promptfoo 0.20.0 → 0.21.0
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/README.md +1 -1
- package/dist/package.json +4 -4
- package/dist/src/assertions.d.ts.map +1 -1
- package/dist/src/assertions.js +5 -0
- package/dist/src/assertions.js.map +1 -1
- package/dist/src/evaluator.js +1 -1
- package/dist/src/evaluator.js.map +1 -1
- package/dist/src/index.d.ts +1 -5
- package/dist/src/index.d.ts.map +1 -1
- package/dist/src/index.js +1 -1
- package/dist/src/index.js.map +1 -1
- package/dist/src/matchers.d.ts +3 -2
- package/dist/src/matchers.d.ts.map +1 -1
- package/dist/src/matchers.js +37 -9
- package/dist/src/matchers.js.map +1 -1
- package/dist/src/providers/anthropic.d.ts +5 -3
- package/dist/src/providers/anthropic.d.ts.map +1 -1
- package/dist/src/providers/anthropic.js +8 -10
- package/dist/src/providers/anthropic.js.map +1 -1
- package/dist/src/providers/azureopenai.d.ts +9 -8
- package/dist/src/providers/azureopenai.d.ts.map +1 -1
- package/dist/src/providers/azureopenai.js +33 -36
- package/dist/src/providers/azureopenai.js.map +1 -1
- package/dist/src/providers/openai.d.ts +12 -12
- package/dist/src/providers/openai.d.ts.map +1 -1
- package/dist/src/providers/openai.js +54 -65
- package/dist/src/providers/openai.js.map +1 -1
- package/dist/src/providers/replicate.d.ts +4 -2
- package/dist/src/providers/replicate.d.ts.map +1 -1
- package/dist/src/providers/replicate.js +10 -8
- package/dist/src/providers/replicate.js.map +1 -1
- package/dist/src/providers/webhook.d.ts +9 -0
- package/dist/src/providers/webhook.d.ts.map +1 -0
- package/dist/src/providers/webhook.js +54 -0
- package/dist/src/providers/webhook.js.map +1 -0
- package/dist/src/providers.d.ts +1 -1
- package/dist/src/providers.d.ts.map +1 -1
- package/dist/src/providers.js +36 -28
- package/dist/src/providers.js.map +1 -1
- package/dist/src/suggestions.d.ts.map +1 -1
- package/dist/src/suggestions.js +1 -3
- package/dist/src/suggestions.js.map +1 -1
- package/dist/src/types.d.ts +7 -1
- package/dist/src/types.d.ts.map +1 -1
- package/dist/src/util.js +1 -1
- package/dist/src/util.js.map +1 -1
- package/dist/src/web/nextui/404/index.html +1 -1
- package/dist/src/web/nextui/404.html +1 -1
- package/dist/src/web/nextui/_next/static/Bl3o5lF4ON7Fjki46lPhr/_buildManifest.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/226-7bbb6c98a19542fd.js +37 -0
- package/dist/src/web/nextui/_next/static/chunks/249-ea9c0f034888ccff.js +125 -0
- package/dist/src/web/nextui/_next/static/chunks/339-501c32916b785ef1.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/365-e426ea5bc7e815fc.js +8 -0
- package/dist/src/web/nextui/_next/static/chunks/396-0a51429a01e24cdd.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/596-297f7ff4a0436e87.js +25 -0
- package/dist/src/web/nextui/_next/static/chunks/613-572c22424de64659.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/706-ae1d3352d28419e9.js +9 -0
- package/dist/src/web/nextui/_next/static/chunks/891-7035926a62c1c4e0.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/eval/[id]/not-found-366629541fd598e9.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/eval/[id]/page-319d2ee38d37574e.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/eval/page-a6b1ff91723b7beb.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/layout-024c4adc71c9feb0.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/page-1ae60660130041b2.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/setup/page-6ef16148040bf4f4.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/{ca377847-cb6ae6a6a073aebb.js → ca377847-26b462611379a4f7.js} +3 -3
- package/dist/src/web/nextui/_next/static/chunks/{fd9d1056-ac777be631f5a9e9.js → fd9d1056-fba4b53a2f01213b.js} +1 -1
- package/dist/src/web/nextui/_next/static/chunks/framework-8883d1e9be70c3da.js +25 -0
- package/dist/src/web/nextui/_next/static/chunks/main-8ea85465d428ecfe.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/main-app-581ccf0003955b21.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/pages/_app-52924524f99094ab.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/pages/_error-c92d5c4bb2b49926.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/webpack-55c264ce2fd85eb7.js +1 -0
- package/dist/src/web/nextui/_next/static/css/4d399fceacd06992.css +1 -0
- package/dist/src/web/nextui/eval/index.html +1 -1
- package/dist/src/web/nextui/eval/index.txt +6 -6
- package/dist/src/web/nextui/index.html +1 -1
- package/dist/src/web/nextui/index.txt +5 -5
- package/dist/src/web/nextui/setup/index.html +27 -1
- package/dist/src/web/nextui/setup/index.txt +9 -9
- package/dist/src/web/server.d.ts.map +1 -1
- package/dist/src/web/server.js +9 -5
- package/dist/src/web/server.js.map +1 -1
- package/package.json +4 -4
- package/dist/src/web/nextui/_next/static/US6gOx8LHTX_Hzm9aYNrC/_buildManifest.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/339-4fc8a80fa840e771.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/373-8a280796c0f2d1af.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/583-125d32af505e9bc4.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/596-07e4a23a5c6cdf04.js +0 -25
- package/dist/src/web/nextui/_next/static/chunks/658-a62210d07dc4dcb6.js +0 -15
- package/dist/src/web/nextui/_next/static/chunks/707-699cbd84b259c37b.js +0 -37
- package/dist/src/web/nextui/_next/static/chunks/858-ceb6fa22e614492b.js +0 -125
- package/dist/src/web/nextui/_next/static/chunks/891-3000ea7c0a292558.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/eval/[id]/not-found-50e40614fa05600e.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/eval/[id]/page-c19c44ed1b2dfb58.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/eval/page-d4a1813b2f8c4532.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/layout-664a8d716d2d24b1.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/page-1f8ef6a00a2355f0.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/setup/page-182018a3c6397345.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/framework-43665103d101a22d.js +0 -25
- package/dist/src/web/nextui/_next/static/chunks/main-50cc0a98559591ce.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/main-app-c9dc13756d166550.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/pages/_app-6b79a29ad0d63b21.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/pages/_error-9aeb3e4d490fe4b8.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/webpack-6e474e42be502dd7.js +0 -1
- package/dist/src/web/nextui/_next/static/css/a35c840ac696f161.css +0 -1
- package/dist/src/web/nextui/api +0 -1
- package/src/__mocks__/esm.ts +0 -3
- package/src/assertions.ts +0 -580
- package/src/cache.ts +0 -109
- package/src/esm.ts +0 -13
- package/src/evaluator.ts +0 -500
- package/src/index.ts +0 -52
- package/src/logger.ts +0 -46
- package/src/main.ts +0 -442
- package/src/matchers.ts +0 -120
- package/src/onboarding.ts +0 -69
- package/src/prompts.ts +0 -39
- package/src/providers/anthropic.ts +0 -88
- package/src/providers/azureopenai.ts +0 -299
- package/src/providers/llama.ts +0 -95
- package/src/providers/localai.ts +0 -111
- package/src/providers/ollama.ts +0 -89
- package/src/providers/openai.ts +0 -337
- package/src/providers/replicate.ts +0 -99
- package/src/providers/scriptCompletion.ts +0 -35
- package/src/providers/shared.ts +0 -34
- package/src/providers.ts +0 -192
- package/src/share.ts +0 -27
- package/src/suggestions.ts +0 -63
- package/src/table.ts +0 -43
- package/src/tableOutput.html +0 -52
- package/src/telemetry.ts +0 -70
- package/src/types.ts +0 -299
- package/src/updates.ts +0 -46
- package/src/util.ts +0 -543
- package/src/web/nextui/.eslintrc.json +0 -3
- package/src/web/nextui/next.config.js +0 -14
- package/src/web/nextui/package-lock.json +0 -4644
- package/src/web/nextui/package.json +0 -47
- package/src/web/nextui/public/favicon.ico +0 -0
- package/src/web/nextui/public/logo.svg +0 -30
- package/src/web/nextui/src/app/Home.css +0 -3
- package/src/web/nextui/src/app/api/route.ts +0 -6
- package/src/web/nextui/src/app/components/DarkMode.css +0 -22
- package/src/web/nextui/src/app/components/DarkMode.tsx +0 -17
- package/src/web/nextui/src/app/components/Logo.css +0 -32
- package/src/web/nextui/src/app/components/Logo.tsx +0 -11
- package/src/web/nextui/src/app/components/PageShell.css +0 -33
- package/src/web/nextui/src/app/components/PageShell.tsx +0 -87
- package/src/web/nextui/src/app/eval/ConfigModal.tsx +0 -84
- package/src/web/nextui/src/app/eval/Eval.css +0 -13
- package/src/web/nextui/src/app/eval/Eval.tsx +0 -79
- package/src/web/nextui/src/app/eval/EvalOutputPromptDialog.tsx +0 -127
- package/src/web/nextui/src/app/eval/ResultsCharts.tsx +0 -355
- package/src/web/nextui/src/app/eval/ResultsTable.css +0 -179
- package/src/web/nextui/src/app/eval/ResultsTable.tsx +0 -503
- package/src/web/nextui/src/app/eval/ResultsView.tsx +0 -301
- package/src/web/nextui/src/app/eval/ShareModal.tsx +0 -70
- package/src/web/nextui/src/app/eval/[id]/not-found.tsx +0 -5
- package/src/web/nextui/src/app/eval/[id]/page.css +0 -9
- package/src/web/nextui/src/app/eval/[id]/page.tsx +0 -20
- package/src/web/nextui/src/app/eval/index.css +0 -0
- package/src/web/nextui/src/app/eval/page.tsx +0 -8
- package/src/web/nextui/src/app/eval/store.ts +0 -18
- package/src/web/nextui/src/app/eval/types.ts +0 -20
- package/src/web/nextui/src/app/globals.css +0 -58
- package/src/web/nextui/src/app/layout.tsx +0 -25
- package/src/web/nextui/src/app/page.tsx +0 -7
- package/src/web/nextui/src/app/setup/AssertsForm.tsx +0 -118
- package/src/web/nextui/src/app/setup/PromptDialog.tsx +0 -77
- package/src/web/nextui/src/app/setup/PromptsSection.tsx +0 -190
- package/src/web/nextui/src/app/setup/ProviderConfigDialog.tsx +0 -99
- package/src/web/nextui/src/app/setup/ProviderSelector.tsx +0 -149
- package/src/web/nextui/src/app/setup/RunTestSuiteButton.tsx +0 -88
- package/src/web/nextui/src/app/setup/TestCaseDialog.tsx +0 -108
- package/src/web/nextui/src/app/setup/TestCasesSection.tsx +0 -154
- package/src/web/nextui/src/app/setup/VarsForm.tsx +0 -57
- package/src/web/nextui/src/app/setup/page.css +0 -3
- package/src/web/nextui/src/app/setup/page.tsx +0 -160
- package/src/web/nextui/src/util/api.ts +0 -1
- package/src/web/nextui/src/util/store.ts +0 -53
- package/src/web/nextui/tsconfig.json +0 -28
- package/src/web/server.ts +0 -151
- /package/dist/src/web/nextui/_next/static/{US6gOx8LHTX_Hzm9aYNrC → Bl3o5lF4ON7Fjki46lPhr}/_ssgManifest.js +0 -0
|
@@ -1,108 +0,0 @@
|
|
|
1
|
-
import React, { useState } from 'react';
|
|
2
|
-
import {
|
|
3
|
-
Button,
|
|
4
|
-
TextField,
|
|
5
|
-
Box,
|
|
6
|
-
Dialog,
|
|
7
|
-
DialogTitle,
|
|
8
|
-
DialogContent,
|
|
9
|
-
DialogActions,
|
|
10
|
-
} from '@mui/material';
|
|
11
|
-
import VarsForm from './VarsForm';
|
|
12
|
-
import AssertsForm from './AssertsForm';
|
|
13
|
-
import type { TestCase } from '../../../../../types';
|
|
14
|
-
|
|
15
|
-
interface TestCaseFormProps {
|
|
16
|
-
open: boolean;
|
|
17
|
-
onAdd: (testCase: TestCase, shouldClose: boolean) => void;
|
|
18
|
-
varsList: string[];
|
|
19
|
-
initialValues?: TestCase;
|
|
20
|
-
onCancel: () => void;
|
|
21
|
-
}
|
|
22
|
-
|
|
23
|
-
const TestCaseForm: React.FC<TestCaseFormProps> = ({
|
|
24
|
-
open,
|
|
25
|
-
onAdd,
|
|
26
|
-
varsList,
|
|
27
|
-
initialValues,
|
|
28
|
-
onCancel,
|
|
29
|
-
}) => {
|
|
30
|
-
const [description, setDescription] = useState(initialValues?.description || '');
|
|
31
|
-
const [vars, setVars] = useState(initialValues?.vars || {});
|
|
32
|
-
const [asserts, setAsserts] = useState(initialValues?.assert || []);
|
|
33
|
-
const [assertsFormKey, setAssertsFormKey] = useState(0);
|
|
34
|
-
|
|
35
|
-
React.useEffect(() => {
|
|
36
|
-
if (initialValues) {
|
|
37
|
-
setDescription(initialValues.description || '');
|
|
38
|
-
setVars(initialValues.vars || {});
|
|
39
|
-
setAsserts(initialValues.assert || []);
|
|
40
|
-
} else {
|
|
41
|
-
setDescription('');
|
|
42
|
-
setVars({});
|
|
43
|
-
setAsserts([]);
|
|
44
|
-
}
|
|
45
|
-
}, [initialValues]);
|
|
46
|
-
|
|
47
|
-
const handleAdd = (close: boolean) => {
|
|
48
|
-
onAdd(
|
|
49
|
-
{
|
|
50
|
-
description,
|
|
51
|
-
vars,
|
|
52
|
-
assert: asserts,
|
|
53
|
-
},
|
|
54
|
-
close,
|
|
55
|
-
);
|
|
56
|
-
if (close) {
|
|
57
|
-
onCancel();
|
|
58
|
-
}
|
|
59
|
-
setDescription('');
|
|
60
|
-
setVars({});
|
|
61
|
-
setAsserts([]);
|
|
62
|
-
setAssertsFormKey((prevKey) => prevKey + 1);
|
|
63
|
-
};
|
|
64
|
-
|
|
65
|
-
return (
|
|
66
|
-
<Dialog open={open} onClose={onCancel} fullWidth maxWidth="md">
|
|
67
|
-
<DialogTitle>{initialValues ? 'Edit Test Case' : 'Add Test Case'}</DialogTitle>
|
|
68
|
-
<DialogContent>
|
|
69
|
-
<Box>
|
|
70
|
-
{/*
|
|
71
|
-
<TextField
|
|
72
|
-
label="Description"
|
|
73
|
-
value={description}
|
|
74
|
-
onChange={(e) => setDescription(e.target.value)}
|
|
75
|
-
fullWidth
|
|
76
|
-
margin="normal"
|
|
77
|
-
/>
|
|
78
|
-
*/}
|
|
79
|
-
<VarsForm
|
|
80
|
-
onAdd={(vars) => setVars(vars)}
|
|
81
|
-
varsList={varsList}
|
|
82
|
-
initialValues={initialValues?.vars as Record<string, string>}
|
|
83
|
-
/>
|
|
84
|
-
<AssertsForm
|
|
85
|
-
key={assertsFormKey}
|
|
86
|
-
onAdd={(asserts) => setAsserts(asserts)}
|
|
87
|
-
initialValues={initialValues?.assert || []}
|
|
88
|
-
/>
|
|
89
|
-
</Box>
|
|
90
|
-
</DialogContent>
|
|
91
|
-
<DialogActions>
|
|
92
|
-
<Button onClick={handleAdd.bind(this, true)} color="primary" variant="contained">
|
|
93
|
-
{initialValues ? 'Update Test Case' : 'Add Test Case'}
|
|
94
|
-
</Button>
|
|
95
|
-
{!initialValues && (
|
|
96
|
-
<Button onClick={handleAdd.bind(this, false)} color="primary" variant="contained">
|
|
97
|
-
Add Another
|
|
98
|
-
</Button>
|
|
99
|
-
)}
|
|
100
|
-
<Button onClick={onCancel} color="secondary">
|
|
101
|
-
Cancel
|
|
102
|
-
</Button>
|
|
103
|
-
</DialogActions>
|
|
104
|
-
</Dialog>
|
|
105
|
-
);
|
|
106
|
-
};
|
|
107
|
-
|
|
108
|
-
export default TestCaseForm;
|
|
@@ -1,154 +0,0 @@
|
|
|
1
|
-
import React from 'react';
|
|
2
|
-
import Button from '@mui/material/Button';
|
|
3
|
-
import Copy from '@mui/icons-material/ContentCopy';
|
|
4
|
-
import Delete from '@mui/icons-material/Delete';
|
|
5
|
-
import Edit from '@mui/icons-material/Edit';
|
|
6
|
-
import IconButton from '@mui/material/IconButton';
|
|
7
|
-
import Stack from '@mui/material/Stack';
|
|
8
|
-
import Table from '@mui/material/Table';
|
|
9
|
-
import TableBody from '@mui/material/TableBody';
|
|
10
|
-
import TableCell from '@mui/material/TableCell';
|
|
11
|
-
import TableContainer from '@mui/material/TableContainer';
|
|
12
|
-
import TableHead from '@mui/material/TableHead';
|
|
13
|
-
import TableRow from '@mui/material/TableRow';
|
|
14
|
-
import Typography from '@mui/material/Typography';
|
|
15
|
-
|
|
16
|
-
import TestCaseDialog from './TestCaseDialog';
|
|
17
|
-
import { useStore } from '../../util/store';
|
|
18
|
-
|
|
19
|
-
import type { TestCase } from '../../../../../types';
|
|
20
|
-
|
|
21
|
-
interface TestCasesSectionProps {
|
|
22
|
-
varsList: string[];
|
|
23
|
-
}
|
|
24
|
-
|
|
25
|
-
const TestCasesSection: React.FC<TestCasesSectionProps> = ({ varsList }) => {
|
|
26
|
-
const { testCases, setTestCases } = useStore();
|
|
27
|
-
const [editingTestCaseIndex, setEditingTestCaseIndex] = React.useState<number | null>(null);
|
|
28
|
-
const [testCaseDialogOpen, setTestCaseDialogOpen] = React.useState(false);
|
|
29
|
-
|
|
30
|
-
const handleAddTestCase = (testCase: TestCase, shouldClose: boolean) => {
|
|
31
|
-
if (editingTestCaseIndex === null) {
|
|
32
|
-
setTestCases([...testCases, testCase]);
|
|
33
|
-
} else {
|
|
34
|
-
const updatedTestCases = testCases.map((tc, index) =>
|
|
35
|
-
index === editingTestCaseIndex ? testCase : tc,
|
|
36
|
-
);
|
|
37
|
-
setTestCases(updatedTestCases);
|
|
38
|
-
setEditingTestCaseIndex(null);
|
|
39
|
-
}
|
|
40
|
-
|
|
41
|
-
if (shouldClose) {
|
|
42
|
-
setTestCaseDialogOpen(false);
|
|
43
|
-
}
|
|
44
|
-
};
|
|
45
|
-
|
|
46
|
-
const handleRemoveTestCase = (event: React.MouseEvent, index: number) => {
|
|
47
|
-
event.stopPropagation();
|
|
48
|
-
|
|
49
|
-
if (confirm('Are you sure you want to delete this test case?')) {
|
|
50
|
-
setTestCases(testCases.filter((_, i) => i !== index));
|
|
51
|
-
}
|
|
52
|
-
};
|
|
53
|
-
|
|
54
|
-
const handleDuplicateTestCase = (event: React.MouseEvent, index: number) => {
|
|
55
|
-
event.stopPropagation();
|
|
56
|
-
const duplicatedTestCase = JSON.parse(JSON.stringify(testCases[index]));
|
|
57
|
-
setTestCases([...testCases, duplicatedTestCase]);
|
|
58
|
-
};
|
|
59
|
-
|
|
60
|
-
return (
|
|
61
|
-
<>
|
|
62
|
-
<Stack direction="row" spacing={2} justifyContent="space-between">
|
|
63
|
-
<Typography variant="h5">Test Cases</Typography>
|
|
64
|
-
<Button color="primary" onClick={() => setTestCaseDialogOpen(true)} variant="contained">
|
|
65
|
-
Add Test Case
|
|
66
|
-
</Button>
|
|
67
|
-
</Stack>
|
|
68
|
-
<TableContainer>
|
|
69
|
-
<Table>
|
|
70
|
-
<TableHead>
|
|
71
|
-
<TableRow>
|
|
72
|
-
<TableCell>Description</TableCell>
|
|
73
|
-
<TableCell>Assertions</TableCell>
|
|
74
|
-
<TableCell>Variables</TableCell>
|
|
75
|
-
<TableCell align="right"></TableCell>
|
|
76
|
-
</TableRow>
|
|
77
|
-
</TableHead>
|
|
78
|
-
<TableBody>
|
|
79
|
-
{testCases.length === 0 ? (
|
|
80
|
-
<TableRow>
|
|
81
|
-
<TableCell colSpan={4} align="center">
|
|
82
|
-
No test cases added yet.
|
|
83
|
-
</TableCell>
|
|
84
|
-
</TableRow>
|
|
85
|
-
) : (
|
|
86
|
-
testCases.map((testCase, index) => (
|
|
87
|
-
<TableRow
|
|
88
|
-
key={index}
|
|
89
|
-
sx={{
|
|
90
|
-
'&:hover': {
|
|
91
|
-
backgroundColor: 'rgba(0, 0, 0, 0.04)',
|
|
92
|
-
cursor: 'pointer',
|
|
93
|
-
},
|
|
94
|
-
}}
|
|
95
|
-
onClick={() => {
|
|
96
|
-
setEditingTestCaseIndex(index);
|
|
97
|
-
setTestCaseDialogOpen(true);
|
|
98
|
-
}}
|
|
99
|
-
>
|
|
100
|
-
<TableCell>
|
|
101
|
-
<Typography variant="body2">
|
|
102
|
-
{testCase.description || `Test Case #${index + 1}`}
|
|
103
|
-
</Typography>
|
|
104
|
-
</TableCell>
|
|
105
|
-
<TableCell>{testCase.assert?.length || 0} assertions</TableCell>
|
|
106
|
-
<TableCell>
|
|
107
|
-
{Object.entries(testCase.vars || {})
|
|
108
|
-
.map(([k, v]) => k + '=' + v)
|
|
109
|
-
.join(', ')}
|
|
110
|
-
</TableCell>
|
|
111
|
-
<TableCell align="right" sx={{ minWidth: 150 }}>
|
|
112
|
-
<IconButton
|
|
113
|
-
onClick={() => {
|
|
114
|
-
setEditingTestCaseIndex(index);
|
|
115
|
-
setTestCaseDialogOpen(true);
|
|
116
|
-
}}
|
|
117
|
-
size="small"
|
|
118
|
-
>
|
|
119
|
-
<Edit />
|
|
120
|
-
</IconButton>
|
|
121
|
-
<IconButton
|
|
122
|
-
onClick={(event) => handleDuplicateTestCase(event, index)}
|
|
123
|
-
size="small"
|
|
124
|
-
>
|
|
125
|
-
<Copy />
|
|
126
|
-
</IconButton>
|
|
127
|
-
<IconButton
|
|
128
|
-
onClick={(event) => handleRemoveTestCase(event, index)}
|
|
129
|
-
size="small"
|
|
130
|
-
>
|
|
131
|
-
<Delete />
|
|
132
|
-
</IconButton>
|
|
133
|
-
</TableCell>
|
|
134
|
-
</TableRow>
|
|
135
|
-
))
|
|
136
|
-
)}
|
|
137
|
-
</TableBody>
|
|
138
|
-
</Table>
|
|
139
|
-
</TableContainer>
|
|
140
|
-
<TestCaseDialog
|
|
141
|
-
open={testCaseDialogOpen}
|
|
142
|
-
onAdd={handleAddTestCase}
|
|
143
|
-
varsList={varsList}
|
|
144
|
-
initialValues={editingTestCaseIndex !== null ? testCases[editingTestCaseIndex] : undefined}
|
|
145
|
-
onCancel={() => {
|
|
146
|
-
setEditingTestCaseIndex(null);
|
|
147
|
-
setTestCaseDialogOpen(false);
|
|
148
|
-
}}
|
|
149
|
-
/>
|
|
150
|
-
</>
|
|
151
|
-
);
|
|
152
|
-
};
|
|
153
|
-
|
|
154
|
-
export default TestCasesSection;
|
|
@@ -1,57 +0,0 @@
|
|
|
1
|
-
import React, { useEffect } from 'react';
|
|
2
|
-
import { Box, TextField, Typography, Stack } from '@mui/material';
|
|
3
|
-
|
|
4
|
-
interface VarsFormProps {
|
|
5
|
-
onAdd: (vars: Record<string, string>) => void;
|
|
6
|
-
varsList: string[];
|
|
7
|
-
initialValues?: Record<string, string>;
|
|
8
|
-
}
|
|
9
|
-
|
|
10
|
-
const VarsForm: React.FC<VarsFormProps> = ({ onAdd, varsList, initialValues }) => {
|
|
11
|
-
const [vars, setVars] = React.useState<Record<string, string>>(initialValues || {});
|
|
12
|
-
|
|
13
|
-
useEffect(() => {
|
|
14
|
-
const newVars: Record<string, string> = {};
|
|
15
|
-
varsList.forEach((v) => {
|
|
16
|
-
newVars[v] = initialValues?.[v] || '';
|
|
17
|
-
});
|
|
18
|
-
setVars(newVars);
|
|
19
|
-
}, [varsList, initialValues]);
|
|
20
|
-
|
|
21
|
-
return (
|
|
22
|
-
<Box my={2}>
|
|
23
|
-
<Typography variant="h6" mb={2}>
|
|
24
|
-
Vars
|
|
25
|
-
</Typography>
|
|
26
|
-
{varsList.length > 0 ? (
|
|
27
|
-
<Stack direction="row" spacing={2} alignItems="center">
|
|
28
|
-
{Object.keys(vars).map((varName, index) => (
|
|
29
|
-
<Stack key={index} direction="row" spacing={2} alignItems="center">
|
|
30
|
-
<TextField
|
|
31
|
-
placeholder={varName}
|
|
32
|
-
label={varName}
|
|
33
|
-
value={vars[varName]}
|
|
34
|
-
fullWidth
|
|
35
|
-
onChange={(e) => {
|
|
36
|
-
const newValue = e.target.value;
|
|
37
|
-
const newVars = {
|
|
38
|
-
...vars,
|
|
39
|
-
[varName]: newValue,
|
|
40
|
-
};
|
|
41
|
-
setVars(newVars);
|
|
42
|
-
onAdd(newVars);
|
|
43
|
-
}}
|
|
44
|
-
/>
|
|
45
|
-
</Stack>
|
|
46
|
-
))}
|
|
47
|
-
</Stack>
|
|
48
|
-
) : (
|
|
49
|
-
<Typography variant="subtitle1" gutterBottom>
|
|
50
|
-
Add variables to your prompt using the {'{{varname}}'} syntax.
|
|
51
|
-
</Typography>
|
|
52
|
-
)}
|
|
53
|
-
</Box>
|
|
54
|
-
);
|
|
55
|
-
};
|
|
56
|
-
|
|
57
|
-
export default VarsForm;
|
|
@@ -1,160 +0,0 @@
|
|
|
1
|
-
'use client';
|
|
2
|
-
|
|
3
|
-
import React, { useState, useEffect } from 'react';
|
|
4
|
-
import Link from 'next/link';
|
|
5
|
-
import yaml from 'js-yaml';
|
|
6
|
-
import { Light as SyntaxHighlighter } from 'react-syntax-highlighter';
|
|
7
|
-
import { docco } from 'react-syntax-highlighter/dist/cjs/styles/hljs';
|
|
8
|
-
import Button from '@mui/material/Button';
|
|
9
|
-
import Container from '@mui/material/Container';
|
|
10
|
-
import Typography from '@mui/material/Typography';
|
|
11
|
-
import Box from '@mui/material/Box';
|
|
12
|
-
import Stack from '@mui/material/Stack';
|
|
13
|
-
import Dialog from '@mui/material/Dialog';
|
|
14
|
-
import DialogActions from '@mui/material/DialogActions';
|
|
15
|
-
import DialogContent from '@mui/material/DialogContent';
|
|
16
|
-
import DialogContentText from '@mui/material/DialogContentText';
|
|
17
|
-
import DialogTitle from '@mui/material/DialogTitle';
|
|
18
|
-
|
|
19
|
-
import RunTestSuiteButton from './RunTestSuiteButton';
|
|
20
|
-
import PromptsSection from './PromptsSection';
|
|
21
|
-
import TestCasesSection from './TestCasesSection';
|
|
22
|
-
import ProviderSelector from './ProviderSelector';
|
|
23
|
-
import { useStore } from '../../util/store';
|
|
24
|
-
|
|
25
|
-
import './page.css';
|
|
26
|
-
|
|
27
|
-
const EvaluateTestSuiteCreator: React.FC = () => {
|
|
28
|
-
const [yamlString, setYamlString] = useState('');
|
|
29
|
-
const [resetDialogOpen, setResetDialogOpen] = useState(false);
|
|
30
|
-
|
|
31
|
-
const {
|
|
32
|
-
description,
|
|
33
|
-
setDescription,
|
|
34
|
-
providers,
|
|
35
|
-
setProviders,
|
|
36
|
-
prompts,
|
|
37
|
-
setPrompts,
|
|
38
|
-
testCases,
|
|
39
|
-
setTestCases,
|
|
40
|
-
} = useStore();
|
|
41
|
-
|
|
42
|
-
useEffect(() => {
|
|
43
|
-
useStore.persist.rehydrate();
|
|
44
|
-
}, []);
|
|
45
|
-
|
|
46
|
-
useEffect(() => {
|
|
47
|
-
const testSuite = {
|
|
48
|
-
description,
|
|
49
|
-
providers,
|
|
50
|
-
prompts,
|
|
51
|
-
tests: testCases,
|
|
52
|
-
};
|
|
53
|
-
setYamlString(yaml.dump(testSuite));
|
|
54
|
-
}, [description, providers, prompts, testCases]);
|
|
55
|
-
|
|
56
|
-
if (process.env.NEXT_PUBLIC_NO_BROWSING) {
|
|
57
|
-
return null;
|
|
58
|
-
}
|
|
59
|
-
|
|
60
|
-
const extractVarsFromPrompts = (prompts: string[]): string[] => {
|
|
61
|
-
const varRegex = /{{(\w+)}}/g;
|
|
62
|
-
const varsSet = new Set<string>();
|
|
63
|
-
|
|
64
|
-
prompts.forEach((prompt) => {
|
|
65
|
-
let match;
|
|
66
|
-
while ((match = varRegex.exec(prompt)) !== null) {
|
|
67
|
-
varsSet.add(match[1]);
|
|
68
|
-
}
|
|
69
|
-
});
|
|
70
|
-
|
|
71
|
-
return Array.from(varsSet);
|
|
72
|
-
};
|
|
73
|
-
|
|
74
|
-
const varsList = extractVarsFromPrompts(prompts);
|
|
75
|
-
|
|
76
|
-
const handleReset = () => {
|
|
77
|
-
setDescription('');
|
|
78
|
-
setProviders([]);
|
|
79
|
-
setPrompts([]);
|
|
80
|
-
setTestCases([]);
|
|
81
|
-
setYamlString('');
|
|
82
|
-
setResetDialogOpen(false);
|
|
83
|
-
};
|
|
84
|
-
|
|
85
|
-
return (
|
|
86
|
-
<Container maxWidth="lg" sx={{ marginTop: '2rem' }}>
|
|
87
|
-
<Stack direction="row" spacing={2} justifyContent="space-between">
|
|
88
|
-
<Typography variant="h4">Set up an evaluation</Typography>
|
|
89
|
-
<Stack direction="row" spacing={2}>
|
|
90
|
-
<RunTestSuiteButton />
|
|
91
|
-
<Button variant="outlined" color="primary" onClick={() => setResetDialogOpen(true)}>
|
|
92
|
-
Reset
|
|
93
|
-
</Button>
|
|
94
|
-
</Stack>
|
|
95
|
-
</Stack>
|
|
96
|
-
<Box mt={4} />
|
|
97
|
-
{/*
|
|
98
|
-
<Box mt={4}>
|
|
99
|
-
<TextField
|
|
100
|
-
label="Description"
|
|
101
|
-
value={description}
|
|
102
|
-
onChange={(e) => {
|
|
103
|
-
setDescription(e.target.value);
|
|
104
|
-
}}
|
|
105
|
-
fullWidth
|
|
106
|
-
margin="normal"
|
|
107
|
-
/>
|
|
108
|
-
</Box>
|
|
109
|
-
*/}
|
|
110
|
-
<Box mt={2}>
|
|
111
|
-
<Stack direction="column" spacing={2} justifyContent="space-between">
|
|
112
|
-
<Typography variant="h5">Providers</Typography>
|
|
113
|
-
<ProviderSelector providers={providers} onChange={setProviders} />
|
|
114
|
-
</Stack>
|
|
115
|
-
</Box>
|
|
116
|
-
<Box mt={4} />
|
|
117
|
-
<PromptsSection />
|
|
118
|
-
<Box mt={6} />
|
|
119
|
-
<TestCasesSection varsList={varsList} />
|
|
120
|
-
<Box mt={8}>
|
|
121
|
-
{yamlString && (
|
|
122
|
-
<Box mt={4}>
|
|
123
|
-
<Typography variant="h5" gutterBottom>
|
|
124
|
-
YAML config
|
|
125
|
-
</Typography>
|
|
126
|
-
<Typography variant="body1" gutterBottom>
|
|
127
|
-
This is the evaluation config that is run by promptfoo. See{' '}
|
|
128
|
-
<Link href="https://promptfoo.dev/docs/configuration/guide">configuration docs</Link>{' '}
|
|
129
|
-
to learn more.
|
|
130
|
-
</Typography>
|
|
131
|
-
<SyntaxHighlighter className="yaml-config" language="yaml" style={docco}>
|
|
132
|
-
{yamlString}
|
|
133
|
-
</SyntaxHighlighter>
|
|
134
|
-
</Box>
|
|
135
|
-
)}
|
|
136
|
-
</Box>
|
|
137
|
-
<Dialog
|
|
138
|
-
open={resetDialogOpen}
|
|
139
|
-
onClose={() => setResetDialogOpen(false)}
|
|
140
|
-
aria-labelledby="alert-dialog-title"
|
|
141
|
-
aria-describedby="alert-dialog-description"
|
|
142
|
-
>
|
|
143
|
-
<DialogTitle id="alert-dialog-title">{'Confirm Reset'}</DialogTitle>
|
|
144
|
-
<DialogContent>
|
|
145
|
-
<DialogContentText id="alert-dialog-description">
|
|
146
|
-
Are you sure you want to reset all the fields? This action cannot be undone.
|
|
147
|
-
</DialogContentText>
|
|
148
|
-
</DialogContent>
|
|
149
|
-
<DialogActions>
|
|
150
|
-
<Button onClick={() => setResetDialogOpen(false)}>Cancel</Button>
|
|
151
|
-
<Button onClick={handleReset} autoFocus>
|
|
152
|
-
Reset
|
|
153
|
-
</Button>
|
|
154
|
-
</DialogActions>
|
|
155
|
-
</Dialog>
|
|
156
|
-
</Container>
|
|
157
|
-
);
|
|
158
|
-
};
|
|
159
|
-
|
|
160
|
-
export default EvaluateTestSuiteCreator;
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
export const API_BASE_URL = `http://localhost:15500`;
|
|
@@ -1,53 +0,0 @@
|
|
|
1
|
-
import { create } from 'zustand';
|
|
2
|
-
import { persist } from 'zustand/middleware';
|
|
3
|
-
|
|
4
|
-
import type { Assertion, ProviderOptions, TestCase } from '../../../../types';
|
|
5
|
-
|
|
6
|
-
export interface State {
|
|
7
|
-
asserts: Assertion[];
|
|
8
|
-
testCases: TestCase[];
|
|
9
|
-
description: string;
|
|
10
|
-
providers: ProviderOptions[];
|
|
11
|
-
prompts: string[];
|
|
12
|
-
setAsserts: (asserts: Assertion[]) => void;
|
|
13
|
-
setTestCases: (testCases: TestCase[]) => void;
|
|
14
|
-
setDescription: (description: string) => void;
|
|
15
|
-
setProviders: (providers: ProviderOptions[]) => void;
|
|
16
|
-
setPrompts: (prompts: string[]) => void;
|
|
17
|
-
}
|
|
18
|
-
|
|
19
|
-
export const useStore = create<State>()(
|
|
20
|
-
persist(
|
|
21
|
-
(set) => ({
|
|
22
|
-
asserts: [],
|
|
23
|
-
testCases: [],
|
|
24
|
-
description: '',
|
|
25
|
-
providers: [],
|
|
26
|
-
prompts: [],
|
|
27
|
-
setAsserts: (asserts) => set({ asserts }),
|
|
28
|
-
setTestCases: (testCases) => set({ testCases }),
|
|
29
|
-
setDescription: (description) => set({ description }),
|
|
30
|
-
setProviders: (providers) => set({ providers }),
|
|
31
|
-
setPrompts: (prompts) => set({ prompts }),
|
|
32
|
-
}),
|
|
33
|
-
{
|
|
34
|
-
name: 'promptfoo',
|
|
35
|
-
skipHydration: true,
|
|
36
|
-
},
|
|
37
|
-
),
|
|
38
|
-
);
|
|
39
|
-
|
|
40
|
-
/*
|
|
41
|
-
export const useStore = create<State>((set) => ({
|
|
42
|
-
asserts: [],
|
|
43
|
-
testCases: [],
|
|
44
|
-
description: '',
|
|
45
|
-
providers: [],
|
|
46
|
-
prompts: [],
|
|
47
|
-
setAsserts: (asserts) => set({ asserts }),
|
|
48
|
-
setTestCases: (testCases) => set({ testCases }),
|
|
49
|
-
setDescription: (description) => set({ description }),
|
|
50
|
-
setProviders: (providers) => set({ providers }),
|
|
51
|
-
setPrompts: (prompts) => set({ prompts }),
|
|
52
|
-
}));
|
|
53
|
-
*/
|
|
@@ -1,28 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"compilerOptions": {
|
|
3
|
-
"target": "es5",
|
|
4
|
-
"lib": ["dom", "dom.iterable", "esnext"],
|
|
5
|
-
"allowJs": true,
|
|
6
|
-
"skipLibCheck": true,
|
|
7
|
-
"strict": true,
|
|
8
|
-
"forceConsistentCasingInFileNames": true,
|
|
9
|
-
"noEmit": true,
|
|
10
|
-
"esModuleInterop": true,
|
|
11
|
-
"module": "esnext",
|
|
12
|
-
"moduleResolution": "bundler",
|
|
13
|
-
"resolveJsonModule": true,
|
|
14
|
-
"isolatedModules": true,
|
|
15
|
-
"jsx": "preserve",
|
|
16
|
-
"incremental": true,
|
|
17
|
-
"plugins": [
|
|
18
|
-
{
|
|
19
|
-
"name": "next"
|
|
20
|
-
}
|
|
21
|
-
],
|
|
22
|
-
"paths": {
|
|
23
|
-
"@/*": ["./src/*"]
|
|
24
|
-
}
|
|
25
|
-
},
|
|
26
|
-
"include": ["next-env.d.ts", "**/*.ts", "**/*.tsx", ".next/types/**/*.ts"],
|
|
27
|
-
"exclude": ["node_modules"]
|
|
28
|
-
}
|