC,oBAAoB,CAAC,CAgC/B"}
|
|
@@ -15,12 +15,22 @@ exports.generateClassification = generateClassification;
|
|
|
15
15
|
const ai_1 = require("ai");
|
|
16
16
|
const zod_1 = require("zod");
|
|
17
17
|
const telemetry_1 = require("../telemetry");
|
|
18
|
+
const isEvaluationModel_1 = require("../utils/isEvaluationModel");
|
|
18
19
|
/**
|
|
19
20
|
* A function that leverages an llm to perform a classification
|
|
21
|
+
*
|
|
22
|
+
* When given an AI SDK evaluation model (e.g. TypeSafe's Jev) the
|
|
23
|
+
* classification is routed through `experimental_evaluate` as a single
|
|
24
|
+
* choice question. Evaluation models cannot generate text, so no explanation
|
|
25
|
+
* is returned in that case. `telemetry`, `schemaName` and `schemaDescription`
|
|
26
|
+
* are ignored on that path: `experimental_evaluate` does not emit spans yet.
|
|
20
27
|
*/
|
|
21
28
|
async function generateClassification(args) {
|
|
22
29
|
var _a, _b;
|
|
23
30
|
const { labels, model, schemaName, schemaDescription, telemetry } = args, prompt = __rest(args, ["labels", "model", "schemaName", "schemaDescription", "telemetry"]);
|
|
31
|
+
if ((0, isEvaluationModel_1.isEvaluationModel)(model)) {
|
|
32
|
+
return evaluateClassification({ model, labels, prompt });
|
|
33
|
+
}
|
|
24
34
|
const telemetryOptions = {
|
|
25
35
|
isEnabled: (_a = telemetry === null || telemetry === void 0 ? void 0 : telemetry.isEnabled) !== null && _a !== void 0 ? _a : true,
|
|
26
36
|
functionId: "generateClassification",
|
|
@@ -40,4 +50,59 @@ async function generateClassification(args) {
|
|
|
40
50
|
explanation: result.object.explanation,
|
|
41
51
|
};
|
|
42
52
|
}
|
|
53
|
+
/**
|
|
54
|
+
* Classify using an AI SDK evaluation model. The rendered prompt becomes the
|
|
55
|
+
* shared evaluation state and the labels become the options of one choice
|
|
56
|
+
* question. The model's label probabilities and resolved model id are
|
|
57
|
+
* returned as metadata.
|
|
58
|
+
*/
|
|
59
|
+
async function evaluateClassification(args) {
|
|
60
|
+
const { model, labels, prompt } = args;
|
|
61
|
+
const result = await (0, ai_1.experimental_evaluate)({
|
|
62
|
+
model,
|
|
63
|
+
state: toEvaluationState(prompt),
|
|
64
|
+
questions: {
|
|
65
|
+
label: {
|
|
66
|
+
type: "choice",
|
|
67
|
+
instructions: "Read the prompt in the state and answer it by selecting exactly one of the choices.",
|
|
68
|
+
criteria: Object.fromEntries(labels.map((label) => [label, null])),
|
|
69
|
+
},
|
|
70
|
+
},
|
|
71
|
+
});
|
|
72
|
+
const answer = result.answers.label;
|
|
73
|
+
return {
|
|
74
|
+
label: answer.choice,
|
|
75
|
+
metadata: {
|
|
76
|
+
probabilities: answer.probabilities,
|
|
77
|
+
modelId: result.response.modelId,
|
|
78
|
+
},
|
|
79
|
+
};
|
|
80
|
+
}
|
|
81
|
+
/**
|
|
82
|
+
* Convert an AI SDK prompt into JSON-compatible evaluation state.
|
|
83
|
+
* A bare text prompt is passed through as a string; anything else
|
|
84
|
+
* (instructions, system messages, message arrays) is serialized to JSON.
|
|
85
|
+
*/
|
|
86
|
+
function toEvaluationState(prompt) {
|
|
87
|
+
const { instructions, system, messages, prompt: promptContent } = prompt;
|
|
88
|
+
if (typeof promptContent === "string" &&
|
|
89
|
+
instructions === undefined &&
|
|
90
|
+
system === undefined &&
|
|
91
|
+
messages === undefined) {
|
|
92
|
+
return promptContent;
|
|
93
|
+
}
|
|
94
|
+
const state = {};
|
|
95
|
+
if (instructions !== undefined)
|
|
96
|
+
state.instructions = instructions;
|
|
97
|
+
if (system !== undefined)
|
|
98
|
+
state.system = system;
|
|
99
|
+
if (promptContent !== undefined)
|
|
100
|
+
state.prompt = promptContent;
|
|
101
|
+
if (messages !== undefined)
|
|
102
|
+
state.messages = messages;
|
|
103
|
+
// Round-trip through JSON so the state only contains JSON values
|
|
104
|
+
// (strips `undefined`, turns Date/URL into strings). Binary file parts
|
|
105
|
+
// are not supported by evaluation models and are not stripped here.
|
|
106
|
+
return JSON.parse(JSON.stringify(state));
|
|
107
|
+
}
|
|
43
108
|
//# sourceMappingURL=generateClassification.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"generateClassification.js","sourceRoot":"","sources":["../../../src/llm/generateClassification.ts"],"names":[],"mappings":";;;;;;;;;;;;;;AAAA,
|
|
1
|
+
{"version":3,"file":"generateClassification.js","sourceRoot":"","sources":["../../../src/llm/generateClassification.ts"],"names":[],"mappings":";;;;;;;;;;;;;;AAAA,2BAA2D;AAC3D,6BAAwB;AAExB,4CAAgE;AAIhE,kEAGoC;AAkBpC;;;;;;;;GAQG;AACI,KAAK,iCACV,IAAkB;;IAElB,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,iBAAiB,EAAE,SAAS,KAC7D,IAAI,EAD8D,MAAM,UACxE,IAAI,EADA,mEAAsE,CACtE,CAAC;IAEP,IAAI,IAAA,qCAAiB,EAAC,KAAK,CAAC,EAAE,CAAC;QAC7B,OAAO,sBAAsB,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;IAC3D,CAAC;IAED,MAAM,gBAAgB,GAAG;QACvB,SAAS,QAAE,SAAS,aAAT,SAAS,uBAAT,SAAS,CAAE,SAAS,mCAAI,IAAI;QACvC,UAAU,EAAE,wBAAwB;QACpC,YAAY,EAAE,IAAA,oCAAwB,QAAC,SAAS,aAAT,SAAS,uBAAT,SAAS,CAAE,MAAM,mCAAI,kBAAM,CAAC;KACpE,CAAC;IAEF,MAAM,MAAM,GAAG,MAAM,IAAA,mBAAc,kBACjC,KAAK;QACL,UAAU;QACV,iBAAiB,EACjB,MAAM,EAAE,OAAC,CAAC,MAAM,CAAC;YACf,WAAW,EAAE,OAAC,CAAC,MAAM,EAAE,EAAE,4EAA4E;YACrG,KAAK,EAAE,OAAC,CAAC,IAAI,CAAC,MAAM,CAAC;SACtB,CAAC,EACF,SAAS,EAAE,gBAAgB;QAC3B,sEAAsE;QACtE,qEAAqE;QACrE,qBAAqB,EAAE,IAAI,IACxB,MAAM,EACT,CAAC;IACH,OAAO;QACL,KAAK,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK;QAC1B,WAAW,EAAE,MAAM,CAAC,MAAM,CAAC,WAAW;KACvC,CAAC;AACJ,CAAC;AAID;;;;;GAKG;AACH,KAAK,UAAU,sBAAsB,CAAC,IAIrC;IACC,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,GAAG,IAAI,CAAC;IACvC,MAAM,MAAM,GAAG,MAAM,IAAA,0BAAqB,EAAC;QACzC,KAAK;QACL,KAAK,EAAE,iBAAiB,CAAC,MAAM,CAAC;QAChC,SAAS,EAAE;YACT,KAAK,EAAE;gBACL,IAAI,EAAE,QAAQ;gBACd,YAAY,EACV,qFAAqF;gBACvF,QAAQ,EAAE,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,CAAC;aACnE;SACF;KACF,CAAC,CAAC;IACH,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC;IACpC,OAAO;QACL,KAAK,EAAE,MAAM,CAAC,MAAM;QACpB,QAAQ,EAAE;YACR,aAAa,EAAE,MAAM,CAAC,aAAa;YACnC,OAAO,EAAE,MAAM,CAAC,QAAQ,CAAC,OAAO;SACjC;KACF,CAAC;AACJ,CAAC;AAED;;;;GAIG;AACH,SAAS,iBAAiB,CAAC,MAAkB;IAC3C,MAAM,EAAE,YAAY,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,aAAa,EAAE,GAAG,MAAM,CAAC;IACzE,IACE,OAAO,aAAa,KAAK,QAAQ;QACjC,YAAY,KAAK,SAAS;QAC1B,MAAM,KAAK,SAAS;QACpB,QAAQ,KAAK,SAAS,EACtB,CAAC;QACD,OAAO,aAAa,CAAC;IACvB,CAAC;IACD,MAAM,KAAK,GAA4B,EAAE,CAAC;IAC1C,IAAI,YAAY,KAAK,SAAS;QAAE,KAAK,CAAC,YAAY,GAAG,YAAY,CAAC;IAClE,IAAI,MAAM,KAAK,SAAS;QAAE,KAAK,CAAC,MAAM,GAAG,MAAM,CAAC;IAChD,IAAI,aAAa,KAAK,SAAS;QAAE,KAAK,CAAC,MAAM,GAAG,aAAa,CAAC;IAC9D,IAAI,QAAQ,KAAK,SAAS;QAAE,KAAK,CAAC,QAAQ,GAAG,QAAQ,CAAC;IACtD,iEAAiE;IACjE,uEAAuE;IACvE,oEAAoE;IACpE,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAoB,CAAC;AAC9D,CAAC"}
|
package/dist/src/llm/index.d.ts
CHANGED
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
export * from "./ClassificationEvaluator";
|
|
2
2
|
export * from "./createClassificationEvaluator";
|
|
3
3
|
export * from "./createClassifierFn";
|
|
4
|
+
export * from "./createCompletenessEvaluator";
|
|
4
5
|
export * from "./createConcisenessEvaluator";
|
|
5
6
|
export * from "./createCorrectnessEvaluator";
|
|
6
7
|
export * from "./createDocumentRelevanceEvaluator";
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/llm/index.ts"],"names":[],"mappings":"AAAA,cAAc,2BAA2B,CAAC;AAC1C,cAAc,iCAAiC,CAAC;AAChD,cAAc,sBAAsB,CAAC;AACrC,cAAc,8BAA8B,CAAC;AAC7C,cAAc,8BAA8B,CAAC;AAC7C,cAAc,oCAAoC,CAAC;AACnD,cAAc,+BAA+B,CAAC;AAC9C,cAAc,gCAAgC,CAAC;AAC/C,cAAc,+BAA+B,CAAC;AAC9C,cAAc,0BAA0B,CAAC;AACzC,cAAc,qCAAqC,CAAC;AACpD,cAAc,iCAAiC,CAAC;AAChD,cAAc,uCAAuC,CAAC;AACtD,cAAc,gCAAgC,CAAC;AAC/C,cAAc,2BAA2B,CAAC;AAC1C,cAAc,+BAA+B,CAAC;AAC9C,cAAc,0BAA0B,CAAC;AACzC,cAAc,gBAAgB,CAAC"}
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/llm/index.ts"],"names":[],"mappings":"AAAA,cAAc,2BAA2B,CAAC;AAC1C,cAAc,iCAAiC,CAAC;AAChD,cAAc,sBAAsB,CAAC;AACrC,cAAc,+BAA+B,CAAC;AAC9C,cAAc,8BAA8B,CAAC;AAC7C,cAAc,8BAA8B,CAAC;AAC7C,cAAc,oCAAoC,CAAC;AACnD,cAAc,+BAA+B,CAAC;AAC9C,cAAc,gCAAgC,CAAC;AAC/C,cAAc,+BAA+B,CAAC;AAC9C,cAAc,0BAA0B,CAAC;AACzC,cAAc,qCAAqC,CAAC;AACpD,cAAc,iCAAiC,CAAC;AAChD,cAAc,uCAAuC,CAAC;AACtD,cAAc,gCAAgC,CAAC;AAC/C,cAAc,2BAA2B,CAAC;AAC1C,cAAc,+BAA+B,CAAC;AAC9C,cAAc,0BAA0B,CAAC;AACzC,cAAc,gBAAgB,CAAC"}
|
package/dist/src/llm/index.js
CHANGED
|
@@ -17,6 +17,7 @@ Object.defineProperty(exports, "__esModule", { value: true });
|
|
|
17
17
|
__exportStar(require("./ClassificationEvaluator"), exports);
|
|
18
18
|
__exportStar(require("./createClassificationEvaluator"), exports);
|
|
19
19
|
__exportStar(require("./createClassifierFn"), exports);
|
|
20
|
+
__exportStar(require("./createCompletenessEvaluator"), exports);
|
|
20
21
|
__exportStar(require("./createConcisenessEvaluator"), exports);
|
|
21
22
|
__exportStar(require("./createCorrectnessEvaluator"), exports);
|
|
22
23
|
__exportStar(require("./createDocumentRelevanceEvaluator"), exports);
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/llm/index.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;AAAA,4DAA0C;AAC1C,kEAAgD;AAChD,uDAAqC;AACrC,+DAA6C;AAC7C,+DAA6C;AAC7C,qEAAmD;AACnD,gEAA8C;AAC9C,iEAA+C;AAC/C,gEAA8C;AAC9C,2DAAyC;AACzC,sEAAoD;AACpD,kEAAgD;AAChD,wEAAsD;AACtD,iEAA+C;AAC/C,4DAA0C;AAC1C,gEAA8C;AAC9C,2DAAyC;AACzC,iDAA+B"}
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/llm/index.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;AAAA,4DAA0C;AAC1C,kEAAgD;AAChD,uDAAqC;AACrC,gEAA8C;AAC9C,+DAA6C;AAC7C,+DAA6C;AAC7C,qEAAmD;AACnD,gEAA8C;AAC9C,iEAA+C;AAC/C,gEAA8C;AAC9C,2DAAyC;AACzC,sEAAoD;AACpD,kEAAgD;AAChD,wEAAsD;AACtD,iEAA+C;AAC/C,4DAA0C;AAC1C,gEAA8C;AAC9C,2DAAyC;AACzC,iDAA+B"}
|
|
@@ -1,4 +1,5 @@
|
|
|
1
1
|
import type { LanguageModel } from "ai";
|
|
2
|
+
import type { EvaluationModel } from "../utils/isEvaluationModel";
|
|
2
3
|
import type { ObjectMapping } from "./data";
|
|
3
4
|
import type { WithTelemetry } from "./otel";
|
|
4
5
|
import type { PromptTemplate } from "./templating";
|
|
@@ -12,7 +13,12 @@ export interface ExampleRecord<OutputType, InputType> {
|
|
|
12
13
|
[key: string]: unknown;
|
|
13
14
|
}
|
|
14
15
|
export interface WithLLM {
|
|
15
|
-
|
|
16
|
+
/**
|
|
17
|
+
* Either a regular AI SDK {@link LanguageModel}, or an AI SDK
|
|
18
|
+
* {@link EvaluationModel} (e.g. TypeSafe's Jev) which classifies without
|
|
19
|
+
* generating text and therefore cannot produce an explanation.
|
|
20
|
+
*/
|
|
21
|
+
model: LanguageModel | EvaluationModel;
|
|
16
22
|
}
|
|
17
23
|
export interface LLMEvaluationArgs extends WithLLM {
|
|
18
24
|
}
|
|
@@ -35,6 +41,11 @@ export interface EvaluationResult {
|
|
|
35
41
|
* @example "The model correctly identified the sentiment of the text."
|
|
36
42
|
*/
|
|
37
43
|
explanation?: string;
|
|
44
|
+
/**
|
|
45
|
+
* Additional information about how the result was produced, e.g. the
|
|
46
|
+
* label probabilities reported by an evaluation model.
|
|
47
|
+
*/
|
|
48
|
+
metadata?: Record<string, unknown>;
|
|
38
49
|
}
|
|
39
50
|
/**
|
|
40
51
|
* The result of a classification
|
|
@@ -42,6 +53,7 @@ export interface EvaluationResult {
|
|
|
42
53
|
export interface ClassificationResult {
|
|
43
54
|
label: string;
|
|
44
55
|
explanation?: string;
|
|
56
|
+
metadata?: Record<string, unknown>;
|
|
45
57
|
}
|
|
46
58
|
/**
|
|
47
59
|
* The choice (e.g. the label and score mapping) of a classification based evaluation
|
|
@@ -58,7 +70,7 @@ export type ClassificationChoicesMap = Record<string, number>;
|
|
|
58
70
|
* The arguments for creating a classification-based evaluator
|
|
59
71
|
*/
|
|
60
72
|
export interface CreateClassifierArgs extends WithTelemetry {
|
|
61
|
-
model: LanguageModel;
|
|
73
|
+
model: LanguageModel | EvaluationModel;
|
|
62
74
|
/**
|
|
63
75
|
* The choices to classify the example into.
|
|
64
76
|
* e.g. { "correct": 1, "incorrect": 0 }
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evals.d.ts","sourceRoot":"","sources":["../../../src/types/evals.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAExC,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,QAAQ,CAAC;AAC5C,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,QAAQ,CAAC;AAC5C,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAEnD;;GAEG;AACH,MAAM,WAAW,aAAa,CAAC,UAAU,EAAE,SAAS;IAClD,MAAM,EAAE,UAAU,CAAC;IACnB,QAAQ,CAAC,EAAE,UAAU,CAAC;IACtB,KAAK,CAAC,EAAE,SAAS,CAAC;IAClB,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACxB;AAED,MAAM,WAAW,OAAO;IACtB,KAAK,EAAE,aAAa,CAAC;
|
|
1
|
+
{"version":3,"file":"evals.d.ts","sourceRoot":"","sources":["../../../src/types/evals.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAExC,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,QAAQ,CAAC;AAC5C,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,QAAQ,CAAC;AAC5C,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAEnD;;GAEG;AACH,MAAM,WAAW,aAAa,CAAC,UAAU,EAAE,SAAS;IAClD,MAAM,EAAE,UAAU,CAAC;IACnB,QAAQ,CAAC,EAAE,UAAU,CAAC;IACtB,KAAK,CAAC,EAAE,SAAS,CAAC;IAClB,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACxB;AAED,MAAM,WAAW,OAAO;IACtB;;;;OAIG;IACH,KAAK,EAAE,aAAa,GAAG,eAAe,CAAC;CACxC;AAGD,MAAM,WAAW,iBAAkB,SAAQ,OAAO;CAAG;AAErD;;GAEG;AACH,MAAM,WAAW,gBAAgB;IAC/B;;;OAGG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;;;OAGG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC;AAED;;GAEG;AACH,MAAM,WAAW,oBAAoB;IACnC,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACpC;AAED;;GAEG;AACH,MAAM,WAAW,oBAAoB;IACnC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;CACf;AAED;;GAEG;AACH,MAAM,MAAM,wBAAwB,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;AAE9D;;GAEG;AACH,MAAM,WAAW,oBAAqB,SAAQ,aAAa;IAIzD,KAAK,EAAE,aAAa,GAAG,eAAe,CAAC;IACvC;;;OAGG;IACH,OAAO,EAAE,wBAAwB,CAAC;IAClC;;OAEG;IACH,cAAc,EAAE,cAAc,CAAC;CAChC;AAED,MAAM,WAAW,mBAAmB,CAClC,WAAW,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CACrE,SAAQ,aAAa;IACrB;;;OAGG;IACH,IAAI,EAAE,MAAM,CAAC;IACb;;OAEG;IACH,IAAI,EAAE,cAAc,CAAC;IACrB;;;OAGG;IACH,qBAAqB,CAAC,EAAE,qBAAqB,CAAC;IAC9C;;OAEG;IACH,YAAY,CAAC,EAAE,aAAa,CAAC,WAAW,CAAC,CAAC;CAC3C;AAED,MAAM,MAAM,sBAAsB,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAC3E,IAAI,CAAC,mBAAmB,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC,CAAC;AAEhD,MAAM,WAAW,iCAAiC,CAChD,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAE1C,SAAQ,oBAAoB,EAAE,sBAAsB,CAAC,UAAU,CAAC;IAChE;;OAEG;IACH,cAAc,EAAE,cAAc,CAAC;CAChC;AAED,MAAM,MAAM,WAAW,CAAC,WAAW,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,CACrE,IAAI,EAAE,WAAW,KACd,OAAO,CAAC,gBAAgB,CAAC,CAAC;AAE/B;;GAEG;AACH,MAAM,MAAM,cAAc,GAAG,KAAK,GAAG,MAAM,CAAC;AAE5C;;;GAGG;AACH,MAAM,MAAM,qBAAqB,GAAG,UAAU,GAAG,UAAU,GAAG,SAAS,CAAC;AAExE;;GAEG;AACH,UAAU,oBAAoB;IAC5B;;OAEG;IACH,IAAI,EAAE,MAAM,CAAC;IACb;;OAEG;IACH,IAAI,EAAE,cAAc,CAAC;IACrB;;;OAGG;IACH,qBAAqB,CAAC,EAAE,qBAAqB,CAAC;CAC/C;AAED;;;GAGG;AACH,MAAM,WAAW,kBAAkB,CACjC,WAAW,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAC3C,SAAQ,oBAAoB;IAC5B;;OAEG;IACH,QAAQ,EAAE,WAAW,CAAC,WAAW,CAAC,CAAC;CACpC"}
|
|
@@ -202,14 +202,13 @@ export type BindingContext<RecordType extends Record<string, unknown>> = {
|
|
|
202
202
|
* }
|
|
203
203
|
* );
|
|
204
204
|
*
|
|
205
|
-
* //
|
|
205
|
+
* // Retrieval relevance evaluator
|
|
206
206
|
* const relevancyEvaluator = bindEvaluator<EvaluationData>(
|
|
207
|
-
*
|
|
207
|
+
* createRetrievalRelevanceEvaluator<EvaluationData>({ model: openai("gpt-4") }),
|
|
208
208
|
* {
|
|
209
209
|
* inputMapping: {
|
|
210
|
-
*
|
|
211
|
-
*
|
|
212
|
-
* output: "modelOutput",
|
|
210
|
+
* input: "userQuery",
|
|
211
|
+
* context: "systemContext",
|
|
213
212
|
* },
|
|
214
213
|
* }
|
|
215
214
|
* );
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"bindEvaluator.d.ts","sourceRoot":"","sources":["../../../src/utils/bindEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAC3D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAEnD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2CG;AACH,MAAM,MAAM,cAAc,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI;IACvE;;;;;;;;;;;;OAYG;IACH,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,CAAC;CACzC,CAAC;AAEF
|
|
1
|
+
{"version":3,"file":"bindEvaluator.d.ts","sourceRoot":"","sources":["../../../src/utils/bindEvaluator.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAC3D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAEnD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2CG;AACH,MAAM,MAAM,cAAc,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI;IACvE;;;;;;;;;;;;OAYG;IACH,YAAY,EAAE,aAAa,CAAC,UAAU,CAAC,CAAC;CACzC,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAyJG;AACH,wBAAgB,aAAa,CAAC,UAAU,SAAS,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACtE,SAAS,EAAE,aAAa,CAAC,UAAU,CAAC,EACpC,OAAO,EAAE,cAAc,CAAC,UAAU,CAAC,GAClC,aAAa,CAAC,UAAU,CAAC,CAM3B"}
|
|
@@ -143,14 +143,13 @@ exports.bindEvaluator = bindEvaluator;
|
|
|
143
143
|
* }
|
|
144
144
|
* );
|
|
145
145
|
*
|
|
146
|
-
* //
|
|
146
|
+
* // Retrieval relevance evaluator
|
|
147
147
|
* const relevancyEvaluator = bindEvaluator<EvaluationData>(
|
|
148
|
-
*
|
|
148
|
+
* createRetrievalRelevanceEvaluator<EvaluationData>({ model: openai("gpt-4") }),
|
|
149
149
|
* {
|
|
150
150
|
* inputMapping: {
|
|
151
|
-
*
|
|
152
|
-
*
|
|
153
|
-
* output: "modelOutput",
|
|
151
|
+
* input: "userQuery",
|
|
152
|
+
* context: "systemContext",
|
|
154
153
|
* },
|
|
155
154
|
* }
|
|
156
155
|
* );
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"bindEvaluator.js","sourceRoot":"","sources":["../../../src/utils/bindEvaluator.ts"],"names":[],"mappings":";;;AAgEA
|
|
1
|
+
{"version":3,"file":"bindEvaluator.js","sourceRoot":"","sources":["../../../src/utils/bindEvaluator.ts"],"names":[],"mappings":";;;AAgEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAyJG;AACH,uBACE,SAAoC,EACpC,OAAmC;IAEnC,IAAI,cAAc,GAA8B,SAAS,CAAC;IAC1D,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;QACzB,cAAc,GAAG,cAAc,CAAC,gBAAgB,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC;IACzE,CAAC;IACD,OAAO,cAAc,CAAC;AACxB,CAAC"}
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
import type { Experimental_EvaluationModel, LanguageModel } from "ai";
|
|
2
|
+
/**
|
|
3
|
+
* An AI SDK evaluation model instance (e.g. TypeSafe's Jev). Unlike a
|
|
4
|
+
* {@link LanguageModel}, evaluation models only answer typed questions and
|
|
5
|
+
* cannot generate free-form text.
|
|
6
|
+
*/
|
|
7
|
+
export type EvaluationModel = Exclude<Experimental_EvaluationModel, string>;
|
|
8
|
+
/**
|
|
9
|
+
* Type guard that distinguishes an AI SDK evaluation model from a language model.
|
|
10
|
+
* Evaluation models expose `doEvaluate` instead of `doGenerate`.
|
|
11
|
+
*/
|
|
12
|
+
export declare function isEvaluationModel(model: LanguageModel | EvaluationModel): model is EvaluationModel;
|
|
13
|
+
//# sourceMappingURL=isEvaluationModel.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"isEvaluationModel.d.ts","sourceRoot":"","sources":["../../../src/utils/isEvaluationModel.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,4BAA4B,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AAEtE;;;;GAIG;AACH,MAAM,MAAM,eAAe,GAAG,OAAO,CAAC,4BAA4B,EAAE,MAAM,CAAC,CAAC;AAE5E;;;GAGG;AACH,wBAAgB,iBAAiB,CAC/B,KAAK,EAAE,aAAa,GAAG,eAAe,GACrC,KAAK,IAAI,eAAe,CAO1B"}
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
+
exports.isEvaluationModel = isEvaluationModel;
|
|
4
|
+
/**
|
|
5
|
+
* Type guard that distinguishes an AI SDK evaluation model from a language model.
|
|
6
|
+
* Evaluation models expose `doEvaluate` instead of `doGenerate`.
|
|
7
|
+
*/
|
|
8
|
+
function isEvaluationModel(model) {
|
|
9
|
+
return (typeof model === "object" &&
|
|
10
|
+
model !== null &&
|
|
11
|
+
"doEvaluate" in model &&
|
|
12
|
+
typeof model.doEvaluate === "function");
|
|
13
|
+
}
|
|
14
|
+
//# sourceMappingURL=isEvaluationModel.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"isEvaluationModel.js","sourceRoot":"","sources":["../../../src/utils/isEvaluationModel.ts"],"names":[],"mappings":";;;AASA;;;GAGG;AACH,2BACE,KAAsC;IAEtC,OAAO,CACL,OAAO,KAAK,KAAK,QAAQ;QACzB,KAAK,KAAK,IAAI;QACd,YAAY,IAAI,KAAK;QACrB,OAAO,KAAK,CAAC,UAAU,KAAK,UAAU,CACvC,CAAC;AACJ,CAAC"}
|