promptfoo 0.103.13 → 0.103.15
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/package.json +19 -18
- package/dist/src/app/assets/{index-CNZfEf8i.js → index-CejUHECf.js} +276 -275
- package/dist/src/app/assets/{index.es-vrsSPryr.js → index.es-DWMnMsuU.js} +1 -1
- package/dist/src/app/assets/{sync-B_Y0n4tJ.js → sync-OIzapIKF.js} +1 -1
- package/dist/src/app/index.html +1 -1
- package/dist/src/assertions/refusal.d.ts.map +1 -1
- package/dist/src/assertions/refusal.js +14 -3
- package/dist/src/assertions/refusal.js.map +1 -1
- package/dist/src/fetch.d.ts +2 -0
- package/dist/src/fetch.d.ts.map +1 -1
- package/dist/src/fetch.js +2 -1
- package/dist/src/fetch.js.map +1 -1
- package/dist/src/index.d.ts +1 -0
- package/dist/src/index.d.ts.map +1 -1
- package/dist/src/providers/adaline.gateway.d.ts +13 -1
- package/dist/src/providers/adaline.gateway.d.ts.map +1 -1
- package/dist/src/providers/adaline.gateway.js +4 -2
- package/dist/src/providers/adaline.gateway.js.map +1 -1
- package/dist/src/providers/azure.d.ts.map +1 -1
- package/dist/src/providers/azure.js +22 -5
- package/dist/src/providers/azure.js.map +1 -1
- package/dist/src/providers/cloudera.d.ts +15 -0
- package/dist/src/providers/cloudera.d.ts.map +1 -0
- package/dist/src/providers/cloudera.js +22 -0
- package/dist/src/providers/cloudera.js.map +1 -0
- package/dist/src/providers/http.d.ts +5 -3
- package/dist/src/providers/http.d.ts.map +1 -1
- package/dist/src/providers/http.js +7 -6
- package/dist/src/providers/http.js.map +1 -1
- package/dist/src/providers/openai.d.ts.map +1 -1
- package/dist/src/providers/openai.js +7 -0
- package/dist/src/providers/openai.js.map +1 -1
- package/dist/src/providers/portkey.d.ts +3 -2
- package/dist/src/providers/portkey.d.ts.map +1 -1
- package/dist/src/providers/portkey.js +1 -0
- package/dist/src/providers/portkey.js.map +1 -1
- package/dist/src/providers/watsonx.d.ts +2 -2
- package/dist/src/providers.d.ts +2 -0
- package/dist/src/providers.d.ts.map +1 -1
- package/dist/src/providers.js +36 -3
- package/dist/src/providers.js.map +1 -1
- package/dist/src/redteam/commands/generate.d.ts.map +1 -1
- package/dist/src/redteam/commands/generate.js +31 -10
- package/dist/src/redteam/commands/generate.js.map +1 -1
- package/dist/src/redteam/commands/run.d.ts.map +1 -1
- package/dist/src/redteam/commands/run.js +1 -1
- package/dist/src/redteam/commands/run.js.map +1 -1
- package/dist/src/redteam/graders.d.ts.map +1 -1
- package/dist/src/redteam/graders.js +13 -13
- package/dist/src/redteam/graders.js.map +1 -1
- package/dist/src/redteam/index.d.ts +1 -0
- package/dist/src/redteam/index.d.ts.map +1 -1
- package/dist/src/redteam/index.js +1 -1
- package/dist/src/redteam/index.js.map +1 -1
- package/dist/src/redteam/plugins/harmful/graders.d.ts +52 -0
- package/dist/src/redteam/plugins/harmful/graders.d.ts.map +1 -1
- package/dist/src/redteam/plugins/harmful/graders.js +577 -22
- package/dist/src/redteam/plugins/harmful/graders.js.map +1 -1
- package/dist/src/redteam/providers/crescendo/index.d.ts +2 -1
- package/dist/src/redteam/providers/crescendo/index.d.ts.map +1 -1
- package/dist/src/redteam/providers/crescendo/index.js +5 -5
- package/dist/src/redteam/providers/crescendo/index.js.map +1 -1
- package/dist/src/redteam/providers/goat.d.ts +2 -1
- package/dist/src/redteam/providers/goat.d.ts.map +1 -1
- package/dist/src/redteam/providers/goat.js +5 -5
- package/dist/src/redteam/providers/goat.js.map +1 -1
- package/dist/src/redteam/providers/shared.js +1 -1
- package/dist/src/redteam/providers/shared.js.map +1 -1
- package/dist/src/redteam/sharedFrontend.d.ts.map +1 -1
- package/dist/src/redteam/sharedFrontend.js +18 -2
- package/dist/src/redteam/sharedFrontend.js.map +1 -1
- package/dist/src/redteam/types.d.ts +1 -0
- package/dist/src/redteam/types.d.ts.map +1 -1
- package/dist/src/redteam/util.d.ts.map +1 -1
- package/dist/src/redteam/util.js +2 -0
- package/dist/src/redteam/util.js.map +1 -1
- package/dist/src/server/server.d.ts.map +1 -1
- package/dist/src/server/server.js +1 -7
- package/dist/src/server/server.js.map +1 -1
- package/dist/src/validators/redteam.d.ts +3 -0
- package/dist/src/validators/redteam.d.ts.map +1 -1
- package/dist/src/validators/redteam.js +2 -0
- package/dist/src/validators/redteam.js.map +1 -1
- package/dist/test/assertions/answerRelevance.test.d.ts +2 -0
- package/dist/test/assertions/answerRelevance.test.d.ts.map +1 -0
- package/dist/test/assertions/answerRelevance.test.js +177 -0
- package/dist/test/assertions/answerRelevance.test.js.map +1 -0
- package/dist/test/assertions/contextFaithfulness.test.d.ts +2 -0
- package/dist/test/assertions/contextFaithfulness.test.d.ts.map +1 -0
- package/dist/test/assertions/contextFaithfulness.test.js +226 -0
- package/dist/test/assertions/contextFaithfulness.test.js.map +1 -0
- package/dist/test/assertions/contextRecall.test.d.ts +2 -0
- package/dist/test/assertions/contextRecall.test.d.ts.map +1 -0
- package/dist/test/assertions/contextRecall.test.js +243 -0
- package/dist/test/assertions/contextRecall.test.js.map +1 -0
- package/dist/test/assertions/contextRelevance.test.d.ts +2 -0
- package/dist/test/assertions/contextRelevance.test.d.ts.map +1 -0
- package/dist/test/assertions/contextRelevance.test.js +238 -0
- package/dist/test/assertions/contextRelevance.test.js.map +1 -0
- package/dist/test/assertions/geval.test.d.ts +2 -0
- package/dist/test/assertions/geval.test.d.ts.map +1 -0
- package/dist/test/assertions/geval.test.js +222 -0
- package/dist/test/assertions/geval.test.js.map +1 -0
- package/dist/test/assertions/modelGradedClosedQa.test.d.ts +2 -0
- package/dist/test/assertions/modelGradedClosedQa.test.d.ts.map +1 -0
- package/dist/test/assertions/modelGradedClosedQa.test.js +200 -0
- package/dist/test/assertions/modelGradedClosedQa.test.js.map +1 -0
- package/dist/test/assertions/refusal.test.d.ts +2 -0
- package/dist/test/assertions/refusal.test.d.ts.map +1 -0
- package/dist/test/assertions/refusal.test.js +300 -0
- package/dist/test/assertions/refusal.test.js.map +1 -0
- package/dist/test/fetch.test.js +66 -18
- package/dist/test/fetch.test.js.map +1 -1
- package/dist/test/providers/adaline.gateway.test.d.ts +2 -0
- package/dist/test/providers/adaline.gateway.test.d.ts.map +1 -0
- package/dist/test/providers/adaline.gateway.test.js +245 -0
- package/dist/test/providers/adaline.gateway.test.js.map +1 -0
- package/dist/test/providers/azure.test.js +73 -12
- package/dist/test/providers/azure.test.js.map +1 -1
- package/dist/test/providers/http.test.js +70 -2
- package/dist/test/providers/http.test.js.map +1 -1
- package/dist/test/providers/index.test.js +35 -459
- package/dist/test/providers/index.test.js.map +1 -1
- package/dist/test/providers/openai.test.js +509 -0
- package/dist/test/providers/openai.test.js.map +1 -1
- package/dist/test/providers/portkey.test.js +57 -0
- package/dist/test/providers/portkey.test.js.map +1 -1
- package/dist/test/providers/simulatedUser.test.js +1 -0
- package/dist/test/providers/simulatedUser.test.js.map +1 -1
- package/dist/test/providers.test.js +10 -0
- package/dist/test/providers.test.js.map +1 -1
- package/dist/test/redteam/commands/generate.test.js +7 -0
- package/dist/test/redteam/commands/generate.test.js.map +1 -1
- package/dist/test/redteam/extraction/purpose.test.js +1 -0
- package/dist/test/redteam/extraction/purpose.test.js.map +1 -1
- package/dist/test/redteam/plugins/asciiSmuggling.test.d.ts +2 -0
- package/dist/test/redteam/plugins/asciiSmuggling.test.d.ts.map +1 -0
- package/dist/test/redteam/plugins/asciiSmuggling.test.js +75 -0
- package/dist/test/redteam/plugins/asciiSmuggling.test.js.map +1 -0
- package/dist/test/redteam/providers/crescendo/index.test.js +3 -3
- package/dist/test/redteam/providers/crescendo/index.test.js.map +1 -1
- package/dist/test/redteam/strategies/index.test.js +1 -0
- package/dist/test/redteam/strategies/index.test.js.map +1 -1
- package/dist/tsconfig.tsbuildinfo +1 -1
- package/package.json +19 -18
|
@@ -0,0 +1,177 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
var __importDefault = (this && this.__importDefault) || function (mod) {
|
|
3
|
+
return (mod && mod.__esModule) ? mod : { "default": mod };
|
|
4
|
+
};
|
|
5
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
6
|
+
const answerRelevance_1 = require("../../src/assertions/answerRelevance");
|
|
7
|
+
const matchers_1 = require("../../src/matchers");
|
|
8
|
+
const invariant_1 = __importDefault(require("../../src/util/invariant"));
|
|
9
|
+
jest.mock('../../src/matchers');
|
|
10
|
+
jest.mock('../../src/util/invariant');
|
|
11
|
+
describe('handleAnswerRelevance', () => {
|
|
12
|
+
beforeEach(() => {
|
|
13
|
+
jest.resetAllMocks();
|
|
14
|
+
jest.mocked(invariant_1.default).mockImplementation((condition, message) => {
|
|
15
|
+
if (!condition) {
|
|
16
|
+
throw new Error(typeof message === 'function' ? message() : message);
|
|
17
|
+
}
|
|
18
|
+
});
|
|
19
|
+
});
|
|
20
|
+
it('should call matchesAnswerRelevance with correct parameters', async () => {
|
|
21
|
+
const mockMatchesAnswerRelevance = jest.mocked(matchers_1.matchesAnswerRelevance);
|
|
22
|
+
mockMatchesAnswerRelevance.mockResolvedValue({
|
|
23
|
+
pass: true,
|
|
24
|
+
score: 0.8,
|
|
25
|
+
reason: 'test reason',
|
|
26
|
+
});
|
|
27
|
+
const result = await (0, answerRelevance_1.handleAnswerRelevance)({
|
|
28
|
+
assertion: {
|
|
29
|
+
type: 'answer-relevance',
|
|
30
|
+
threshold: 0.7,
|
|
31
|
+
},
|
|
32
|
+
output: 'test output',
|
|
33
|
+
prompt: 'test prompt',
|
|
34
|
+
test: {
|
|
35
|
+
vars: {},
|
|
36
|
+
options: {},
|
|
37
|
+
},
|
|
38
|
+
baseType: 'answer-relevance',
|
|
39
|
+
context: {},
|
|
40
|
+
inverse: false,
|
|
41
|
+
outputString: 'test output',
|
|
42
|
+
providerResponse: {
|
|
43
|
+
output: 'test output',
|
|
44
|
+
tokenUsage: {},
|
|
45
|
+
},
|
|
46
|
+
});
|
|
47
|
+
expect(mockMatchesAnswerRelevance).toHaveBeenCalledWith('test prompt', 'test output', 0.7, {});
|
|
48
|
+
expect(result).toEqual({
|
|
49
|
+
assertion: {
|
|
50
|
+
type: 'answer-relevance',
|
|
51
|
+
threshold: 0.7,
|
|
52
|
+
},
|
|
53
|
+
pass: true,
|
|
54
|
+
score: 0.8,
|
|
55
|
+
reason: 'test reason',
|
|
56
|
+
});
|
|
57
|
+
});
|
|
58
|
+
it('should use query from vars if available', async () => {
|
|
59
|
+
const mockMatchesAnswerRelevance = jest.mocked(matchers_1.matchesAnswerRelevance);
|
|
60
|
+
mockMatchesAnswerRelevance.mockResolvedValue({
|
|
61
|
+
pass: true,
|
|
62
|
+
score: 0.8,
|
|
63
|
+
reason: 'test reason',
|
|
64
|
+
});
|
|
65
|
+
const result = await (0, answerRelevance_1.handleAnswerRelevance)({
|
|
66
|
+
assertion: {
|
|
67
|
+
type: 'answer-relevance',
|
|
68
|
+
threshold: 0.7,
|
|
69
|
+
},
|
|
70
|
+
output: 'test output',
|
|
71
|
+
prompt: 'test prompt',
|
|
72
|
+
test: {
|
|
73
|
+
vars: {
|
|
74
|
+
query: 'test query',
|
|
75
|
+
},
|
|
76
|
+
options: {},
|
|
77
|
+
},
|
|
78
|
+
baseType: 'answer-relevance',
|
|
79
|
+
context: {},
|
|
80
|
+
inverse: false,
|
|
81
|
+
outputString: 'test output',
|
|
82
|
+
providerResponse: {
|
|
83
|
+
output: 'test output',
|
|
84
|
+
tokenUsage: {},
|
|
85
|
+
},
|
|
86
|
+
});
|
|
87
|
+
expect(mockMatchesAnswerRelevance).toHaveBeenCalledWith('test query', 'test output', 0.7, {});
|
|
88
|
+
expect(result).toEqual({
|
|
89
|
+
assertion: {
|
|
90
|
+
type: 'answer-relevance',
|
|
91
|
+
threshold: 0.7,
|
|
92
|
+
},
|
|
93
|
+
pass: true,
|
|
94
|
+
score: 0.8,
|
|
95
|
+
reason: 'test reason',
|
|
96
|
+
});
|
|
97
|
+
});
|
|
98
|
+
it('should throw error if output is not string', async () => {
|
|
99
|
+
await expect((0, answerRelevance_1.handleAnswerRelevance)({
|
|
100
|
+
assertion: {
|
|
101
|
+
type: 'answer-relevance',
|
|
102
|
+
},
|
|
103
|
+
output: {},
|
|
104
|
+
prompt: 'test prompt',
|
|
105
|
+
test: {
|
|
106
|
+
vars: {},
|
|
107
|
+
options: {},
|
|
108
|
+
},
|
|
109
|
+
baseType: 'answer-relevance',
|
|
110
|
+
context: {},
|
|
111
|
+
inverse: false,
|
|
112
|
+
outputString: 'test output',
|
|
113
|
+
providerResponse: {
|
|
114
|
+
output: 'test output',
|
|
115
|
+
tokenUsage: {},
|
|
116
|
+
},
|
|
117
|
+
})).rejects.toThrow('answer-relevance assertion type must evaluate a string output');
|
|
118
|
+
});
|
|
119
|
+
it('should throw error if prompt is missing', async () => {
|
|
120
|
+
await expect((0, answerRelevance_1.handleAnswerRelevance)({
|
|
121
|
+
assertion: {
|
|
122
|
+
type: 'answer-relevance',
|
|
123
|
+
},
|
|
124
|
+
output: 'test output',
|
|
125
|
+
prompt: '',
|
|
126
|
+
test: {
|
|
127
|
+
vars: {},
|
|
128
|
+
options: {},
|
|
129
|
+
},
|
|
130
|
+
baseType: 'answer-relevance',
|
|
131
|
+
context: {},
|
|
132
|
+
inverse: false,
|
|
133
|
+
outputString: 'test output',
|
|
134
|
+
providerResponse: {
|
|
135
|
+
output: 'test output',
|
|
136
|
+
tokenUsage: {},
|
|
137
|
+
},
|
|
138
|
+
})).rejects.toThrow('answer-relevance assertion type must have a prompt');
|
|
139
|
+
});
|
|
140
|
+
it('should use default threshold of 0 if not specified', async () => {
|
|
141
|
+
const mockMatchesAnswerRelevance = jest.mocked(matchers_1.matchesAnswerRelevance);
|
|
142
|
+
mockMatchesAnswerRelevance.mockResolvedValue({
|
|
143
|
+
pass: true,
|
|
144
|
+
score: 0.8,
|
|
145
|
+
reason: 'test reason',
|
|
146
|
+
});
|
|
147
|
+
const result = await (0, answerRelevance_1.handleAnswerRelevance)({
|
|
148
|
+
assertion: {
|
|
149
|
+
type: 'answer-relevance',
|
|
150
|
+
},
|
|
151
|
+
output: 'test output',
|
|
152
|
+
prompt: 'test prompt',
|
|
153
|
+
test: {
|
|
154
|
+
vars: {},
|
|
155
|
+
options: {},
|
|
156
|
+
},
|
|
157
|
+
baseType: 'answer-relevance',
|
|
158
|
+
context: {},
|
|
159
|
+
inverse: false,
|
|
160
|
+
outputString: 'test output',
|
|
161
|
+
providerResponse: {
|
|
162
|
+
output: 'test output',
|
|
163
|
+
tokenUsage: {},
|
|
164
|
+
},
|
|
165
|
+
});
|
|
166
|
+
expect(mockMatchesAnswerRelevance).toHaveBeenCalledWith('test prompt', 'test output', 0, {});
|
|
167
|
+
expect(result).toEqual({
|
|
168
|
+
assertion: {
|
|
169
|
+
type: 'answer-relevance',
|
|
170
|
+
},
|
|
171
|
+
pass: true,
|
|
172
|
+
score: 0.8,
|
|
173
|
+
reason: 'test reason',
|
|
174
|
+
});
|
|
175
|
+
});
|
|
176
|
+
});
|
|
177
|
+
//# sourceMappingURL=answerRelevance.test.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"answerRelevance.test.js","sourceRoot":"","sources":["../../../test/assertions/answerRelevance.test.ts"],"names":[],"mappings":";;;;;AAAA,0EAA6E;AAC7E,iDAA4D;AAE5D,yEAAiD;AAEjD,IAAI,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;AAChC,IAAI,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;AAEtC,QAAQ,CAAC,uBAAuB,EAAE,GAAG,EAAE;IACrC,UAAU,CAAC,GAAG,EAAE;QACd,IAAI,CAAC,aAAa,EAAE,CAAC;QACrB,IAAI,CAAC,MAAM,CAAC,mBAAS,CAAC,CAAC,kBAAkB,CAAC,CAAC,SAAS,EAAE,OAAO,EAAE,EAAE;YAC/D,IAAI,CAAC,SAAS,EAAE,CAAC;gBACf,MAAM,IAAI,KAAK,CAAC,OAAO,OAAO,KAAK,UAAU,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;YACvE,CAAC;QACH,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4DAA4D,EAAE,KAAK,IAAI,EAAE;QAC1E,MAAM,0BAA0B,GAAG,IAAI,CAAC,MAAM,CAAC,iCAAsB,CAAC,CAAC;QACvE,0BAA0B,CAAC,iBAAiB,CAAC;YAC3C,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,GAAG;YACV,MAAM,EAAE,aAAa;SACtB,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,IAAA,uCAAqB,EAAC;YACzC,SAAS,EAAE;gBACT,IAAI,EAAE,kBAAkB;gBACxB,SAAS,EAAE,GAAG;aACf;YACD,MAAM,EAAE,aAAa;YACrB,MAAM,EAAE,aAAa;YACrB,IAAI,EAAE;gBACJ,IAAI,EAAE,EAAE;gBACR,OAAO,EAAE,EAAE;aACZ;YACD,QAAQ,EAAE,kBAAkB;YAC5B,OAAO,EAAE,EAAmC;YAC5C,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE;gBAChB,MAAM,EAAE,aAAa;gBACrB,UAAU,EAAE,EAAE;aACf;SACF,CAAC,CAAC;QAEH,MAAM,CAAC,0BAA0B,CAAC,CAAC,oBAAoB,CAAC,aAAa,EAAE,aAAa,EAAE,GAAG,EAAE,EAAE,CAAC,CAAC;QAC/F,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC;YACrB,SAAS,EAAE;gBACT,IAAI,EAAE,kBAAkB;gBACxB,SAAS,EAAE,GAAG;aACf;YACD,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,GAAG;YACV,MAAM,EAAE,aAAa;SACtB,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;QACvD,MAAM,0BAA0B,GAAG,IAAI,CAAC,MAAM,CAAC,iCAAsB,CAAC,CAAC;QACvE,0BAA0B,CAAC,iBAAiB,CAAC;YAC3C,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,GAAG;YACV,MAAM,EAAE,aAAa;SACtB,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,IAAA,uCAAqB,EAAC;YACzC,SAAS,EAAE;gBACT,IAAI,EAAE,kBAAkB;gBACxB,SAAS,EAAE,GAAG;aACf;YACD,MAAM,EAAE,aAAa;YACrB,MAAM,EAAE,aAAa;YACrB,IAAI,EAAE;gBACJ,IAAI,EAAE;oBACJ,KAAK,EAAE,YAAY;iBACpB;gBACD,OAAO,EAAE,EAAE;aACZ;YACD,QAAQ,EAAE,kBAAkB;YAC5B,OAAO,EAAE,EAAmC;YAC5C,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE;gBAChB,MAAM,EAAE,aAAa;gBACrB,UAAU,EAAE,EAAE;aACf;SACF,CAAC,CAAC;QAEH,MAAM,CAAC,0BAA0B,CAAC,CAAC,oBAAoB,CAAC,YAAY,EAAE,aAAa,EAAE,GAAG,EAAE,EAAE,CAAC,CAAC;QAC9F,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC;YACrB,SAAS,EAAE;gBACT,IAAI,EAAE,kBAAkB;gBACxB,SAAS,EAAE,GAAG;aACf;YACD,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,GAAG;YACV,MAAM,EAAE,aAAa;SACtB,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4CAA4C,EAAE,KAAK,IAAI,EAAE;QAC1D,MAAM,MAAM,CACV,IAAA,uCAAqB,EAAC;YACpB,SAAS,EAAE;gBACT,IAAI,EAAE,kBAAkB;aACzB;YACD,MAAM,EAAE,EAAE;YACV,MAAM,EAAE,aAAa;YACrB,IAAI,EAAE;gBACJ,IAAI,EAAE,EAAE;gBACR,OAAO,EAAE,EAAE;aACZ;YACD,QAAQ,EAAE,kBAAkB;YAC5B,OAAO,EAAE,EAAmC;YAC5C,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE;gBAChB,MAAM,EAAE,aAAa;gBACrB,UAAU,EAAE,EAAE;aACf;SACF,CAAC,CACH,CAAC,OAAO,CAAC,OAAO,CAAC,+DAA+D,CAAC,CAAC;IACrF,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;QACvD,MAAM,MAAM,CACV,IAAA,uCAAqB,EAAC;YACpB,SAAS,EAAE;gBACT,IAAI,EAAE,kBAAkB;aACzB;YACD,MAAM,EAAE,aAAa;YACrB,MAAM,EAAE,EAAE;YACV,IAAI,EAAE;gBACJ,IAAI,EAAE,EAAE;gBACR,OAAO,EAAE,EAAE;aACZ;YACD,QAAQ,EAAE,kBAAkB;YAC5B,OAAO,EAAE,EAAmC;YAC5C,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE;gBAChB,MAAM,EAAE,aAAa;gBACrB,UAAU,EAAE,EAAE;aACf;SACF,CAAC,CACH,CAAC,OAAO,CAAC,OAAO,CAAC,oDAAoD,CAAC,CAAC;IAC1E,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oDAAoD,EAAE,KAAK,IAAI,EAAE;QAClE,MAAM,0BAA0B,GAAG,IAAI,CAAC,MAAM,CAAC,iCAAsB,CAAC,CAAC;QACvE,0BAA0B,CAAC,iBAAiB,CAAC;YAC3C,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,GAAG;YACV,MAAM,EAAE,aAAa;SACtB,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,IAAA,uCAAqB,EAAC;YACzC,SAAS,EAAE;gBACT,IAAI,EAAE,kBAAkB;aACzB;YACD,MAAM,EAAE,aAAa;YACrB,MAAM,EAAE,aAAa;YACrB,IAAI,EAAE;gBACJ,IAAI,EAAE,EAAE;gBACR,OAAO,EAAE,EAAE;aACZ;YACD,QAAQ,EAAE,kBAAkB;YAC5B,OAAO,EAAE,EAAmC;YAC5C,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE;gBAChB,MAAM,EAAE,aAAa;gBACrB,UAAU,EAAE,EAAE;aACf;SACF,CAAC,CAAC;QAEH,MAAM,CAAC,0BAA0B,CAAC,CAAC,oBAAoB,CAAC,aAAa,EAAE,aAAa,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;QAC7F,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC;YACrB,SAAS,EAAE;gBACT,IAAI,EAAE,kBAAkB;aACzB;YACD,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,GAAG;YACV,MAAM,EAAE,aAAa;SACtB,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"contextFaithfulness.test.d.ts","sourceRoot":"","sources":["../../../test/assertions/contextFaithfulness.test.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,226 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
|
|
3
|
+
if (k2 === undefined) k2 = k;
|
|
4
|
+
var desc = Object.getOwnPropertyDescriptor(m, k);
|
|
5
|
+
if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
|
|
6
|
+
desc = { enumerable: true, get: function() { return m[k]; } };
|
|
7
|
+
}
|
|
8
|
+
Object.defineProperty(o, k2, desc);
|
|
9
|
+
}) : (function(o, m, k, k2) {
|
|
10
|
+
if (k2 === undefined) k2 = k;
|
|
11
|
+
o[k2] = m[k];
|
|
12
|
+
}));
|
|
13
|
+
var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
|
|
14
|
+
Object.defineProperty(o, "default", { enumerable: true, value: v });
|
|
15
|
+
}) : function(o, v) {
|
|
16
|
+
o["default"] = v;
|
|
17
|
+
});
|
|
18
|
+
var __importStar = (this && this.__importStar) || (function () {
|
|
19
|
+
var ownKeys = function(o) {
|
|
20
|
+
ownKeys = Object.getOwnPropertyNames || function (o) {
|
|
21
|
+
var ar = [];
|
|
22
|
+
for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
|
|
23
|
+
return ar;
|
|
24
|
+
};
|
|
25
|
+
return ownKeys(o);
|
|
26
|
+
};
|
|
27
|
+
return function (mod) {
|
|
28
|
+
if (mod && mod.__esModule) return mod;
|
|
29
|
+
var result = {};
|
|
30
|
+
if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
|
|
31
|
+
__setModuleDefault(result, mod);
|
|
32
|
+
return result;
|
|
33
|
+
};
|
|
34
|
+
})();
|
|
35
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
36
|
+
const contextFaithfulness_1 = require("../../src/assertions/contextFaithfulness");
|
|
37
|
+
const matchers = __importStar(require("../../src/matchers"));
|
|
38
|
+
jest.mock('../../src/matchers');
|
|
39
|
+
describe('handleContextFaithfulness', () => {
|
|
40
|
+
beforeEach(() => {
|
|
41
|
+
jest.resetAllMocks();
|
|
42
|
+
});
|
|
43
|
+
it('should throw error if vars is missing', async () => {
|
|
44
|
+
const params = {
|
|
45
|
+
assertion: { type: 'context-faithfulness' },
|
|
46
|
+
test: {},
|
|
47
|
+
output: 'test output',
|
|
48
|
+
baseType: 'context-faithfulness',
|
|
49
|
+
context: {
|
|
50
|
+
prompt: 'test prompt',
|
|
51
|
+
vars: {},
|
|
52
|
+
test: {},
|
|
53
|
+
logProbs: null,
|
|
54
|
+
tokenUsage: null,
|
|
55
|
+
cached: false,
|
|
56
|
+
provider: null,
|
|
57
|
+
providerResponse: null,
|
|
58
|
+
},
|
|
59
|
+
inverse: false,
|
|
60
|
+
outputString: 'test output',
|
|
61
|
+
providerResponse: null,
|
|
62
|
+
};
|
|
63
|
+
await expect((0, contextFaithfulness_1.handleContextFaithfulness)(params)).rejects.toThrow('context-faithfulness assertion type must have a vars object');
|
|
64
|
+
});
|
|
65
|
+
it('should throw error if query is missing', async () => {
|
|
66
|
+
const params = {
|
|
67
|
+
assertion: { type: 'context-faithfulness' },
|
|
68
|
+
test: {
|
|
69
|
+
vars: {
|
|
70
|
+
context: 'test context',
|
|
71
|
+
},
|
|
72
|
+
},
|
|
73
|
+
output: 'test output',
|
|
74
|
+
baseType: 'context-faithfulness',
|
|
75
|
+
context: {
|
|
76
|
+
prompt: 'test prompt',
|
|
77
|
+
vars: {},
|
|
78
|
+
test: {},
|
|
79
|
+
logProbs: null,
|
|
80
|
+
tokenUsage: null,
|
|
81
|
+
cached: false,
|
|
82
|
+
provider: null,
|
|
83
|
+
providerResponse: null,
|
|
84
|
+
},
|
|
85
|
+
inverse: false,
|
|
86
|
+
outputString: 'test output',
|
|
87
|
+
providerResponse: null,
|
|
88
|
+
};
|
|
89
|
+
await expect((0, contextFaithfulness_1.handleContextFaithfulness)(params)).rejects.toThrow('context-faithfulness assertion type must have a query var');
|
|
90
|
+
});
|
|
91
|
+
it('should throw error if context is missing', async () => {
|
|
92
|
+
const params = {
|
|
93
|
+
assertion: { type: 'context-faithfulness' },
|
|
94
|
+
test: {
|
|
95
|
+
vars: {
|
|
96
|
+
query: 'test query',
|
|
97
|
+
},
|
|
98
|
+
},
|
|
99
|
+
output: 'test output',
|
|
100
|
+
baseType: 'context-faithfulness',
|
|
101
|
+
context: {
|
|
102
|
+
prompt: 'test prompt',
|
|
103
|
+
vars: {},
|
|
104
|
+
test: {},
|
|
105
|
+
logProbs: null,
|
|
106
|
+
tokenUsage: null,
|
|
107
|
+
cached: false,
|
|
108
|
+
provider: null,
|
|
109
|
+
providerResponse: null,
|
|
110
|
+
},
|
|
111
|
+
inverse: false,
|
|
112
|
+
outputString: 'test output',
|
|
113
|
+
providerResponse: null,
|
|
114
|
+
};
|
|
115
|
+
await expect((0, contextFaithfulness_1.handleContextFaithfulness)(params)).rejects.toThrow('context-faithfulness assertion type must have a context var');
|
|
116
|
+
});
|
|
117
|
+
it('should throw error if output is not a string', async () => {
|
|
118
|
+
const params = {
|
|
119
|
+
assertion: { type: 'context-faithfulness' },
|
|
120
|
+
test: {
|
|
121
|
+
vars: {
|
|
122
|
+
query: 'test query',
|
|
123
|
+
context: 'test context',
|
|
124
|
+
},
|
|
125
|
+
},
|
|
126
|
+
output: 123,
|
|
127
|
+
baseType: 'context-faithfulness',
|
|
128
|
+
context: {
|
|
129
|
+
prompt: 'test prompt',
|
|
130
|
+
vars: {},
|
|
131
|
+
test: {},
|
|
132
|
+
logProbs: null,
|
|
133
|
+
tokenUsage: null,
|
|
134
|
+
cached: false,
|
|
135
|
+
provider: null,
|
|
136
|
+
providerResponse: null,
|
|
137
|
+
},
|
|
138
|
+
inverse: false,
|
|
139
|
+
outputString: 'test output',
|
|
140
|
+
providerResponse: null,
|
|
141
|
+
};
|
|
142
|
+
await expect((0, contextFaithfulness_1.handleContextFaithfulness)(params)).rejects.toThrow('context-faithfulness assertion type must have a string output');
|
|
143
|
+
});
|
|
144
|
+
it('should call matchesContextFaithfulness with correct params', async () => {
|
|
145
|
+
const mockResult = {
|
|
146
|
+
pass: true,
|
|
147
|
+
score: 0.9,
|
|
148
|
+
reason: 'test reason',
|
|
149
|
+
};
|
|
150
|
+
jest.mocked(matchers.matchesContextFaithfulness).mockResolvedValue(mockResult);
|
|
151
|
+
const params = {
|
|
152
|
+
assertion: {
|
|
153
|
+
type: 'context-faithfulness',
|
|
154
|
+
threshold: 0.8,
|
|
155
|
+
},
|
|
156
|
+
test: {
|
|
157
|
+
vars: {
|
|
158
|
+
query: 'test query',
|
|
159
|
+
context: 'test context',
|
|
160
|
+
},
|
|
161
|
+
options: {
|
|
162
|
+
provider: 'test-provider',
|
|
163
|
+
},
|
|
164
|
+
},
|
|
165
|
+
output: 'test output',
|
|
166
|
+
baseType: 'context-faithfulness',
|
|
167
|
+
context: {
|
|
168
|
+
prompt: 'test prompt',
|
|
169
|
+
vars: {},
|
|
170
|
+
test: {},
|
|
171
|
+
logProbs: null,
|
|
172
|
+
tokenUsage: null,
|
|
173
|
+
cached: false,
|
|
174
|
+
provider: null,
|
|
175
|
+
providerResponse: null,
|
|
176
|
+
},
|
|
177
|
+
inverse: false,
|
|
178
|
+
outputString: 'test output',
|
|
179
|
+
providerResponse: null,
|
|
180
|
+
};
|
|
181
|
+
const result = await (0, contextFaithfulness_1.handleContextFaithfulness)(params);
|
|
182
|
+
expect(matchers.matchesContextFaithfulness).toHaveBeenCalledWith('test query', 'test output', 'test context', 0.8, { provider: 'test-provider' });
|
|
183
|
+
expect(result).toEqual({
|
|
184
|
+
assertion: params.assertion,
|
|
185
|
+
...mockResult,
|
|
186
|
+
});
|
|
187
|
+
});
|
|
188
|
+
it('should use default threshold of 0 if not provided', async () => {
|
|
189
|
+
const mockResult = {
|
|
190
|
+
pass: true,
|
|
191
|
+
score: 0.9,
|
|
192
|
+
reason: 'test reason',
|
|
193
|
+
};
|
|
194
|
+
jest.mocked(matchers.matchesContextFaithfulness).mockResolvedValue(mockResult);
|
|
195
|
+
const params = {
|
|
196
|
+
assertion: {
|
|
197
|
+
type: 'context-faithfulness',
|
|
198
|
+
},
|
|
199
|
+
test: {
|
|
200
|
+
vars: {
|
|
201
|
+
query: 'test query',
|
|
202
|
+
context: 'test context',
|
|
203
|
+
},
|
|
204
|
+
options: {},
|
|
205
|
+
},
|
|
206
|
+
output: 'test output',
|
|
207
|
+
baseType: 'context-faithfulness',
|
|
208
|
+
context: {
|
|
209
|
+
prompt: 'test prompt',
|
|
210
|
+
vars: {},
|
|
211
|
+
test: {},
|
|
212
|
+
logProbs: null,
|
|
213
|
+
tokenUsage: null,
|
|
214
|
+
cached: false,
|
|
215
|
+
provider: null,
|
|
216
|
+
providerResponse: null,
|
|
217
|
+
},
|
|
218
|
+
inverse: false,
|
|
219
|
+
outputString: 'test output',
|
|
220
|
+
providerResponse: null,
|
|
221
|
+
};
|
|
222
|
+
await (0, contextFaithfulness_1.handleContextFaithfulness)(params);
|
|
223
|
+
expect(matchers.matchesContextFaithfulness).toHaveBeenCalledWith('test query', 'test output', 'test context', 0, {});
|
|
224
|
+
});
|
|
225
|
+
});
|
|
226
|
+
//# sourceMappingURL=contextFaithfulness.test.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"contextFaithfulness.test.js","sourceRoot":"","sources":["../../../test/assertions/contextFaithfulness.test.ts"],"names":[],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAAA,kFAAqF;AACrF,6DAA+C;AAE/C,IAAI,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;AAEhC,QAAQ,CAAC,2BAA2B,EAAE,GAAG,EAAE;IACzC,UAAU,CAAC,GAAG,EAAE;QACd,IAAI,CAAC,aAAa,EAAE,CAAC;IACvB,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uCAAuC,EAAE,KAAK,IAAI,EAAE;QACrD,MAAM,MAAM,GAAG;YACb,SAAS,EAAE,EAAE,IAAI,EAAE,sBAAsB,EAAE;YAC3C,IAAI,EAAE,EAAE;YACR,MAAM,EAAE,aAAa;YACrB,QAAQ,EAAE,sBAA+B;YACzC,OAAO,EAAE;gBACP,MAAM,EAAE,aAAa;gBACrB,IAAI,EAAE,EAAE;gBACR,IAAI,EAAE,EAAE;gBACR,QAAQ,EAAE,IAAI;gBACd,UAAU,EAAE,IAAI;gBAChB,MAAM,EAAE,KAAK;gBACb,QAAQ,EAAE,IAAI;gBACd,gBAAgB,EAAE,IAAI;aACvB;YACD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE,IAAI;SACvB,CAAC;QAEF,MAAM,MAAM,CAAC,IAAA,+CAAyB,EAAC,MAAa,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CACpE,6DAA6D,CAC9D,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wCAAwC,EAAE,KAAK,IAAI,EAAE;QACtD,MAAM,MAAM,GAAG;YACb,SAAS,EAAE,EAAE,IAAI,EAAE,sBAAsB,EAAE;YAC3C,IAAI,EAAE;gBACJ,IAAI,EAAE;oBACJ,OAAO,EAAE,cAAc;iBACxB;aACF;YACD,MAAM,EAAE,aAAa;YACrB,QAAQ,EAAE,sBAA+B;YACzC,OAAO,EAAE;gBACP,MAAM,EAAE,aAAa;gBACrB,IAAI,EAAE,EAAE;gBACR,IAAI,EAAE,EAAE;gBACR,QAAQ,EAAE,IAAI;gBACd,UAAU,EAAE,IAAI;gBAChB,MAAM,EAAE,KAAK;gBACb,QAAQ,EAAE,IAAI;gBACd,gBAAgB,EAAE,IAAI;aACvB;YACD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE,IAAI;SACvB,CAAC;QAEF,MAAM,MAAM,CAAC,IAAA,+CAAyB,EAAC,MAAa,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CACpE,2DAA2D,CAC5D,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0CAA0C,EAAE,KAAK,IAAI,EAAE;QACxD,MAAM,MAAM,GAAG;YACb,SAAS,EAAE,EAAE,IAAI,EAAE,sBAAsB,EAAE;YAC3C,IAAI,EAAE;gBACJ,IAAI,EAAE;oBACJ,KAAK,EAAE,YAAY;iBACpB;aACF;YACD,MAAM,EAAE,aAAa;YACrB,QAAQ,EAAE,sBAA+B;YACzC,OAAO,EAAE;gBACP,MAAM,EAAE,aAAa;gBACrB,IAAI,EAAE,EAAE;gBACR,IAAI,EAAE,EAAE;gBACR,QAAQ,EAAE,IAAI;gBACd,UAAU,EAAE,IAAI;gBAChB,MAAM,EAAE,KAAK;gBACb,QAAQ,EAAE,IAAI;gBACd,gBAAgB,EAAE,IAAI;aACvB;YACD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE,IAAI;SACvB,CAAC;QAEF,MAAM,MAAM,CAAC,IAAA,+CAAyB,EAAC,MAAa,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CACpE,6DAA6D,CAC9D,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;QAC5D,MAAM,MAAM,GAAG;YACb,SAAS,EAAE,EAAE,IAAI,EAAE,sBAAsB,EAAE;YAC3C,IAAI,EAAE;gBACJ,IAAI,EAAE;oBACJ,KAAK,EAAE,YAAY;oBACnB,OAAO,EAAE,cAAc;iBACxB;aACF;YACD,MAAM,EAAE,GAAG;YACX,QAAQ,EAAE,sBAA+B;YACzC,OAAO,EAAE;gBACP,MAAM,EAAE,aAAa;gBACrB,IAAI,EAAE,EAAE;gBACR,IAAI,EAAE,EAAE;gBACR,QAAQ,EAAE,IAAI;gBACd,UAAU,EAAE,IAAI;gBAChB,MAAM,EAAE,KAAK;gBACb,QAAQ,EAAE,IAAI;gBACd,gBAAgB,EAAE,IAAI;aACvB;YACD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE,IAAI;SACvB,CAAC;QAEF,MAAM,MAAM,CAAC,IAAA,+CAAyB,EAAC,MAAa,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CACpE,+DAA+D,CAChE,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4DAA4D,EAAE,KAAK,IAAI,EAAE;QAC1E,MAAM,UAAU,GAAG;YACjB,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,GAAG;YACV,MAAM,EAAE,aAAa;SACtB,CAAC;QAEF,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,0BAA0B,CAAC,CAAC,iBAAiB,CAAC,UAAU,CAAC,CAAC;QAE/E,MAAM,MAAM,GAAG;YACb,SAAS,EAAE;gBACT,IAAI,EAAE,sBAA+B;gBACrC,SAAS,EAAE,GAAG;aACf;YACD,IAAI,EAAE;gBACJ,IAAI,EAAE;oBACJ,KAAK,EAAE,YAAY;oBACnB,OAAO,EAAE,cAAc;iBACxB;gBACD,OAAO,EAAE;oBACP,QAAQ,EAAE,eAAe;iBAC1B;aACF;YACD,MAAM,EAAE,aAAa;YACrB,QAAQ,EAAE,sBAA+B;YACzC,OAAO,EAAE;gBACP,MAAM,EAAE,aAAa;gBACrB,IAAI,EAAE,EAAE;gBACR,IAAI,EAAE,EAAE;gBACR,QAAQ,EAAE,IAAI;gBACd,UAAU,EAAE,IAAI;gBAChB,MAAM,EAAE,KAAK;gBACb,QAAQ,EAAE,IAAI;gBACd,gBAAgB,EAAE,IAAI;aACvB;YACD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE,IAAI;SACvB,CAAC;QAEF,MAAM,MAAM,GAAG,MAAM,IAAA,+CAAyB,EAAC,MAAa,CAAC,CAAC;QAE9D,MAAM,CAAC,QAAQ,CAAC,0BAA0B,CAAC,CAAC,oBAAoB,CAC9D,YAAY,EACZ,aAAa,EACb,cAAc,EACd,GAAG,EACH,EAAE,QAAQ,EAAE,eAAe,EAAE,CAC9B,CAAC;QAEF,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC;YACrB,SAAS,EAAE,MAAM,CAAC,SAAS;YAC3B,GAAG,UAAU;SACd,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,mDAAmD,EAAE,KAAK,IAAI,EAAE;QACjE,MAAM,UAAU,GAAG;YACjB,IAAI,EAAE,IAAI;YACV,KAAK,EAAE,GAAG;YACV,MAAM,EAAE,aAAa;SACtB,CAAC;QAEF,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,0BAA0B,CAAC,CAAC,iBAAiB,CAAC,UAAU,CAAC,CAAC;QAE/E,MAAM,MAAM,GAAG;YACb,SAAS,EAAE;gBACT,IAAI,EAAE,sBAA+B;aACtC;YACD,IAAI,EAAE;gBACJ,IAAI,EAAE;oBACJ,KAAK,EAAE,YAAY;oBACnB,OAAO,EAAE,cAAc;iBACxB;gBACD,OAAO,EAAE,EAAE;aACZ;YACD,MAAM,EAAE,aAAa;YACrB,QAAQ,EAAE,sBAA+B;YACzC,OAAO,EAAE;gBACP,MAAM,EAAE,aAAa;gBACrB,IAAI,EAAE,EAAE;gBACR,IAAI,EAAE,EAAE;gBACR,QAAQ,EAAE,IAAI;gBACd,UAAU,EAAE,IAAI;gBAChB,MAAM,EAAE,KAAK;gBACb,QAAQ,EAAE,IAAI;gBACd,gBAAgB,EAAE,IAAI;aACvB;YACD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,aAAa;YAC3B,gBAAgB,EAAE,IAAI;SACvB,CAAC;QAEF,MAAM,IAAA,+CAAyB,EAAC,MAAa,CAAC,CAAC;QAE/C,MAAM,CAAC,QAAQ,CAAC,0BAA0B,CAAC,CAAC,oBAAoB,CAC9D,YAAY,EACZ,aAAa,EACb,cAAc,EACd,CAAC,EACD,EAAE,CACH,CAAC;IACJ,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"contextRecall.test.d.ts","sourceRoot":"","sources":["../../../test/assertions/contextRecall.test.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,243 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
|
|
3
|
+
if (k2 === undefined) k2 = k;
|
|
4
|
+
var desc = Object.getOwnPropertyDescriptor(m, k);
|
|
5
|
+
if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
|
|
6
|
+
desc = { enumerable: true, get: function() { return m[k]; } };
|
|
7
|
+
}
|
|
8
|
+
Object.defineProperty(o, k2, desc);
|
|
9
|
+
}) : (function(o, m, k, k2) {
|
|
10
|
+
if (k2 === undefined) k2 = k;
|
|
11
|
+
o[k2] = m[k];
|
|
12
|
+
}));
|
|
13
|
+
var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
|
|
14
|
+
Object.defineProperty(o, "default", { enumerable: true, value: v });
|
|
15
|
+
}) : function(o, v) {
|
|
16
|
+
o["default"] = v;
|
|
17
|
+
});
|
|
18
|
+
var __importStar = (this && this.__importStar) || (function () {
|
|
19
|
+
var ownKeys = function(o) {
|
|
20
|
+
ownKeys = Object.getOwnPropertyNames || function (o) {
|
|
21
|
+
var ar = [];
|
|
22
|
+
for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
|
|
23
|
+
return ar;
|
|
24
|
+
};
|
|
25
|
+
return ownKeys(o);
|
|
26
|
+
};
|
|
27
|
+
return function (mod) {
|
|
28
|
+
if (mod && mod.__esModule) return mod;
|
|
29
|
+
var result = {};
|
|
30
|
+
if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
|
|
31
|
+
__setModuleDefault(result, mod);
|
|
32
|
+
return result;
|
|
33
|
+
};
|
|
34
|
+
})();
|
|
35
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
36
|
+
const contextRecall_1 = require("../../src/assertions/contextRecall");
|
|
37
|
+
const matchers = __importStar(require("../../src/matchers"));
|
|
38
|
+
jest.mock('../../src/matchers');
|
|
39
|
+
describe('handleContextRecall', () => {
|
|
40
|
+
const mockMatchesContextRecall = jest.spyOn(matchers, 'matchesContextRecall');
|
|
41
|
+
beforeEach(() => {
|
|
42
|
+
jest.resetAllMocks();
|
|
43
|
+
});
|
|
44
|
+
it('should handle context recall with prompt context', async () => {
|
|
45
|
+
const mockResult = {
|
|
46
|
+
pass: true,
|
|
47
|
+
score: 0.9,
|
|
48
|
+
reason: 'Good recall',
|
|
49
|
+
};
|
|
50
|
+
mockMatchesContextRecall.mockResolvedValue(mockResult);
|
|
51
|
+
const mockProvider = {
|
|
52
|
+
id: () => 'test-provider',
|
|
53
|
+
callApi: jest.fn(),
|
|
54
|
+
};
|
|
55
|
+
const params = {
|
|
56
|
+
assertion: { type: 'context-recall', threshold: 0.8 },
|
|
57
|
+
renderedValue: 'test output',
|
|
58
|
+
prompt: 'test context',
|
|
59
|
+
test: {
|
|
60
|
+
vars: {},
|
|
61
|
+
options: {},
|
|
62
|
+
},
|
|
63
|
+
baseType: 'context-recall',
|
|
64
|
+
context: {
|
|
65
|
+
prompt: 'test context',
|
|
66
|
+
vars: {},
|
|
67
|
+
test: {
|
|
68
|
+
vars: {},
|
|
69
|
+
options: {},
|
|
70
|
+
},
|
|
71
|
+
logProbs: undefined,
|
|
72
|
+
provider: mockProvider,
|
|
73
|
+
providerResponse: undefined,
|
|
74
|
+
},
|
|
75
|
+
inverse: false,
|
|
76
|
+
output: 'test output',
|
|
77
|
+
outputString: 'test output',
|
|
78
|
+
provider: mockProvider,
|
|
79
|
+
providerResponse: {},
|
|
80
|
+
};
|
|
81
|
+
const result = await (0, contextRecall_1.handleContextRecall)(params);
|
|
82
|
+
expect(result).toEqual({
|
|
83
|
+
assertion: { type: 'context-recall', threshold: 0.8 },
|
|
84
|
+
...mockResult,
|
|
85
|
+
});
|
|
86
|
+
expect(mockMatchesContextRecall).toHaveBeenCalledWith('test context', 'test output', 0.8, {}, {});
|
|
87
|
+
});
|
|
88
|
+
it('should handle context recall with vars context', async () => {
|
|
89
|
+
const mockResult = {
|
|
90
|
+
pass: true,
|
|
91
|
+
score: 0.85,
|
|
92
|
+
reason: 'Good recall from vars',
|
|
93
|
+
};
|
|
94
|
+
mockMatchesContextRecall.mockResolvedValue(mockResult);
|
|
95
|
+
const mockProvider = {
|
|
96
|
+
id: () => 'test-provider',
|
|
97
|
+
callApi: jest.fn(),
|
|
98
|
+
};
|
|
99
|
+
const params = {
|
|
100
|
+
assertion: { type: 'context-recall', threshold: 0.7 },
|
|
101
|
+
renderedValue: 'test output',
|
|
102
|
+
prompt: 'original context',
|
|
103
|
+
test: {
|
|
104
|
+
vars: { context: 'context from vars' },
|
|
105
|
+
options: {},
|
|
106
|
+
},
|
|
107
|
+
baseType: 'context-recall',
|
|
108
|
+
context: {
|
|
109
|
+
prompt: 'original context',
|
|
110
|
+
vars: { context: 'context from vars' },
|
|
111
|
+
test: {
|
|
112
|
+
vars: { context: 'context from vars' },
|
|
113
|
+
options: {},
|
|
114
|
+
},
|
|
115
|
+
logProbs: undefined,
|
|
116
|
+
provider: mockProvider,
|
|
117
|
+
providerResponse: undefined,
|
|
118
|
+
},
|
|
119
|
+
inverse: false,
|
|
120
|
+
output: 'test output',
|
|
121
|
+
outputString: 'test output',
|
|
122
|
+
provider: mockProvider,
|
|
123
|
+
providerResponse: {},
|
|
124
|
+
};
|
|
125
|
+
const result = await (0, contextRecall_1.handleContextRecall)(params);
|
|
126
|
+
expect(result).toEqual({
|
|
127
|
+
assertion: { type: 'context-recall', threshold: 0.7 },
|
|
128
|
+
...mockResult,
|
|
129
|
+
});
|
|
130
|
+
expect(mockMatchesContextRecall).toHaveBeenCalledWith('context from vars', 'test output', 0.7, {}, { context: 'context from vars' });
|
|
131
|
+
});
|
|
132
|
+
it('should use default threshold of 0 when not provided', async () => {
|
|
133
|
+
const mockResult = {
|
|
134
|
+
pass: true,
|
|
135
|
+
score: 0.5,
|
|
136
|
+
reason: 'Default threshold test',
|
|
137
|
+
};
|
|
138
|
+
mockMatchesContextRecall.mockResolvedValue(mockResult);
|
|
139
|
+
const mockProvider = {
|
|
140
|
+
id: () => 'test-provider',
|
|
141
|
+
callApi: jest.fn(),
|
|
142
|
+
};
|
|
143
|
+
const params = {
|
|
144
|
+
assertion: { type: 'context-recall' },
|
|
145
|
+
renderedValue: 'test output',
|
|
146
|
+
prompt: 'test context',
|
|
147
|
+
test: {
|
|
148
|
+
vars: {},
|
|
149
|
+
options: {},
|
|
150
|
+
},
|
|
151
|
+
baseType: 'context-recall',
|
|
152
|
+
context: {
|
|
153
|
+
prompt: 'test context',
|
|
154
|
+
vars: {},
|
|
155
|
+
test: {
|
|
156
|
+
vars: {},
|
|
157
|
+
options: {},
|
|
158
|
+
},
|
|
159
|
+
logProbs: undefined,
|
|
160
|
+
provider: mockProvider,
|
|
161
|
+
providerResponse: undefined,
|
|
162
|
+
},
|
|
163
|
+
inverse: false,
|
|
164
|
+
output: 'test output',
|
|
165
|
+
outputString: 'test output',
|
|
166
|
+
provider: mockProvider,
|
|
167
|
+
providerResponse: {},
|
|
168
|
+
};
|
|
169
|
+
const result = await (0, contextRecall_1.handleContextRecall)(params);
|
|
170
|
+
expect(result).toEqual({
|
|
171
|
+
assertion: { type: 'context-recall' },
|
|
172
|
+
...mockResult,
|
|
173
|
+
});
|
|
174
|
+
expect(mockMatchesContextRecall).toHaveBeenCalledWith('test context', 'test output', 0, {}, {});
|
|
175
|
+
});
|
|
176
|
+
it('should throw error when renderedValue is not a string', async () => {
|
|
177
|
+
const mockProvider = {
|
|
178
|
+
id: () => 'test-provider',
|
|
179
|
+
callApi: jest.fn(),
|
|
180
|
+
};
|
|
181
|
+
const params = {
|
|
182
|
+
assertion: { type: 'context-recall' },
|
|
183
|
+
renderedValue: { value: 123 }, // Changed to object to match AssertionValue type
|
|
184
|
+
prompt: 'test context',
|
|
185
|
+
test: {
|
|
186
|
+
vars: {},
|
|
187
|
+
options: {},
|
|
188
|
+
},
|
|
189
|
+
baseType: 'context-recall',
|
|
190
|
+
context: {
|
|
191
|
+
prompt: 'test context',
|
|
192
|
+
vars: {},
|
|
193
|
+
test: {
|
|
194
|
+
vars: {},
|
|
195
|
+
options: {},
|
|
196
|
+
},
|
|
197
|
+
logProbs: undefined,
|
|
198
|
+
provider: mockProvider,
|
|
199
|
+
providerResponse: undefined,
|
|
200
|
+
},
|
|
201
|
+
inverse: false,
|
|
202
|
+
output: '123',
|
|
203
|
+
outputString: '123',
|
|
204
|
+
provider: mockProvider,
|
|
205
|
+
providerResponse: {},
|
|
206
|
+
};
|
|
207
|
+
await expect((0, contextRecall_1.handleContextRecall)(params)).rejects.toThrow('context-recall assertion type must have a string value');
|
|
208
|
+
});
|
|
209
|
+
it('should throw error when prompt is missing', async () => {
|
|
210
|
+
const mockProvider = {
|
|
211
|
+
id: () => 'test-provider',
|
|
212
|
+
callApi: jest.fn(),
|
|
213
|
+
};
|
|
214
|
+
const params = {
|
|
215
|
+
assertion: { type: 'context-recall' },
|
|
216
|
+
renderedValue: 'test output',
|
|
217
|
+
prompt: undefined,
|
|
218
|
+
test: {
|
|
219
|
+
vars: {},
|
|
220
|
+
options: {},
|
|
221
|
+
},
|
|
222
|
+
baseType: 'context-recall',
|
|
223
|
+
context: {
|
|
224
|
+
prompt: undefined,
|
|
225
|
+
vars: {},
|
|
226
|
+
test: {
|
|
227
|
+
vars: {},
|
|
228
|
+
options: {},
|
|
229
|
+
},
|
|
230
|
+
logProbs: undefined,
|
|
231
|
+
provider: mockProvider,
|
|
232
|
+
providerResponse: undefined,
|
|
233
|
+
},
|
|
234
|
+
inverse: false,
|
|
235
|
+
output: 'test output',
|
|
236
|
+
outputString: 'test output',
|
|
237
|
+
provider: mockProvider,
|
|
238
|
+
providerResponse: {},
|
|
239
|
+
};
|
|
240
|
+
await expect((0, contextRecall_1.handleContextRecall)(params)).rejects.toThrow('context-recall assertion type must have a prompt');
|
|
241
|
+
});
|
|
242
|
+
});
|
|
243
|
+
//# sourceMappingURL=contextRecall.test.js.map
|