npm - promptfoo - Versions diffs - 0.19.2 → 0.20.0 - Mend

promptfoo 0.19.2 → 0.20.0

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.

Files changed (80) hide show

package/dist/src/web/nextui/setup/index.html CHANGED Viewed

	@@ -1 +1 @@
1	- <!DOCTYPE html><html lang="en"><head><meta charSet="utf-8"/><link rel="preload" as="font" href="/_next/static/media/0e4fe491bf84089c-s.p.woff2" crossorigin="" type="font/woff2"/><link rel="preload" as="font" href="/_next/static/media/627622453ef56b0d-s.p.woff2" crossorigin="" type="font/woff2"/><link rel="preload" as="font" href="/_next/static/media/934c4b7cb736f2a3-s.p.woff2" crossorigin="" type="font/woff2"/><link rel="stylesheet" href="/_next/static/css/e388dd377baf25ec.css" data-precedence="next"/><link rel="stylesheet" href="/_next/static/css/a35c840ac696f161.css" data-precedence="next"/><link rel="preload" href="/_next/static/chunks/webpack-a886dd767c2e76b7.js" as="script"/><link rel="preload" href="/_next/static/chunks/fd9d1056-d8847af536b5787b.js" as="script"/><link rel="preload" href="/_next/static/chunks/596-9c29c47b8dee7a50.js" as="script"/><link rel="preload" href="/_next/static/chunks/main-app-581ccf0003955b21.js" as="script"/><title>promptfoo</title><meta name="description" content="LLM testing and evaluation"/><meta name="viewport" content="width=device-width, initial-scale=1"/><meta name="next-size-adjust"/><script src="/_next/static/chunks/polyfills-78c92fac7aa8fdd8.js" noModule=""></script></head><body class="__className_979022"><div><style data-emotion="css jj2ztu">.css-jj2ztu{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-flex-direction:row;-ms-flex-direction:row;flex-direction:row;}.css-jj2ztu>:not(style):not(style){margin:0;}.css-jj2ztu>:not(style)~:not(style){margin-left:16px;}</style><div class="MuiStack-root nav css-jj2ztu"><div class="logo MuiBox-root css-0"><img src="/logo.svg" alt="Promptfoo logo"/> <span>promptfoo</span></div><a href="/setup/">New Eval</a><a href="/eval/">View Evals</a><div class="dark-mode-toggle"><style data-emotion="css vubbuv">.css-vubbuv{-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;width:1em;height:1em;display:inline-block;fill:currentColor;-webkit-flex-shrink:0;-ms-flex-negative:0;flex-shrink:0;-webkit-transition:fill 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:fill 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;font-size:1.5rem;}</style><svg class="MuiSvgIcon-root MuiSvgIcon-fontSizeMedium css-vubbuv" focusable="false" aria-hidden="true" viewBox="0 0 24 24" data-testid="LightModeIcon"><path d="M12 7c-2.76 0-5 2.24-5 5s2.24 5 5 5 5-2.24 5-5-2.24-5-5-5zM2 13h2c.55 0 1-.45 1-1s-.45-1-1-1H2c-.55 0-1 .45-1 1s.45 1 1 1zm18 0h2c.55 0 1-.45 1-1s-.45-1-1-1h-2c-.55 0-1 .45-1 1s.45 1 1 1zM11 2v2c0 .55.45 1 1 1s1-.45 1-1V2c0-.55-.45-1-1-1s-1 .45-1 1zm0 18v2c0 .55.45 1 1 1s1-.45 1-1v-2c0-.55-.45-1-1-1s-1 .45-1 1zM5.99 4.58c-.39-.39-1.03-.39-1.41 0-.39.39-.39 1.03 0 1.41l1.06 1.06c.39.39 1.03.39 1.41 0s.39-1.03 0-1.41L5.99 4.58zm12.37 12.37c-.39-.39-1.03-.39-1.41 0-.39.39-.39 1.03 0 1.41l1.06 1.06c.39.39 1.03.39 1.41 0 .39-.39.39-1.03 0-1.41l-1.06-1.06zm1.06-10.96c.39-.39.39-1.03 0-1.41-.39-.39-1.03-.39-1.41 0l-1.06 1.06c-.39.39-.39 1.03 0 1.41s1.03.39 1.41 0l1.06-1.06zM7.05 18.36c.39-.39.39-1.03 0-1.41-.39-.39-1.03-.39-1.41 0l-1.06 1.06c-.39.39-.39 1.03 0 1.41s1.03.39 1.41 0l1.06-1.06z"></path></svg></div></div><div><style data-emotion="css 19t2hy6">.css-19t2hy6{width:100%;margin-left:auto;box-sizing:border-box;margin-right:auto;display:block;padding-left:16px;padding-right:16px;margin-top:2rem;}@media (min-width:600px){.css-19t2hy6{padding-left:24px;padding-right:24px;}}@media (min-width:1200px){.css-19t2hy6{max-width:1200px;}}</style><div class="MuiContainer-root MuiContainer-maxWidthLg css-19t2hy6"><style data-emotion="css 1w57jtv">.css-1w57jtv{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-flex-direction:row;-ms-flex-direction:row;flex-direction:row;-webkit-box-pack:justify;-webkit-justify-content:space-between;justify-content:space-between;}.css-1w57jtv>:not(style):not(style){margin:0;}.css-1w57jtv>:not(style)~:not(style){margin-left:16px;}</style><div class="MuiStack-root css-1w57jtv"><style data-emotion="css jqhduh">.css-jqhduh{margin:0;font-family:inherit;font-weight:400;font-size:2.125rem;line-height:1.235;}</style><h4 class="MuiTypography-root MuiTypography-h4 css-jqhduh">Set up an evaluation</h4><style data-emotion="css jj2ztu">.css-jj2ztu{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-flex-direction:row;-ms-flex-direction:row;flex-direction:row;}.css-jj2ztu>:not(style):not(style){margin:0;}.css-jj2ztu>:not(style)~:not(style){margin-left:16px;}</style><div class="MuiStack-root css-jj2ztu"><style data-emotion="css 11kchfx">.css-11kchfx{font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.75;text-transform:uppercase;min-width:64px;padding:6px 16px;border-radius:4px;-webkit-transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;color:#fff;background-color:#1976d2;box-shadow:0px 3px 1px -2px rgba(0,0,0,0.2),0px 2px 2px 0px rgba(0,0,0,0.14),0px 1px 5px 0px rgba(0,0,0,0.12);}.css-11kchfx:hover{-webkit-text-decoration:none;text-decoration:none;background-color:#1565c0;box-shadow:0px 2px 4px -1px rgba(0,0,0,0.2),0px 4px 5px 0px rgba(0,0,0,0.14),0px 1px 10px 0px rgba(0,0,0,0.12);}@media (hover: none){.css-11kchfx:hover{background-color:#1976d2;}}.css-11kchfx:active{box-shadow:0px 5px 5px -3px rgba(0,0,0,0.2),0px 8px 10px 1px rgba(0,0,0,0.14),0px 3px 14px 2px rgba(0,0,0,0.12);}.css-11kchfx.Mui-focusVisible{box-shadow:0px 3px 5px -1px rgba(0,0,0,0.2),0px 6px 10px 0px rgba(0,0,0,0.14),0px 1px 18px 0px rgba(0,0,0,0.12);}.css-11kchfx.Mui-disabled{color:rgba(0, 0, 0, 0.26);box-shadow:none;background-color:rgba(0, 0, 0, 0.12);}</style><style data-emotion="css 1r4sjx4">.css-1r4sjx4{display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;-webkit-box-pack:center;-ms-flex-pack:center;-webkit-justify-content:center;justify-content:center;position:relative;box-sizing:border-box;-webkit-tap-highlight-color:transparent;background-color:transparent;outline:0;border:0;margin:0;border-radius:0;padding:0;cursor:pointer;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;vertical-align:middle;-moz-appearance:none;-webkit-appearance:none;-webkit-text-decoration:none;text-decoration:none;color:inherit;font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.75;text-transform:uppercase;min-width:64px;padding:6px 16px;border-radius:4px;-webkit-transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;color:#fff;background-color:#1976d2;box-shadow:0px 3px 1px -2px rgba(0,0,0,0.2),0px 2px 2px 0px rgba(0,0,0,0.14),0px 1px 5px 0px rgba(0,0,0,0.12);}.css-1r4sjx4::-moz-focus-inner{border-style:none;}.css-1r4sjx4.Mui-disabled{pointer-events:none;cursor:default;}@media print{.css-1r4sjx4{-webkit-print-color-adjust:exact;color-adjust:exact;}}.css-1r4sjx4:hover{-webkit-text-decoration:none;text-decoration:none;background-color:#1565c0;box-shadow:0px 2px 4px -1px rgba(0,0,0,0.2),0px 4px 5px 0px rgba(0,0,0,0.14),0px 1px 10px 0px rgba(0,0,0,0.12);}@media (hover: none){.css-1r4sjx4:hover{background-color:#1976d2;}}.css-1r4sjx4:active{box-shadow:0px 5px 5px -3px rgba(0,0,0,0.2),0px 8px 10px 1px rgba(0,0,0,0.14),0px 3px 14px 2px rgba(0,0,0,0.12);}.css-1r4sjx4.Mui-focusVisible{box-shadow:0px 3px 5px -1px rgba(0,0,0,0.2),0px 6px 10px 0px rgba(0,0,0,0.14),0px 1px 18px 0px rgba(0,0,0,0.12);}.css-1r4sjx4.Mui-disabled{color:rgba(0, 0, 0, 0.26);box-shadow:none;background-color:rgba(0, 0, 0, 0.12);}</style><button class="MuiButtonBase-root MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium css-1r4sjx4" tabindex="0" type="button">Run Evaluation</button><style data-emotion="css 1l5hj2g">.css-1l5hj2g{font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.75;text-transform:uppercase;min-width:64px;padding:5px 15px;border-radius:4px;-webkit-transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;border:1px solid rgba(25, 118, 210, 0.5);color:#1976d2;}.css-1l5hj2g:hover{-webkit-text-decoration:none;text-decoration:none;background-color:rgba(25, 118, 210, 0.04);border:1px solid #1976d2;}@media (hover: none){.css-1l5hj2g:hover{background-color:transparent;}}.css-1l5hj2g.Mui-disabled{color:rgba(0, 0, 0, 0.26);border:1px solid rgba(0, 0, 0, 0.12);}</style><style data-emotion="css 1avs8v6">.css-1avs8v6{display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;-webkit-box-pack:center;-ms-flex-pack:center;-webkit-justify-content:center;justify-content:center;position:relative;box-sizing:border-box;-webkit-tap-highlight-color:transparent;background-color:transparent;outline:0;border:0;margin:0;border-radius:0;padding:0;cursor:pointer;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;vertical-align:middle;-moz-appearance:none;-webkit-appearance:none;-webkit-text-decoration:none;text-decoration:none;color:inherit;font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.75;text-transform:uppercase;min-width:64px;padding:5px 15px;border-radius:4px;-webkit-transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;border:1px solid rgba(25, 118, 210, 0.5);color:#1976d2;}.css-1avs8v6::-moz-focus-inner{border-style:none;}.css-1avs8v6.Mui-disabled{pointer-events:none;cursor:default;}@media print{.css-1avs8v6{-webkit-print-color-adjust:exact;color-adjust:exact;}}.css-1avs8v6:hover{-webkit-text-decoration:none;text-decoration:none;background-color:rgba(25, 118, 210, 0.04);border:1px solid #1976d2;}@media (hover: none){.css-1avs8v6:hover{background-color:transparent;}}.css-1avs8v6.Mui-disabled{color:rgba(0, 0, 0, 0.26);border:1px solid rgba(0, 0, 0, 0.12);}</style><button class="MuiButtonBase-root MuiButton-root MuiButton-outlined MuiButton-outlinedPrimary MuiButton-sizeMedium MuiButton-outlinedSizeMedium MuiButton-root MuiButton-outlined MuiButton-outlinedPrimary MuiButton-sizeMedium MuiButton-outlinedSizeMedium css-1avs8v6" tabindex="0" type="button">Reset</button></div></div><style data-emotion="css h5fkc8">.css-h5fkc8{margin-top:32px;}</style><div class="MuiBox-root css-h5fkc8"></div><style data-emotion="css 1yuhvjn">.css-1yuhvjn{margin-top:16px;}</style><div class="MuiBox-root css-1yuhvjn"><style data-emotion="css 1nguorl">.css-1nguorl{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-flex-direction:column;-ms-flex-direction:column;flex-direction:column;-webkit-box-pack:justify;-webkit-justify-content:space-between;justify-content:space-between;}.css-1nguorl>:not(style):not(style){margin:0;}.css-1nguorl>:not(style)~:not(style){margin-top:16px;}</style><div class="MuiStack-root css-1nguorl"><style data-emotion="css 1d79rw6">.css-1d79rw6{margin:0;font-family:inherit;font-weight:400;font-size:1.5rem;line-height:1.334;}</style><h5 class="MuiTypography-root MuiTypography-h5 css-1d79rw6">Providers</h5><div class="MuiBox-root css-1yuhvjn"><style data-emotion="css 18nc3u2">.css-18nc3u2.Mui-focused .MuiAutocomplete-clearIndicator{visibility:visible;}@media (pointer: fine){.css-18nc3u2:hover .MuiAutocomplete-clearIndicator{visibility:visible;}}.css-18nc3u2 .MuiAutocomplete-tag{margin:3px;max-width:calc(100% - 6px);}.css-18nc3u2 .MuiAutocomplete-inputRoot{-webkit-box-flex-wrap:wrap;-webkit-flex-wrap:wrap;-ms-flex-wrap:wrap;flex-wrap:wrap;}.MuiAutocomplete-hasPopupIcon.css-18nc3u2 .MuiAutocomplete-inputRoot,.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiAutocomplete-inputRoot{padding-right:30px;}.MuiAutocomplete-hasPopupIcon.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiAutocomplete-inputRoot{padding-right:56px;}.css-18nc3u2 .MuiAutocomplete-inputRoot .MuiAutocomplete-input{width:0;min-width:30px;}.css-18nc3u2 .MuiInput-root{padding-bottom:1px;}.css-18nc3u2 .MuiInput-root .MuiInput-input{padding:4px 4px 4px 0px;}.css-18nc3u2 .MuiInput-root.MuiInputBase-sizeSmall .MuiInput-input{padding:2px 4px 3px 0;}.css-18nc3u2 .MuiOutlinedInput-root{padding:9px;}.MuiAutocomplete-hasPopupIcon.css-18nc3u2 .MuiOutlinedInput-root,.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiOutlinedInput-root{padding-right:39px;}.MuiAutocomplete-hasPopupIcon.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiOutlinedInput-root{padding-right:65px;}.css-18nc3u2 .MuiOutlinedInput-root .MuiAutocomplete-input{padding:7.5px 4px 7.5px 5px;}.css-18nc3u2 .MuiOutlinedInput-root .MuiAutocomplete-endAdornment{right:9px;}.css-18nc3u2 .MuiOutlinedInput-root.MuiInputBase-sizeSmall{padding-top:6px;padding-bottom:6px;padding-left:6px;}.css-18nc3u2 .MuiOutlinedInput-root.MuiInputBase-sizeSmall .MuiAutocomplete-input{padding:2.5px 4px 2.5px 8px;}.css-18nc3u2 .MuiFilledInput-root{padding-top:19px;padding-left:8px;}.MuiAutocomplete-hasPopupIcon.css-18nc3u2 .MuiFilledInput-root,.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiFilledInput-root{padding-right:39px;}.MuiAutocomplete-hasPopupIcon.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiFilledInput-root{padding-right:65px;}.css-18nc3u2 .MuiFilledInput-root .MuiFilledInput-input{padding:7px 4px;}.css-18nc3u2 .MuiFilledInput-root .MuiAutocomplete-endAdornment{right:9px;}.css-18nc3u2 .MuiFilledInput-root.MuiInputBase-sizeSmall{padding-bottom:1px;}.css-18nc3u2 .MuiFilledInput-root.MuiInputBase-sizeSmall .MuiFilledInput-input{padding:2.5px 4px;}.css-18nc3u2 .MuiInputBase-hiddenLabel{padding-top:8px;}.css-18nc3u2 .MuiFilledInput-root.MuiInputBase-hiddenLabel{padding-top:0;padding-bottom:0;}.css-18nc3u2 .MuiFilledInput-root.MuiInputBase-hiddenLabel .MuiAutocomplete-input{padding-top:16px;padding-bottom:17px;}.css-18nc3u2 .MuiFilledInput-root.MuiInputBase-hiddenLabel.MuiInputBase-sizeSmall .MuiAutocomplete-input{padding-top:8px;padding-bottom:9px;}.css-18nc3u2 .MuiAutocomplete-input{-webkit-box-flex:1;-webkit-flex-grow:1;-ms-flex-positive:1;flex-grow:1;text-overflow:ellipsis;opacity:1;}</style><div class="MuiAutocomplete-root css-18nc3u2"><style data-emotion="css feqhe6">.css-feqhe6{display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;-webkit-flex-direction:column;-ms-flex-direction:column;flex-direction:column;position:relative;min-width:0;padding:0;margin:0;border:0;vertical-align:top;width:100%;}</style><div class="MuiFormControl-root MuiFormControl-fullWidth MuiTextField-root css-feqhe6"><style data-emotion="css-global 1prfaxn">@-webkit-keyframes mui-auto-fill{from{display:block;}}@keyframes mui-auto-fill{from{display:block;}}@-webkit-keyframes mui-auto-fill-cancel{from{display:block;}}@keyframes mui-auto-fill-cancel{from{display:block;}}</style><style data-emotion="css 18bxsfn">.css-18bxsfn{font-family:inherit;font-weight:400;font-size:1rem;line-height:1.4375em;color:rgba(0, 0, 0, 0.87);box-sizing:border-box;position:relative;cursor:text;display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;width:100%;position:relative;border-radius:4px;}.css-18bxsfn.Mui-disabled{color:rgba(0, 0, 0, 0.38);cursor:default;}.css-18bxsfn:hover .MuiOutlinedInput-notchedOutline{border-color:rgba(0, 0, 0, 0.87);}@media (hover: none){.css-18bxsfn:hover .MuiOutlinedInput-notchedOutline{border-color:rgba(0, 0, 0, 0.23);}}.css-18bxsfn.Mui-focused .MuiOutlinedInput-notchedOutline{border-color:#1976d2;border-width:2px;}.css-18bxsfn.Mui-error .MuiOutlinedInput-notchedOutline{border-color:#d32f2f;}.css-18bxsfn.Mui-disabled .MuiOutlinedInput-notchedOutline{border-color:rgba(0, 0, 0, 0.26);}</style><div class="MuiInputBase-root MuiOutlinedInput-root MuiInputBase-colorPrimary MuiInputBase-fullWidth MuiInputBase-formControl MuiAutocomplete-inputRoot css-18bxsfn"><style data-emotion="css 1x5jdmq">.css-1x5jdmq{font:inherit;letter-spacing:inherit;color:currentColor;padding:4px 0 5px;border:0;box-sizing:content-box;background:none;height:1.4375em;margin:0;-webkit-tap-highlight-color:transparent;display:block;min-width:0;width:100%;-webkit-animation-name:mui-auto-fill-cancel;animation-name:mui-auto-fill-cancel;-webkit-animation-duration:10ms;animation-duration:10ms;padding:16.5px 14px;}.css-1x5jdmq::-webkit-input-placeholder{color:currentColor;opacity:0.42;-webkit-transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1x5jdmq::-moz-placeholder{color:currentColor;opacity:0.42;-webkit-transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1x5jdmq:-ms-input-placeholder{color:currentColor;opacity:0.42;-webkit-transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1x5jdmq::-ms-input-placeholder{color:currentColor;opacity:0.42;-webkit-transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1x5jdmq:focus{outline:0;}.css-1x5jdmq:invalid{box-shadow:none;}.css-1x5jdmq::-webkit-search-decoration{-webkit-appearance:none;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq::-webkit-input-placeholder{opacity:0!important;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq::-moz-placeholder{opacity:0!important;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq:-ms-input-placeholder{opacity:0!important;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq::-ms-input-placeholder{opacity:0!important;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq:focus::-webkit-input-placeholder{opacity:0.42;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq:focus::-moz-placeholder{opacity:0.42;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq:focus:-ms-input-placeholder{opacity:0.42;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq:focus::-ms-input-placeholder{opacity:0.42;}.css-1x5jdmq.Mui-disabled{opacity:1;-webkit-text-fill-color:rgba(0, 0, 0, 0.38);}.css-1x5jdmq:-webkit-autofill{-webkit-animation-duration:5000s;animation-duration:5000s;-webkit-animation-name:mui-auto-fill;animation-name:mui-auto-fill;}.css-1x5jdmq:-webkit-autofill{border-radius:inherit;}</style><input aria-invalid="false" autoComplete="off" id=":R3ahqmmla:" placeholder="Select LLM providers" type="text" class="MuiInputBase-input MuiOutlinedInput-input MuiAutocomplete-input MuiAutocomplete-inputFocused css-1x5jdmq" aria-autocomplete="list" aria-expanded="false" autoCapitalize="none" spellCheck="false" role="combobox" value=""/><style data-emotion="css 19w1uun">.css-19w1uun{border-color:rgba(0, 0, 0, 0.23);}</style><style data-emotion="css igs3ac">.css-igs3ac{text-align:left;position:absolute;bottom:0;right:0;top:-5px;left:0;margin:0;padding:0 8px;pointer-events:none;border-radius:inherit;border-style:solid;border-width:1px;overflow:hidden;min-width:0%;border-color:rgba(0, 0, 0, 0.23);}</style><fieldset aria-hidden="true" class="MuiOutlinedInput-notchedOutline css-igs3ac"><style data-emotion="css ihdtdm">.css-ihdtdm{float:unset;width:auto;overflow:hidden;padding:0;line-height:11px;-webkit-transition:width 150ms cubic-bezier(0.0, 0, 0.2, 1) 0ms;transition:width 150ms cubic-bezier(0.0, 0, 0.2, 1) 0ms;}</style><legend class="css-ihdtdm"><span class="notranslate"></span></legend></fieldset></div></div></div></div></div></div><div class="MuiBox-root css-h5fkc8"></div><div><div class="MuiStack-root css-1w57jtv"><h5 class="MuiTypography-root MuiTypography-h5 css-1d79rw6">Prompts</h5><div><label for="file-input-add-prompt"><span aria-label="Upload prompt from file" class=""><style data-emotion="css 1wf493t">.css-1wf493t{text-align:center;-webkit-flex:0 0 auto;-ms-flex:0 0 auto;flex:0 0 auto;font-size:1.5rem;padding:8px;border-radius:50%;overflow:visible;color:rgba(0, 0, 0, 0.54);-webkit-transition:background-color 150ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 150ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1wf493t:hover{background-color:rgba(0, 0, 0, 0.04);}@media (hover: none){.css-1wf493t:hover{background-color:transparent;}}.css-1wf493t.Mui-disabled{background-color:transparent;color:rgba(0, 0, 0, 0.26);}</style><style data-emotion="css 1yxmbwk">.css-1yxmbwk{display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;-webkit-box-pack:center;-ms-flex-pack:center;-webkit-justify-content:center;justify-content:center;position:relative;box-sizing:border-box;-webkit-tap-highlight-color:transparent;background-color:transparent;outline:0;border:0;margin:0;border-radius:0;padding:0;cursor:pointer;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;vertical-align:middle;-moz-appearance:none;-webkit-appearance:none;-webkit-text-decoration:none;text-decoration:none;color:inherit;text-align:center;-webkit-flex:0 0 auto;-ms-flex:0 0 auto;flex:0 0 auto;font-size:1.5rem;padding:8px;border-radius:50%;overflow:visible;color:rgba(0, 0, 0, 0.54);-webkit-transition:background-color 150ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 150ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1yxmbwk::-moz-focus-inner{border-style:none;}.css-1yxmbwk.Mui-disabled{pointer-events:none;cursor:default;}@media print{.css-1yxmbwk{-webkit-print-color-adjust:exact;color-adjust:exact;}}.css-1yxmbwk:hover{background-color:rgba(0, 0, 0, 0.04);}@media (hover: none){.css-1yxmbwk:hover{background-color:transparent;}}.css-1yxmbwk.Mui-disabled{background-color:transparent;color:rgba(0, 0, 0, 0.26);}</style><span class="MuiButtonBase-root MuiIconButton-root MuiIconButton-sizeMedium css-1yxmbwk" tabindex="0" role="button"><style data-emotion="css vubbuv">.css-vubbuv{-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;width:1em;height:1em;display:inline-block;fill:currentColor;-webkit-flex-shrink:0;-ms-flex-negative:0;flex-shrink:0;-webkit-transition:fill 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:fill 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;font-size:1.5rem;}</style><svg class="MuiSvgIcon-root MuiSvgIcon-fontSizeMedium css-vubbuv" focusable="false" aria-hidden="true" viewBox="0 0 24 24" data-testid="PublishIcon"><path d="M5 4v2h14V4H5zm0 10h4v6h6v-6h4l-7-7-7 7z"></path></svg></span><input id="file-input-add-prompt" type="file" accept=".txt,.md" style="display:none"/></span><style data-emotion="css 1e2dcm1">.css-1e2dcm1{z-index:1500;pointer-events:none;}</style><style data-emotion="css okvapm">.css-okvapm{z-index:1500;pointer-events:none;}</style></label><button class="MuiButtonBase-root MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium css-1r4sjx4" tabindex="0" type="button">Add Prompt</button></div></div><style data-emotion="css kge0eu">.css-kge0eu{width:100%;overflow-x:auto;}</style><div class="MuiTableContainer-root css-kge0eu"><style data-emotion="css 1od17fy">.css-1od17fy{display:table;width:100%;border-collapse:collapse;border-spacing:0;}.css-1od17fy caption{font-family:inherit;font-weight:400;font-size:0.875rem;line-height:1.43;padding:16px;color:rgba(0, 0, 0, 0.6);text-align:left;caption-side:bottom;}</style><table class="MuiTable-root css-1od17fy"><style data-emotion="css 1xnox0e">.css-1xnox0e{display:table-row-group;}</style><tbody class="MuiTableBody-root css-1xnox0e"><style data-emotion="css 1gqug66">.css-1gqug66{color:inherit;display:table-row;vertical-align:middle;outline:0;}.css-1gqug66.MuiTableRow-hover:hover{background-color:rgba(0, 0, 0, 0.04);}.css-1gqug66.Mui-selected{background-color:rgba(25, 118, 210, 0.08);}.css-1gqug66.Mui-selected:hover{background-color:rgba(25, 118, 210, 0.12);}</style><tr class="MuiTableRow-root css-1gqug66"><style data-emotion="css qzozz">.css-qzozz{font-family:inherit;font-weight:400;font-size:0.875rem;line-height:1.43;display:table-cell;vertical-align:inherit;border-bottom:1px solid rgba(224, 224, 224, 1);text-align:center;padding:16px;color:rgba(0, 0, 0, 0.87);}</style><td class="MuiTableCell-root MuiTableCell-body MuiTableCell-alignCenter MuiTableCell-sizeMedium css-qzozz" colSpan="2">No prompts added yet.</td></tr></tbody></table></div><style data-emotion="css 1k371a6">@media print{.css-1k371a6{position:absolute!important;}}</style></div><style data-emotion="css 1hnm6b6">.css-1hnm6b6{margin-top:48px;}</style><div class="MuiBox-root css-1hnm6b6"></div><div class="MuiStack-root css-1w57jtv"><h5 class="MuiTypography-root MuiTypography-h5 css-1d79rw6">Test Cases</h5><button class="MuiButtonBase-root MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium css-1r4sjx4" tabindex="0" type="button">Add Test Case</button></div><div class="MuiTableContainer-root css-kge0eu"><table class="MuiTable-root css-1od17fy"><style data-emotion="css 1wbz3t9">.css-1wbz3t9{display:table-header-group;}</style><thead class="MuiTableHead-root css-1wbz3t9"><tr class="MuiTableRow-root MuiTableRow-head css-1gqug66"><style data-emotion="css 75le9z">.css-75le9z{font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.5rem;display:table-cell;vertical-align:inherit;border-bottom:1px solid rgba(224, 224, 224, 1);text-align:left;padding:16px;color:rgba(0, 0, 0, 0.87);}</style><th class="MuiTableCell-root MuiTableCell-head MuiTableCell-sizeMedium css-75le9z" scope="col">Description</th><th class="MuiTableCell-root MuiTableCell-head MuiTableCell-sizeMedium css-75le9z" scope="col">Assertions</th><th class="MuiTableCell-root MuiTableCell-head MuiTableCell-sizeMedium css-75le9z" scope="col">Variables</th><style data-emotion="css qnurhq">.css-qnurhq{font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.5rem;display:table-cell;vertical-align:inherit;border-bottom:1px solid rgba(224, 224, 224, 1);text-align:right;padding:16px;color:rgba(0, 0, 0, 0.87);-webkit-flex-direction:row-reverse;-ms-flex-direction:row-reverse;flex-direction:row-reverse;}</style><th class="MuiTableCell-root MuiTableCell-head MuiTableCell-alignRight MuiTableCell-sizeMedium css-qnurhq" scope="col"></th></tr></thead><tbody class="MuiTableBody-root css-1xnox0e"><tr class="MuiTableRow-root css-1gqug66"><td class="MuiTableCell-root MuiTableCell-body MuiTableCell-alignCenter MuiTableCell-sizeMedium css-qzozz" colSpan="4">No test cases added yet.</td></tr></tbody></table></div><style data-emotion="css h8m1q6">.css-h8m1q6{margin-top:64px;}</style><div class="MuiBox-root css-h8m1q6"></div></div></div></div><script src="/_next/static/chunks/webpack-a886dd767c2e76b7.js" async=""></script><script src="/_next/static/chunks/fd9d1056-d8847af536b5787b.js" async=""></script><script src="/_next/static/chunks/596-9c29c47b8dee7a50.js" async=""></script><script src="/_next/static/chunks/main-app-581ccf0003955b21.js" async=""></script><script>(self.__next_f=self.__next_f\|\|[]).push([0])</script><script>self.__next_f.push([1,"1:HL[\"/_next/static/media/0e4fe491bf84089c-s.p.woff2\",{\"as\":\"font\",\"type\":\"font/woff2\"}]\n2:HL[\"/_next/static/media/627622453ef56b0d-s.p.woff2\",{\"as\":\"font\",\"type\":\"font/woff2\"}]\n3:HL[\"/_next/static/media/934c4b7cb736f2a3-s.p.woff2\",{\"as\":\"font\",\"type\":\"font/woff2\"}]\n4:HL[\"/_next/static/css/e388dd377baf25ec.css\",{\"as\":\"style\"}]\n0:\"$L5\"\n"])</script><script>self.__next_f.push([1,"6:HL[\"/_next/static/css/a35c840ac696f161.css\",{\"as\":\"style\"}]\n"])</script><script>self.__next_f.push([1,"7:I{\"id\":7948,\"chunks\":[\"272:static/chunks/webpack-a886dd767c2e76b7.js\",\"971:static/chunks/fd9d1056-d8847af536b5787b.js\",\"596:static/chunks/596-9c29c47b8dee7a50.js\"],\"name\":\"default\",\"async\":false}\n9:I{\"id\":6628,\"chunks\":[\"272:static/chunks/webpack-a886dd767c2e76b7.js\",\"971:static/chunks/fd9d1056-d8847af536b5787b.js\",\"596:static/chunks/596-9c29c47b8dee7a50.js\"],\"name\":\"\",\"async\":false}\na:I{\"id\":2152,\"chunks\":[\"658:static/chunks/658-f8f9d18540505edc.js\",\"583:static/chunks/583-507e6d8883bb85ff.js\",\"185:static"])</script><script>self.__next_f.push([1,"/chunks/app/layout-4c714b1a5a3a768d.js\"],\"name\":\"PageShell\",\"async\":false}\nb:I{\"id\":7767,\"chunks\":[\"272:static/chunks/webpack-a886dd767c2e76b7.js\",\"971:static/chunks/fd9d1056-d8847af536b5787b.js\",\"596:static/chunks/596-9c29c47b8dee7a50.js\"],\"name\":\"default\",\"async\":false}\nc:I{\"id\":7920,\"chunks\":[\"272:static/chunks/webpack-a886dd767c2e76b7.js\",\"971:static/chunks/fd9d1056-d8847af536b5787b.js\",\"596:static/chunks/596-9c29c47b8dee7a50.js\"],\"name\":\"default\",\"async\":false}\ne:I{\"id\":4839,\"chunks\":[\"272:static/chunk"])</script><script>self.__next_f.push([1,"s/webpack-a886dd767c2e76b7.js\",\"971:static/chunks/fd9d1056-d8847af536b5787b.js\",\"596:static/chunks/596-9c29c47b8dee7a50.js\"],\"name\":\"default\",\"async\":false}\nf:I{\"id\":5302,\"chunks\":[\"658:static/chunks/658-f8f9d18540505edc.js\",\"858:static/chunks/858-7255df6dbc44dff9.js\",\"339:static/chunks/339-501c32916b785ef1.js\",\"583:static/chunks/583-507e6d8883bb85ff.js\",\"373:static/chunks/373-6a411db0b05027d3.js\",\"413:static/chunks/app/setup/page-cd35686fe6c12be8.js\"],\"name\":\"\",\"async\":false}\n"])</script><script>self.__next_f.push([1,"5:[[[\"$\",\"link\",\"0\",{\"rel\":\"stylesheet\",\"href\":\"/_next/static/css/e388dd377baf25ec.css\",\"precedence\":\"next\"}]],[\"$\",\"$L7\",null,{\"buildId\":\"MaiSPL4P4l2b_4yoQGI8T\",\"assetPrefix\":\"\",\"initialCanonicalUrl\":\"/setup/\",\"initialTree\":[\"\",{\"children\":[\"setup\",{\"children\":[\"__PAGE__\",{}]}]},\"$undefined\",\"$undefined\",true],\"initialHead\":\"$L8\",\"globalErrorComponent\":\"$9\",\"children\":[null,[\"$\",\"html\",null,{\"lang\":\"en\",\"children\":[\"$\",\"body\",null,{\"className\":\"__className_979022\",\"children\":[\"$\",\"$La\",null,{\"children\":[\"$\",\"$Lb\",null,{\"parallelRouterKey\":\"children\",\"segmentPath\":[\"children\"],\"error\":\"$undefined\",\"errorStyles\":\"$undefined\",\"loading\":\"$undefined\",\"loadingStyles\":\"$undefined\",\"hasLoading\":false,\"template\":[\"$\",\"$Lc\",null,{}],\"templateStyles\":\"$undefined\",\"notFound\":[[\"$\",\"title\",null,{\"children\":\"404: This page could not be found.\"}],[\"$\",\"div\",null,{\"style\":{\"fontFamily\":\"system-ui,\\\"Segoe UI\\\",Roboto,Helvetica,Arial,sans-serif,\\\"Apple Color Emoji\\\",\\\"Segoe UI Emoji\\\"\",\"height\":\"100vh\",\"textAlign\":\"center\",\"display\":\"flex\",\"flexDirection\":\"column\",\"alignItems\":\"center\",\"justifyContent\":\"center\"},\"children\":[\"$\",\"div\",null,{\"children\":[[\"$\",\"style\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}\"}}],[\"$\",\"h1\",null,{\"className\":\"next-error-h1\",\"style\":{\"display\":\"inline-block\",\"margin\":\"0 20px 0 0\",\"padding\":\"0 23px 0 0\",\"fontSize\":24,\"fontWeight\":500,\"verticalAlign\":\"top\",\"lineHeight\":\"49px\"},\"children\":\"404\"}],[\"$\",\"div\",null,{\"style\":{\"display\":\"inline-block\"},\"children\":[\"$\",\"h2\",null,{\"style\":{\"fontSize\":14,\"fontWeight\":400,\"lineHeight\":\"49px\",\"margin\":0},\"children\":\"This page could not be found.\"}]}]]}]}]],\"notFoundStyles\":\"$undefined\",\"childProp\":{\"current\":[\"$\",\"$Lb\",null,{\"parallelRouterKey\":\"children\",\"segmentPath\":[\"children\",\"setup\",\"children\"],\"error\":\"$undefined\",\"errorStyles\":\"$undefined\",\"loading\":\"$undefined\",\"loadingStyles\":\"$undefined\",\"hasLoading\":false,\"template\":[\"$\",\"$Lc\",null,{}],\"templateStyles\":\"$undefined\",\"notFound\":\"$undefined\",\"notFoundStyles\":\"$undefined\",\"childProp\":{\"current\":[\"$Ld\",[\"$\",\"$Le\",null,{\"propsForComponent\":{\"params\":{}},\"Component\":\"$f\"}],null],\"segment\":\"__PAGE__\"},\"styles\":[[\"$\",\"link\",\"0\",{\"rel\":\"stylesheet\",\"href\":\"/_next/static/css/a35c840ac696f161.css\",\"precedence\":\"next\"}]]}],\"segment\":\"setup\"},\"styles\":[]}]}]}]}],null]}]]\n"])</script><script>self.__next_f.push([1,"8:[[\"$\",\"meta\",\"0\",{\"charSet\":\"utf-8\"}],[\"$\",\"title\",\"1\",{\"children\":\"promptfoo\"}],[\"$\",\"meta\",\"2\",{\"name\":\"description\",\"content\":\"LLM testing and evaluation\"}],[\"$\",\"meta\",\"3\",{\"name\":\"viewport\",\"content\":\"width=device-width, initial-scale=1\"}],[\"$\",\"meta\",\"4\",{\"name\":\"next-size-adjust\"}]]\nd:null\n"])</script></body></html>
1	+ <!DOCTYPE html><html lang="en"><head><meta charSet="utf-8"/><link rel="preload" as="font" href="/_next/static/media/0e4fe491bf84089c-s.p.woff2" crossorigin="" type="font/woff2"/><link rel="preload" as="font" href="/_next/static/media/627622453ef56b0d-s.p.woff2" crossorigin="" type="font/woff2"/><link rel="preload" as="font" href="/_next/static/media/934c4b7cb736f2a3-s.p.woff2" crossorigin="" type="font/woff2"/><link rel="stylesheet" href="/_next/static/css/e388dd377baf25ec.css" data-precedence="next"/><link rel="stylesheet" href="/_next/static/css/a35c840ac696f161.css" data-precedence="next"/><link rel="preload" href="/_next/static/chunks/webpack-6e474e42be502dd7.js" as="script"/><link rel="preload" href="/_next/static/chunks/fd9d1056-ac777be631f5a9e9.js" as="script"/><link rel="preload" href="/_next/static/chunks/596-07e4a23a5c6cdf04.js" as="script"/><link rel="preload" href="/_next/static/chunks/main-app-c9dc13756d166550.js" as="script"/><title>promptfoo</title><meta name="description" content="LLM testing and evaluation"/><meta name="viewport" content="width=device-width, initial-scale=1"/><meta name="next-size-adjust"/><script src="/_next/static/chunks/polyfills-78c92fac7aa8fdd8.js" noModule=""></script></head><body class="__className_979022"><div><style data-emotion="css jj2ztu">.css-jj2ztu{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-flex-direction:row;-ms-flex-direction:row;flex-direction:row;}.css-jj2ztu>:not(style):not(style){margin:0;}.css-jj2ztu>:not(style)~:not(style){margin-left:16px;}</style><div class="MuiStack-root nav css-jj2ztu"><div class="logo MuiBox-root css-0"><img src="/logo.svg" alt="Promptfoo logo"/> <span>promptfoo</span></div><a href="/setup/">New Eval</a><a href="/eval/">View Evals</a><div class="dark-mode-toggle"><style data-emotion="css vubbuv">.css-vubbuv{-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;width:1em;height:1em;display:inline-block;fill:currentColor;-webkit-flex-shrink:0;-ms-flex-negative:0;flex-shrink:0;-webkit-transition:fill 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:fill 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;font-size:1.5rem;}</style><svg class="MuiSvgIcon-root MuiSvgIcon-fontSizeMedium css-vubbuv" focusable="false" aria-hidden="true" viewBox="0 0 24 24" data-testid="LightModeIcon"><path d="M12 7c-2.76 0-5 2.24-5 5s2.24 5 5 5 5-2.24 5-5-2.24-5-5-5zM2 13h2c.55 0 1-.45 1-1s-.45-1-1-1H2c-.55 0-1 .45-1 1s.45 1 1 1zm18 0h2c.55 0 1-.45 1-1s-.45-1-1-1h-2c-.55 0-1 .45-1 1s.45 1 1 1zM11 2v2c0 .55.45 1 1 1s1-.45 1-1V2c0-.55-.45-1-1-1s-1 .45-1 1zm0 18v2c0 .55.45 1 1 1s1-.45 1-1v-2c0-.55-.45-1-1-1s-1 .45-1 1zM5.99 4.58c-.39-.39-1.03-.39-1.41 0-.39.39-.39 1.03 0 1.41l1.06 1.06c.39.39 1.03.39 1.41 0s.39-1.03 0-1.41L5.99 4.58zm12.37 12.37c-.39-.39-1.03-.39-1.41 0-.39.39-.39 1.03 0 1.41l1.06 1.06c.39.39 1.03.39 1.41 0 .39-.39.39-1.03 0-1.41l-1.06-1.06zm1.06-10.96c.39-.39.39-1.03 0-1.41-.39-.39-1.03-.39-1.41 0l-1.06 1.06c-.39.39-.39 1.03 0 1.41s1.03.39 1.41 0l1.06-1.06zM7.05 18.36c.39-.39.39-1.03 0-1.41-.39-.39-1.03-.39-1.41 0l-1.06 1.06c-.39.39-.39 1.03 0 1.41s1.03.39 1.41 0l1.06-1.06z"></path></svg></div></div><div><style data-emotion="css 19t2hy6">.css-19t2hy6{width:100%;margin-left:auto;box-sizing:border-box;margin-right:auto;display:block;padding-left:16px;padding-right:16px;margin-top:2rem;}@media (min-width:600px){.css-19t2hy6{padding-left:24px;padding-right:24px;}}@media (min-width:1200px){.css-19t2hy6{max-width:1200px;}}</style><div class="MuiContainer-root MuiContainer-maxWidthLg css-19t2hy6"><style data-emotion="css 1w57jtv">.css-1w57jtv{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-flex-direction:row;-ms-flex-direction:row;flex-direction:row;-webkit-box-pack:justify;-webkit-justify-content:space-between;justify-content:space-between;}.css-1w57jtv>:not(style):not(style){margin:0;}.css-1w57jtv>:not(style)~:not(style){margin-left:16px;}</style><div class="MuiStack-root css-1w57jtv"><style data-emotion="css jqhduh">.css-jqhduh{margin:0;font-family:inherit;font-weight:400;font-size:2.125rem;line-height:1.235;}</style><h4 class="MuiTypography-root MuiTypography-h4 css-jqhduh">Set up an evaluation</h4><style data-emotion="css jj2ztu">.css-jj2ztu{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-flex-direction:row;-ms-flex-direction:row;flex-direction:row;}.css-jj2ztu>:not(style):not(style){margin:0;}.css-jj2ztu>:not(style)~:not(style){margin-left:16px;}</style><div class="MuiStack-root css-jj2ztu"><style data-emotion="css 11kchfx">.css-11kchfx{font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.75;text-transform:uppercase;min-width:64px;padding:6px 16px;border-radius:4px;-webkit-transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;color:#fff;background-color:#1976d2;box-shadow:0px 3px 1px -2px rgba(0,0,0,0.2),0px 2px 2px 0px rgba(0,0,0,0.14),0px 1px 5px 0px rgba(0,0,0,0.12);}.css-11kchfx:hover{-webkit-text-decoration:none;text-decoration:none;background-color:#1565c0;box-shadow:0px 2px 4px -1px rgba(0,0,0,0.2),0px 4px 5px 0px rgba(0,0,0,0.14),0px 1px 10px 0px rgba(0,0,0,0.12);}@media (hover: none){.css-11kchfx:hover{background-color:#1976d2;}}.css-11kchfx:active{box-shadow:0px 5px 5px -3px rgba(0,0,0,0.2),0px 8px 10px 1px rgba(0,0,0,0.14),0px 3px 14px 2px rgba(0,0,0,0.12);}.css-11kchfx.Mui-focusVisible{box-shadow:0px 3px 5px -1px rgba(0,0,0,0.2),0px 6px 10px 0px rgba(0,0,0,0.14),0px 1px 18px 0px rgba(0,0,0,0.12);}.css-11kchfx.Mui-disabled{color:rgba(0, 0, 0, 0.26);box-shadow:none;background-color:rgba(0, 0, 0, 0.12);}</style><style data-emotion="css 1r4sjx4">.css-1r4sjx4{display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;-webkit-box-pack:center;-ms-flex-pack:center;-webkit-justify-content:center;justify-content:center;position:relative;box-sizing:border-box;-webkit-tap-highlight-color:transparent;background-color:transparent;outline:0;border:0;margin:0;border-radius:0;padding:0;cursor:pointer;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;vertical-align:middle;-moz-appearance:none;-webkit-appearance:none;-webkit-text-decoration:none;text-decoration:none;color:inherit;font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.75;text-transform:uppercase;min-width:64px;padding:6px 16px;border-radius:4px;-webkit-transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;color:#fff;background-color:#1976d2;box-shadow:0px 3px 1px -2px rgba(0,0,0,0.2),0px 2px 2px 0px rgba(0,0,0,0.14),0px 1px 5px 0px rgba(0,0,0,0.12);}.css-1r4sjx4::-moz-focus-inner{border-style:none;}.css-1r4sjx4.Mui-disabled{pointer-events:none;cursor:default;}@media print{.css-1r4sjx4{-webkit-print-color-adjust:exact;color-adjust:exact;}}.css-1r4sjx4:hover{-webkit-text-decoration:none;text-decoration:none;background-color:#1565c0;box-shadow:0px 2px 4px -1px rgba(0,0,0,0.2),0px 4px 5px 0px rgba(0,0,0,0.14),0px 1px 10px 0px rgba(0,0,0,0.12);}@media (hover: none){.css-1r4sjx4:hover{background-color:#1976d2;}}.css-1r4sjx4:active{box-shadow:0px 5px 5px -3px rgba(0,0,0,0.2),0px 8px 10px 1px rgba(0,0,0,0.14),0px 3px 14px 2px rgba(0,0,0,0.12);}.css-1r4sjx4.Mui-focusVisible{box-shadow:0px 3px 5px -1px rgba(0,0,0,0.2),0px 6px 10px 0px rgba(0,0,0,0.14),0px 1px 18px 0px rgba(0,0,0,0.12);}.css-1r4sjx4.Mui-disabled{color:rgba(0, 0, 0, 0.26);box-shadow:none;background-color:rgba(0, 0, 0, 0.12);}</style><button class="MuiButtonBase-root MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium css-1r4sjx4" tabindex="0" type="button">Run Evaluation</button><style data-emotion="css 1l5hj2g">.css-1l5hj2g{font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.75;text-transform:uppercase;min-width:64px;padding:5px 15px;border-radius:4px;-webkit-transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;border:1px solid rgba(25, 118, 210, 0.5);color:#1976d2;}.css-1l5hj2g:hover{-webkit-text-decoration:none;text-decoration:none;background-color:rgba(25, 118, 210, 0.04);border:1px solid #1976d2;}@media (hover: none){.css-1l5hj2g:hover{background-color:transparent;}}.css-1l5hj2g.Mui-disabled{color:rgba(0, 0, 0, 0.26);border:1px solid rgba(0, 0, 0, 0.12);}</style><style data-emotion="css 1avs8v6">.css-1avs8v6{display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;-webkit-box-pack:center;-ms-flex-pack:center;-webkit-justify-content:center;justify-content:center;position:relative;box-sizing:border-box;-webkit-tap-highlight-color:transparent;background-color:transparent;outline:0;border:0;margin:0;border-radius:0;padding:0;cursor:pointer;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;vertical-align:middle;-moz-appearance:none;-webkit-appearance:none;-webkit-text-decoration:none;text-decoration:none;color:inherit;font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.75;text-transform:uppercase;min-width:64px;padding:5px 15px;border-radius:4px;-webkit-transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,box-shadow 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,border-color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms,color 250ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;border:1px solid rgba(25, 118, 210, 0.5);color:#1976d2;}.css-1avs8v6::-moz-focus-inner{border-style:none;}.css-1avs8v6.Mui-disabled{pointer-events:none;cursor:default;}@media print{.css-1avs8v6{-webkit-print-color-adjust:exact;color-adjust:exact;}}.css-1avs8v6:hover{-webkit-text-decoration:none;text-decoration:none;background-color:rgba(25, 118, 210, 0.04);border:1px solid #1976d2;}@media (hover: none){.css-1avs8v6:hover{background-color:transparent;}}.css-1avs8v6.Mui-disabled{color:rgba(0, 0, 0, 0.26);border:1px solid rgba(0, 0, 0, 0.12);}</style><button class="MuiButtonBase-root MuiButton-root MuiButton-outlined MuiButton-outlinedPrimary MuiButton-sizeMedium MuiButton-outlinedSizeMedium MuiButton-root MuiButton-outlined MuiButton-outlinedPrimary MuiButton-sizeMedium MuiButton-outlinedSizeMedium css-1avs8v6" tabindex="0" type="button">Reset</button></div></div><style data-emotion="css h5fkc8">.css-h5fkc8{margin-top:32px;}</style><div class="MuiBox-root css-h5fkc8"></div><style data-emotion="css 1yuhvjn">.css-1yuhvjn{margin-top:16px;}</style><div class="MuiBox-root css-1yuhvjn"><style data-emotion="css 1nguorl">.css-1nguorl{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-flex-direction:column;-ms-flex-direction:column;flex-direction:column;-webkit-box-pack:justify;-webkit-justify-content:space-between;justify-content:space-between;}.css-1nguorl>:not(style):not(style){margin:0;}.css-1nguorl>:not(style)~:not(style){margin-top:16px;}</style><div class="MuiStack-root css-1nguorl"><style data-emotion="css 1d79rw6">.css-1d79rw6{margin:0;font-family:inherit;font-weight:400;font-size:1.5rem;line-height:1.334;}</style><h5 class="MuiTypography-root MuiTypography-h5 css-1d79rw6">Providers</h5><div class="MuiBox-root css-1yuhvjn"><style data-emotion="css 18nc3u2">.css-18nc3u2.Mui-focused .MuiAutocomplete-clearIndicator{visibility:visible;}@media (pointer: fine){.css-18nc3u2:hover .MuiAutocomplete-clearIndicator{visibility:visible;}}.css-18nc3u2 .MuiAutocomplete-tag{margin:3px;max-width:calc(100% - 6px);}.css-18nc3u2 .MuiAutocomplete-inputRoot{-webkit-box-flex-wrap:wrap;-webkit-flex-wrap:wrap;-ms-flex-wrap:wrap;flex-wrap:wrap;}.MuiAutocomplete-hasPopupIcon.css-18nc3u2 .MuiAutocomplete-inputRoot,.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiAutocomplete-inputRoot{padding-right:30px;}.MuiAutocomplete-hasPopupIcon.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiAutocomplete-inputRoot{padding-right:56px;}.css-18nc3u2 .MuiAutocomplete-inputRoot .MuiAutocomplete-input{width:0;min-width:30px;}.css-18nc3u2 .MuiInput-root{padding-bottom:1px;}.css-18nc3u2 .MuiInput-root .MuiInput-input{padding:4px 4px 4px 0px;}.css-18nc3u2 .MuiInput-root.MuiInputBase-sizeSmall .MuiInput-input{padding:2px 4px 3px 0;}.css-18nc3u2 .MuiOutlinedInput-root{padding:9px;}.MuiAutocomplete-hasPopupIcon.css-18nc3u2 .MuiOutlinedInput-root,.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiOutlinedInput-root{padding-right:39px;}.MuiAutocomplete-hasPopupIcon.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiOutlinedInput-root{padding-right:65px;}.css-18nc3u2 .MuiOutlinedInput-root .MuiAutocomplete-input{padding:7.5px 4px 7.5px 5px;}.css-18nc3u2 .MuiOutlinedInput-root .MuiAutocomplete-endAdornment{right:9px;}.css-18nc3u2 .MuiOutlinedInput-root.MuiInputBase-sizeSmall{padding-top:6px;padding-bottom:6px;padding-left:6px;}.css-18nc3u2 .MuiOutlinedInput-root.MuiInputBase-sizeSmall .MuiAutocomplete-input{padding:2.5px 4px 2.5px 8px;}.css-18nc3u2 .MuiFilledInput-root{padding-top:19px;padding-left:8px;}.MuiAutocomplete-hasPopupIcon.css-18nc3u2 .MuiFilledInput-root,.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiFilledInput-root{padding-right:39px;}.MuiAutocomplete-hasPopupIcon.MuiAutocomplete-hasClearIcon.css-18nc3u2 .MuiFilledInput-root{padding-right:65px;}.css-18nc3u2 .MuiFilledInput-root .MuiFilledInput-input{padding:7px 4px;}.css-18nc3u2 .MuiFilledInput-root .MuiAutocomplete-endAdornment{right:9px;}.css-18nc3u2 .MuiFilledInput-root.MuiInputBase-sizeSmall{padding-bottom:1px;}.css-18nc3u2 .MuiFilledInput-root.MuiInputBase-sizeSmall .MuiFilledInput-input{padding:2.5px 4px;}.css-18nc3u2 .MuiInputBase-hiddenLabel{padding-top:8px;}.css-18nc3u2 .MuiFilledInput-root.MuiInputBase-hiddenLabel{padding-top:0;padding-bottom:0;}.css-18nc3u2 .MuiFilledInput-root.MuiInputBase-hiddenLabel .MuiAutocomplete-input{padding-top:16px;padding-bottom:17px;}.css-18nc3u2 .MuiFilledInput-root.MuiInputBase-hiddenLabel.MuiInputBase-sizeSmall .MuiAutocomplete-input{padding-top:8px;padding-bottom:9px;}.css-18nc3u2 .MuiAutocomplete-input{-webkit-box-flex:1;-webkit-flex-grow:1;-ms-flex-positive:1;flex-grow:1;text-overflow:ellipsis;opacity:1;}</style><div class="MuiAutocomplete-root css-18nc3u2"><style data-emotion="css feqhe6">.css-feqhe6{display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;-webkit-flex-direction:column;-ms-flex-direction:column;flex-direction:column;position:relative;min-width:0;padding:0;margin:0;border:0;vertical-align:top;width:100%;}</style><div class="MuiFormControl-root MuiFormControl-fullWidth MuiTextField-root css-feqhe6"><style data-emotion="css-global 1prfaxn">@-webkit-keyframes mui-auto-fill{from{display:block;}}@keyframes mui-auto-fill{from{display:block;}}@-webkit-keyframes mui-auto-fill-cancel{from{display:block;}}@keyframes mui-auto-fill-cancel{from{display:block;}}</style><style data-emotion="css 18bxsfn">.css-18bxsfn{font-family:inherit;font-weight:400;font-size:1rem;line-height:1.4375em;color:rgba(0, 0, 0, 0.87);box-sizing:border-box;position:relative;cursor:text;display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;width:100%;position:relative;border-radius:4px;}.css-18bxsfn.Mui-disabled{color:rgba(0, 0, 0, 0.38);cursor:default;}.css-18bxsfn:hover .MuiOutlinedInput-notchedOutline{border-color:rgba(0, 0, 0, 0.87);}@media (hover: none){.css-18bxsfn:hover .MuiOutlinedInput-notchedOutline{border-color:rgba(0, 0, 0, 0.23);}}.css-18bxsfn.Mui-focused .MuiOutlinedInput-notchedOutline{border-color:#1976d2;border-width:2px;}.css-18bxsfn.Mui-error .MuiOutlinedInput-notchedOutline{border-color:#d32f2f;}.css-18bxsfn.Mui-disabled .MuiOutlinedInput-notchedOutline{border-color:rgba(0, 0, 0, 0.26);}</style><div class="MuiInputBase-root MuiOutlinedInput-root MuiInputBase-colorPrimary MuiInputBase-fullWidth MuiInputBase-formControl MuiAutocomplete-inputRoot css-18bxsfn"><style data-emotion="css 1x5jdmq">.css-1x5jdmq{font:inherit;letter-spacing:inherit;color:currentColor;padding:4px 0 5px;border:0;box-sizing:content-box;background:none;height:1.4375em;margin:0;-webkit-tap-highlight-color:transparent;display:block;min-width:0;width:100%;-webkit-animation-name:mui-auto-fill-cancel;animation-name:mui-auto-fill-cancel;-webkit-animation-duration:10ms;animation-duration:10ms;padding:16.5px 14px;}.css-1x5jdmq::-webkit-input-placeholder{color:currentColor;opacity:0.42;-webkit-transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1x5jdmq::-moz-placeholder{color:currentColor;opacity:0.42;-webkit-transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1x5jdmq:-ms-input-placeholder{color:currentColor;opacity:0.42;-webkit-transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1x5jdmq::-ms-input-placeholder{color:currentColor;opacity:0.42;-webkit-transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:opacity 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1x5jdmq:focus{outline:0;}.css-1x5jdmq:invalid{box-shadow:none;}.css-1x5jdmq::-webkit-search-decoration{-webkit-appearance:none;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq::-webkit-input-placeholder{opacity:0!important;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq::-moz-placeholder{opacity:0!important;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq:-ms-input-placeholder{opacity:0!important;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq::-ms-input-placeholder{opacity:0!important;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq:focus::-webkit-input-placeholder{opacity:0.42;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq:focus::-moz-placeholder{opacity:0.42;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq:focus:-ms-input-placeholder{opacity:0.42;}label[data-shrink=false]+.MuiInputBase-formControl .css-1x5jdmq:focus::-ms-input-placeholder{opacity:0.42;}.css-1x5jdmq.Mui-disabled{opacity:1;-webkit-text-fill-color:rgba(0, 0, 0, 0.38);}.css-1x5jdmq:-webkit-autofill{-webkit-animation-duration:5000s;animation-duration:5000s;-webkit-animation-name:mui-auto-fill;animation-name:mui-auto-fill;}.css-1x5jdmq:-webkit-autofill{border-radius:inherit;}</style><input aria-invalid="false" autoComplete="off" id=":R3ahqmmla:" placeholder="Select LLM providers" type="text" class="MuiInputBase-input MuiOutlinedInput-input MuiAutocomplete-input MuiAutocomplete-inputFocused css-1x5jdmq" aria-autocomplete="list" aria-expanded="false" autoCapitalize="none" spellCheck="false" role="combobox" value=""/><style data-emotion="css 19w1uun">.css-19w1uun{border-color:rgba(0, 0, 0, 0.23);}</style><style data-emotion="css igs3ac">.css-igs3ac{text-align:left;position:absolute;bottom:0;right:0;top:-5px;left:0;margin:0;padding:0 8px;pointer-events:none;border-radius:inherit;border-style:solid;border-width:1px;overflow:hidden;min-width:0%;border-color:rgba(0, 0, 0, 0.23);}</style><fieldset aria-hidden="true" class="MuiOutlinedInput-notchedOutline css-igs3ac"><style data-emotion="css ihdtdm">.css-ihdtdm{float:unset;width:auto;overflow:hidden;padding:0;line-height:11px;-webkit-transition:width 150ms cubic-bezier(0.0, 0, 0.2, 1) 0ms;transition:width 150ms cubic-bezier(0.0, 0, 0.2, 1) 0ms;}</style><legend class="css-ihdtdm"><span class="notranslate"></span></legend></fieldset></div></div></div></div></div></div><div class="MuiBox-root css-h5fkc8"></div><div><div class="MuiStack-root css-1w57jtv"><h5 class="MuiTypography-root MuiTypography-h5 css-1d79rw6">Prompts</h5><div><label for="file-input-add-prompt"><span aria-label="Upload prompt from file" class=""><style data-emotion="css 1wf493t">.css-1wf493t{text-align:center;-webkit-flex:0 0 auto;-ms-flex:0 0 auto;flex:0 0 auto;font-size:1.5rem;padding:8px;border-radius:50%;overflow:visible;color:rgba(0, 0, 0, 0.54);-webkit-transition:background-color 150ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 150ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1wf493t:hover{background-color:rgba(0, 0, 0, 0.04);}@media (hover: none){.css-1wf493t:hover{background-color:transparent;}}.css-1wf493t.Mui-disabled{background-color:transparent;color:rgba(0, 0, 0, 0.26);}</style><style data-emotion="css 1yxmbwk">.css-1yxmbwk{display:-webkit-inline-box;display:-webkit-inline-flex;display:-ms-inline-flexbox;display:inline-flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;-webkit-box-pack:center;-ms-flex-pack:center;-webkit-justify-content:center;justify-content:center;position:relative;box-sizing:border-box;-webkit-tap-highlight-color:transparent;background-color:transparent;outline:0;border:0;margin:0;border-radius:0;padding:0;cursor:pointer;-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;vertical-align:middle;-moz-appearance:none;-webkit-appearance:none;-webkit-text-decoration:none;text-decoration:none;color:inherit;text-align:center;-webkit-flex:0 0 auto;-ms-flex:0 0 auto;flex:0 0 auto;font-size:1.5rem;padding:8px;border-radius:50%;overflow:visible;color:rgba(0, 0, 0, 0.54);-webkit-transition:background-color 150ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:background-color 150ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;}.css-1yxmbwk::-moz-focus-inner{border-style:none;}.css-1yxmbwk.Mui-disabled{pointer-events:none;cursor:default;}@media print{.css-1yxmbwk{-webkit-print-color-adjust:exact;color-adjust:exact;}}.css-1yxmbwk:hover{background-color:rgba(0, 0, 0, 0.04);}@media (hover: none){.css-1yxmbwk:hover{background-color:transparent;}}.css-1yxmbwk.Mui-disabled{background-color:transparent;color:rgba(0, 0, 0, 0.26);}</style><span class="MuiButtonBase-root MuiIconButton-root MuiIconButton-sizeMedium css-1yxmbwk" tabindex="0" role="button"><style data-emotion="css vubbuv">.css-vubbuv{-webkit-user-select:none;-moz-user-select:none;-ms-user-select:none;user-select:none;width:1em;height:1em;display:inline-block;fill:currentColor;-webkit-flex-shrink:0;-ms-flex-negative:0;flex-shrink:0;-webkit-transition:fill 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;transition:fill 200ms cubic-bezier(0.4, 0, 0.2, 1) 0ms;font-size:1.5rem;}</style><svg class="MuiSvgIcon-root MuiSvgIcon-fontSizeMedium css-vubbuv" focusable="false" aria-hidden="true" viewBox="0 0 24 24" data-testid="PublishIcon"><path d="M5 4v2h14V4H5zm0 10h4v6h6v-6h4l-7-7-7 7z"></path></svg></span><input id="file-input-add-prompt" type="file" accept=".txt,.md" style="display:none"/></span><style data-emotion="css 1e2dcm1">.css-1e2dcm1{z-index:1500;pointer-events:none;}</style><style data-emotion="css okvapm">.css-okvapm{z-index:1500;pointer-events:none;}</style></label><button class="MuiButtonBase-root MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium css-1r4sjx4" tabindex="0" type="button">Add Prompt</button></div></div><style data-emotion="css kge0eu">.css-kge0eu{width:100%;overflow-x:auto;}</style><div class="MuiTableContainer-root css-kge0eu"><style data-emotion="css 1od17fy">.css-1od17fy{display:table;width:100%;border-collapse:collapse;border-spacing:0;}.css-1od17fy caption{font-family:inherit;font-weight:400;font-size:0.875rem;line-height:1.43;padding:16px;color:rgba(0, 0, 0, 0.6);text-align:left;caption-side:bottom;}</style><table class="MuiTable-root css-1od17fy"><style data-emotion="css 1xnox0e">.css-1xnox0e{display:table-row-group;}</style><tbody class="MuiTableBody-root css-1xnox0e"><style data-emotion="css 1gqug66">.css-1gqug66{color:inherit;display:table-row;vertical-align:middle;outline:0;}.css-1gqug66.MuiTableRow-hover:hover{background-color:rgba(0, 0, 0, 0.04);}.css-1gqug66.Mui-selected{background-color:rgba(25, 118, 210, 0.08);}.css-1gqug66.Mui-selected:hover{background-color:rgba(25, 118, 210, 0.12);}</style><tr class="MuiTableRow-root css-1gqug66"><style data-emotion="css qzozz">.css-qzozz{font-family:inherit;font-weight:400;font-size:0.875rem;line-height:1.43;display:table-cell;vertical-align:inherit;border-bottom:1px solid rgba(224, 224, 224, 1);text-align:center;padding:16px;color:rgba(0, 0, 0, 0.87);}</style><td class="MuiTableCell-root MuiTableCell-body MuiTableCell-alignCenter MuiTableCell-sizeMedium css-qzozz" colSpan="2">No prompts added yet.</td></tr></tbody></table></div><style data-emotion="css 1k371a6">@media print{.css-1k371a6{position:absolute!important;}}</style></div><style data-emotion="css 1hnm6b6">.css-1hnm6b6{margin-top:48px;}</style><div class="MuiBox-root css-1hnm6b6"></div><div class="MuiStack-root css-1w57jtv"><h5 class="MuiTypography-root MuiTypography-h5 css-1d79rw6">Test Cases</h5><button class="MuiButtonBase-root MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium MuiButton-root MuiButton-contained MuiButton-containedPrimary MuiButton-sizeMedium MuiButton-containedSizeMedium css-1r4sjx4" tabindex="0" type="button">Add Test Case</button></div><div class="MuiTableContainer-root css-kge0eu"><table class="MuiTable-root css-1od17fy"><style data-emotion="css 1wbz3t9">.css-1wbz3t9{display:table-header-group;}</style><thead class="MuiTableHead-root css-1wbz3t9"><tr class="MuiTableRow-root MuiTableRow-head css-1gqug66"><style data-emotion="css 75le9z">.css-75le9z{font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.5rem;display:table-cell;vertical-align:inherit;border-bottom:1px solid rgba(224, 224, 224, 1);text-align:left;padding:16px;color:rgba(0, 0, 0, 0.87);}</style><th class="MuiTableCell-root MuiTableCell-head MuiTableCell-sizeMedium css-75le9z" scope="col">Description</th><th class="MuiTableCell-root MuiTableCell-head MuiTableCell-sizeMedium css-75le9z" scope="col">Assertions</th><th class="MuiTableCell-root MuiTableCell-head MuiTableCell-sizeMedium css-75le9z" scope="col">Variables</th><style data-emotion="css qnurhq">.css-qnurhq{font-family:inherit;font-weight:500;font-size:0.875rem;line-height:1.5rem;display:table-cell;vertical-align:inherit;border-bottom:1px solid rgba(224, 224, 224, 1);text-align:right;padding:16px;color:rgba(0, 0, 0, 0.87);-webkit-flex-direction:row-reverse;-ms-flex-direction:row-reverse;flex-direction:row-reverse;}</style><th class="MuiTableCell-root MuiTableCell-head MuiTableCell-alignRight MuiTableCell-sizeMedium css-qnurhq" scope="col"></th></tr></thead><tbody class="MuiTableBody-root css-1xnox0e"><tr class="MuiTableRow-root css-1gqug66"><td class="MuiTableCell-root MuiTableCell-body MuiTableCell-alignCenter MuiTableCell-sizeMedium css-qzozz" colSpan="4">No test cases added yet.</td></tr></tbody></table></div><style data-emotion="css h8m1q6">.css-h8m1q6{margin-top:64px;}</style><div class="MuiBox-root css-h8m1q6"></div></div></div></div><script src="/_next/static/chunks/webpack-6e474e42be502dd7.js" async=""></script><script src="/_next/static/chunks/fd9d1056-ac777be631f5a9e9.js" async=""></script><script src="/_next/static/chunks/596-07e4a23a5c6cdf04.js" async=""></script><script src="/_next/static/chunks/main-app-c9dc13756d166550.js" async=""></script><script>(self.__next_f=self.__next_f\|\|[]).push([0])</script><script>self.__next_f.push([1,"1:HL[\"/_next/static/media/0e4fe491bf84089c-s.p.woff2\",{\"as\":\"font\",\"type\":\"font/woff2\"}]\n2:HL[\"/_next/static/media/627622453ef56b0d-s.p.woff2\",{\"as\":\"font\",\"type\":\"font/woff2\"}]\n3:HL[\"/_next/static/media/934c4b7cb736f2a3-s.p.woff2\",{\"as\":\"font\",\"type\":\"font/woff2\"}]\n4:HL[\"/_next/static/css/e388dd377baf25ec.css\",{\"as\":\"style\"}]\n0:\"$L5\"\n"])</script><script>self.__next_f.push([1,"6:HL[\"/_next/static/css/a35c840ac696f161.css\",{\"as\":\"style\"}]\n"])</script><script>self.__next_f.push([1,"7:I{\"id\":57948,\"chunks\":[\"272:static/chunks/webpack-6e474e42be502dd7.js\",\"971:static/chunks/fd9d1056-ac777be631f5a9e9.js\",\"596:static/chunks/596-07e4a23a5c6cdf04.js\"],\"name\":\"default\",\"async\":false}\n9:I{\"id\":56628,\"chunks\":[\"272:static/chunks/webpack-6e474e42be502dd7.js\",\"971:static/chunks/fd9d1056-ac777be631f5a9e9.js\",\"596:static/chunks/596-07e4a23a5c6cdf04.js\"],\"name\":\"\",\"async\":false}\na:I{\"id\":82152,\"chunks\":[\"658:static/chunks/658-a62210d07dc4dcb6.js\",\"583:static/chunks/583-125d32af505e9bc4.js\",\"185:sta"])</script><script>self.__next_f.push([1,"tic/chunks/app/layout-664a8d716d2d24b1.js\"],\"name\":\"PageShell\",\"async\":false}\nb:I{\"id\":47767,\"chunks\":[\"272:static/chunks/webpack-6e474e42be502dd7.js\",\"971:static/chunks/fd9d1056-ac777be631f5a9e9.js\",\"596:static/chunks/596-07e4a23a5c6cdf04.js\"],\"name\":\"default\",\"async\":false}\nc:I{\"id\":57920,\"chunks\":[\"272:static/chunks/webpack-6e474e42be502dd7.js\",\"971:static/chunks/fd9d1056-ac777be631f5a9e9.js\",\"596:static/chunks/596-07e4a23a5c6cdf04.js\"],\"name\":\"default\",\"async\":false}\ne:I{\"id\":44839,\"chunks\":[\"272:static"])</script><script>self.__next_f.push([1,"/chunks/webpack-6e474e42be502dd7.js\",\"971:static/chunks/fd9d1056-ac777be631f5a9e9.js\",\"596:static/chunks/596-07e4a23a5c6cdf04.js\"],\"name\":\"default\",\"async\":false}\nf:I{\"id\":95302,\"chunks\":[\"658:static/chunks/658-a62210d07dc4dcb6.js\",\"858:static/chunks/858-ceb6fa22e614492b.js\",\"339:static/chunks/339-4fc8a80fa840e771.js\",\"583:static/chunks/583-125d32af505e9bc4.js\",\"373:static/chunks/373-8a280796c0f2d1af.js\",\"413:static/chunks/app/setup/page-182018a3c6397345.js\"],\"name\":\"\",\"async\":false}\n"])</script><script>self.__next_f.push([1,"5:[[[\"$\",\"link\",\"0\",{\"rel\":\"stylesheet\",\"href\":\"/_next/static/css/e388dd377baf25ec.css\",\"precedence\":\"next\"}]],[\"$\",\"$L7\",null,{\"buildId\":\"US6gOx8LHTX_Hzm9aYNrC\",\"assetPrefix\":\"\",\"initialCanonicalUrl\":\"/setup/\",\"initialTree\":[\"\",{\"children\":[\"setup\",{\"children\":[\"__PAGE__\",{}]}]},\"$undefined\",\"$undefined\",true],\"initialHead\":\"$L8\",\"globalErrorComponent\":\"$9\",\"children\":[null,[\"$\",\"html\",null,{\"lang\":\"en\",\"children\":[\"$\",\"body\",null,{\"className\":\"__className_979022\",\"children\":[\"$\",\"$La\",null,{\"children\":[\"$\",\"$Lb\",null,{\"parallelRouterKey\":\"children\",\"segmentPath\":[\"children\"],\"error\":\"$undefined\",\"errorStyles\":\"$undefined\",\"loading\":\"$undefined\",\"loadingStyles\":\"$undefined\",\"hasLoading\":false,\"template\":[\"$\",\"$Lc\",null,{}],\"templateStyles\":\"$undefined\",\"notFound\":[[\"$\",\"title\",null,{\"children\":\"404: This page could not be found.\"}],[\"$\",\"div\",null,{\"style\":{\"fontFamily\":\"system-ui,\\\"Segoe UI\\\",Roboto,Helvetica,Arial,sans-serif,\\\"Apple Color Emoji\\\",\\\"Segoe UI Emoji\\\"\",\"height\":\"100vh\",\"textAlign\":\"center\",\"display\":\"flex\",\"flexDirection\":\"column\",\"alignItems\":\"center\",\"justifyContent\":\"center\"},\"children\":[\"$\",\"div\",null,{\"children\":[[\"$\",\"style\",null,{\"dangerouslySetInnerHTML\":{\"__html\":\"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}\"}}],[\"$\",\"h1\",null,{\"className\":\"next-error-h1\",\"style\":{\"display\":\"inline-block\",\"margin\":\"0 20px 0 0\",\"padding\":\"0 23px 0 0\",\"fontSize\":24,\"fontWeight\":500,\"verticalAlign\":\"top\",\"lineHeight\":\"49px\"},\"children\":\"404\"}],[\"$\",\"div\",null,{\"style\":{\"display\":\"inline-block\"},\"children\":[\"$\",\"h2\",null,{\"style\":{\"fontSize\":14,\"fontWeight\":400,\"lineHeight\":\"49px\",\"margin\":0},\"children\":\"This page could not be found.\"}]}]]}]}]],\"notFoundStyles\":\"$undefined\",\"childProp\":{\"current\":[\"$\",\"$Lb\",null,{\"parallelRouterKey\":\"children\",\"segmentPath\":[\"children\",\"setup\",\"children\"],\"error\":\"$undefined\",\"errorStyles\":\"$undefined\",\"loading\":\"$undefined\",\"loadingStyles\":\"$undefined\",\"hasLoading\":false,\"template\":[\"$\",\"$Lc\",null,{}],\"templateStyles\":\"$undefined\",\"notFound\":\"$undefined\",\"notFoundStyles\":\"$undefined\",\"childProp\":{\"current\":[\"$Ld\",[\"$\",\"$Le\",null,{\"propsForComponent\":{\"params\":{}},\"Component\":\"$f\"}],null],\"segment\":\"__PAGE__\"},\"styles\":[[\"$\",\"link\",\"0\",{\"rel\":\"stylesheet\",\"href\":\"/_next/static/css/a35c840ac696f161.css\",\"precedence\":\"next\"}]]}],\"segment\":\"setup\"},\"styles\":[]}]}]}]}],null]}]]\n"])</script><script>self.__next_f.push([1,"8:[[\"$\",\"meta\",\"0\",{\"charSet\":\"utf-8\"}],[\"$\",\"title\",\"1\",{\"children\":\"promptfoo\"}],[\"$\",\"meta\",\"2\",{\"name\":\"description\",\"content\":\"LLM testing and evaluation\"}],[\"$\",\"meta\",\"3\",{\"name\":\"viewport\",\"content\":\"width=device-width, initial-scale=1\"}],[\"$\",\"meta\",\"4\",{\"name\":\"next-size-adjust\"}]]\nd:null\n"])</script></body></html>

package/dist/src/web/nextui/setup/index.txt CHANGED Viewed

@@ -2,13 +2,13 @@
 2:HL["/_next/static/media/627622453ef56b0d-s.p.woff2",{"as":"font","type":"font/woff2"}]
 3:HL["/_next/static/media/934c4b7cb736f2a3-s.p.woff2",{"as":"font","type":"font/woff2"}]
 4:HL["/_next/static/css/e388dd377baf25ec.css",{"as":"style"}]
-0:["MaiSPL4P4l2b_4yoQGI8T",[[["",{"children":["setup",{"children":["__PAGE__",{}]}]},"$undefined","$undefined",true],"$L5",[[["$","link","0",{"rel":"stylesheet","href":"/_next/static/css/e388dd377baf25ec.css","precedence":"next"}]],"$L6"]]]]
+0:["US6gOx8LHTX_Hzm9aYNrC",[[["",{"children":["setup",{"children":["__PAGE__",{}]}]},"$undefined","$undefined",true],"$L5",[[["$","link","0",{"rel":"stylesheet","href":"/_next/static/css/e388dd377baf25ec.css","precedence":"next"}]],"$L6"]]]]
 7:HL["/_next/static/css/a35c840ac696f161.css",{"as":"style"}]
-8:I{"id":2152,"chunks":["658:static/chunks/658-f8f9d18540505edc.js","583:static/chunks/583-507e6d8883bb85ff.js","185:static/chunks/app/layout-4c714b1a5a3a768d.js"],"name":"PageShell","async":false}
-9:I{"id":7767,"chunks":["272:static/chunks/webpack-a886dd767c2e76b7.js","971:static/chunks/fd9d1056-d8847af536b5787b.js","596:static/chunks/596-9c29c47b8dee7a50.js"],"name":"default","async":false}
-a:I{"id":7920,"chunks":["272:static/chunks/webpack-a886dd767c2e76b7.js","971:static/chunks/fd9d1056-d8847af536b5787b.js","596:static/chunks/596-9c29c47b8dee7a50.js"],"name":"default","async":false}
-c:I{"id":4839,"chunks":["272:static/chunks/webpack-a886dd767c2e76b7.js","971:static/chunks/fd9d1056-d8847af536b5787b.js","596:static/chunks/596-9c29c47b8dee7a50.js"],"name":"default","async":false}
-d:I{"id":5302,"chunks":["658:static/chunks/658-f8f9d18540505edc.js","858:static/chunks/858-7255df6dbc44dff9.js","339:static/chunks/339-501c32916b785ef1.js","583:static/chunks/583-507e6d8883bb85ff.js","373:static/chunks/373-6a411db0b05027d3.js","413:static/chunks/app/setup/page-cd35686fe6c12be8.js"],"name":"","async":false}
+8:I{"id":82152,"chunks":["658:static/chunks/658-a62210d07dc4dcb6.js","583:static/chunks/583-125d32af505e9bc4.js","185:static/chunks/app/layout-664a8d716d2d24b1.js"],"name":"PageShell","async":false}
+9:I{"id":47767,"chunks":["272:static/chunks/webpack-6e474e42be502dd7.js","971:static/chunks/fd9d1056-ac777be631f5a9e9.js","596:static/chunks/596-07e4a23a5c6cdf04.js"],"name":"default","async":false}
+a:I{"id":57920,"chunks":["272:static/chunks/webpack-6e474e42be502dd7.js","971:static/chunks/fd9d1056-ac777be631f5a9e9.js","596:static/chunks/596-07e4a23a5c6cdf04.js"],"name":"default","async":false}
+c:I{"id":44839,"chunks":["272:static/chunks/webpack-6e474e42be502dd7.js","971:static/chunks/fd9d1056-ac777be631f5a9e9.js","596:static/chunks/596-07e4a23a5c6cdf04.js"],"name":"default","async":false}
+d:I{"id":95302,"chunks":["658:static/chunks/658-a62210d07dc4dcb6.js","858:static/chunks/858-ceb6fa22e614492b.js","339:static/chunks/339-4fc8a80fa840e771.js","583:static/chunks/583-125d32af505e9bc4.js","373:static/chunks/373-8a280796c0f2d1af.js","413:static/chunks/app/setup/page-182018a3c6397345.js"],"name":"","async":false}
 5:[null,["$","html",null,{"lang":"en","children":["$","body",null,{"className":"__className_979022","children":["$","$L8",null,{"children":["$","$L9",null,{"parallelRouterKey":"children","segmentPath":["children"],"error":"$undefined","errorStyles":"$undefined","loading":"$undefined","loadingStyles":"$undefined","hasLoading":false,"template":["$","$La",null,{}],"templateStyles":"$undefined","notFound":[["$","title",null,{"children":"404: This page could not be found."}],["$","div",null,{"style":{"fontFamily":"system-ui,\"Segoe UI\",Roboto,Helvetica,Arial,sans-serif,\"Apple Color Emoji\",\"Segoe UI Emoji\"","height":"100vh","textAlign":"center","display":"flex","flexDirection":"column","alignItems":"center","justifyContent":"center"},"children":["$","div",null,{"children":[["$","style",null,{"dangerouslySetInnerHTML":{"__html":"body{color:#000;background:#fff;margin:0}.next-error-h1{border-right:1px solid rgba(0,0,0,.3)}@media (prefers-color-scheme:dark){body{color:#fff;background:#000}.next-error-h1{border-right:1px solid rgba(255,255,255,.3)}}"}}],["$","h1",null,{"className":"next-error-h1","style":{"display":"inline-block","margin":"0 20px 0 0","padding":"0 23px 0 0","fontSize":24,"fontWeight":500,"verticalAlign":"top","lineHeight":"49px"},"children":"404"}],["$","div",null,{"style":{"display":"inline-block"},"children":["$","h2",null,{"style":{"fontSize":14,"fontWeight":400,"lineHeight":"49px","margin":0},"children":"This page could not be found."}]}]]}]}]],"notFoundStyles":"$undefined","childProp":{"current":["$","$L9",null,{"parallelRouterKey":"children","segmentPath":["children","setup","children"],"error":"$undefined","errorStyles":"$undefined","loading":"$undefined","loadingStyles":"$undefined","hasLoading":false,"template":["$","$La",null,{}],"templateStyles":"$undefined","notFound":"$undefined","notFoundStyles":"$undefined","childProp":{"current":["$Lb",["$","$Lc",null,{"propsForComponent":{"params":{}},"Component":"$d"}],null],"segment":"__PAGE__"},"styles":[["$","link","0",{"rel":"stylesheet","href":"/_next/static/css/a35c840ac696f161.css","precedence":"next"}]]}],"segment":"setup"},"styles":[]}]}]}]}],null]
 6:[["$","meta","0",{"charSet":"utf-8"}],["$","title","1",{"children":"promptfoo"}],["$","meta","2",{"name":"description","content":"LLM testing and evaluation"}],["$","meta","3",{"name":"viewport","content":"width=device-width, initial-scale=1"}],["$","meta","4",{"name":"next-size-adjust"}]]
 b:null

package/package.json CHANGED Viewed

@@ -2,7 +2,7 @@
   "name": "promptfoo",
   "description": "LLM eval & testing toolkit",
   "author": "Ian Webster",
-  "version": "0.19.2",
+  "version": "0.20.0",
   "license": "MIT",
   "type": "commonjs",
   "main": "dist/src/index.js",
@@ -79,6 +79,7 @@
     "csv-stringify": "^6.3.2",
     "debounce": "^1.2.1",
     "express": "^4.18.2",
+    "fastest-levenshtein": "^1.0.16",
     "glob": "^10.2.6",
     "js-yaml": "^4.1.0",
     "node-fetch": "^2.6.7",

package/src/assertions.ts CHANGED Viewed

@@ -1,20 +1,13 @@
 import rouge from 'rouge';
 import invariant from 'tiny-invariant';
 import Ajv from 'ajv';
+import { distance as levenshtein } from 'fastest-levenshtein';
 import telemetry from './telemetry';
-import { DefaultEmbeddingProvider, DefaultGradingProvider } from './providers/openai';
-import { cosineSimilarity, fetchWithRetries, getNunjucksEngine } from './util';
-import { loadApiProvider } from './providers';
-import { DEFAULT_GRADING_PROMPT } from './prompts';
-import type {
-  Assertion,
-  AssertionType,
-  GradingConfig,
-  GradingResult,
-  AtomicTestCase,
-} from './types';
+import { fetchWithRetries, getNunjucksEngine } from './util';
+import { matchesSimilarity, matchesLlmRubric } from './matchers';
+import type { Assertion, AssertionType, GradingResult, AtomicTestCase } from './types';
 const DEFAULT_SEMANTIC_SIMILARITY_THRESHOLD = 0.8;
@@ -309,7 +302,9 @@ export async function runAssertion(
       if (typeof assertion.value === 'function') {
         return assertion.value(output, test, assertion);
       }
-      const customFunction = new Function('output', 'context', `return ${renderedValue}`);
+      invariant(typeof renderedValue === 'string', 'javascript assertion must have a string value');
+      const functionBody = renderedValue.includes('\n') ? renderedValue : `return ${renderedValue}`;
+      const customFunction = new Function('output', 'context', functionBody);
       const result = customFunction(output, context) as any;
       if (typeof result === 'boolean') {
         pass = result !== inverse;
@@ -317,6 +312,8 @@ export async function runAssertion(
       } else if (typeof result === 'number') {
         pass = true;
         score = result;
+      } else if (typeof result === 'object') {
+        return result;
       } else {
         throw new Error('Custom function must return a boolean or number');
       }
@@ -404,6 +401,11 @@ ${assertion.value}`,
       typeof renderedValue === 'string',
       '"contains" assertion type must have a string value',
     );
+    // Assertion provider overrides test provider
+    test.options = test.options || {};
+    test.options.provider = assertion.provider || test.options.provider;
     return {
       assertion,
       ...(await matchesLlmRubric(renderedValue, output, test.options)),
@@ -469,6 +471,25 @@ ${assertion.value}`,
     return handleRougeScore(baseType, assertion, renderedValue, output, inverse);
   }
+  if (baseType === 'levenshtein') {
+    invariant(
+      typeof renderedValue === 'string',
+      '"levenshtein" assertion type must have a string value',
+    );
+    const levDistance = levenshtein(output, renderedValue);
+    pass = levDistance <= (assertion.threshold || 5);
+    return {
+      pass,
+      score: pass ? 1 : 0,
+      reason: pass
+        ? 'Assertion passed'
+        : `Levenshtein distance ${levDistance} is greater than threshold ${
+            assertion.threshold || 5
+          }`,
+      assertion,
+    };
+  }
   throw new Error('Unknown assertion type: ' + assertion.type);
 }
@@ -489,118 +510,6 @@ function containsJSON(str: string): boolean {
   }
 }
-export async function matchesSimilarity(
-  expected: string,
-  output: string,
-  threshold: number,
-  inverse: boolean = false,
-): Promise<Omit<GradingResult, 'assertion'>> {
-  const expectedEmbedding = await DefaultEmbeddingProvider.callEmbeddingApi(expected);
-  const outputEmbedding = await DefaultEmbeddingProvider.callEmbeddingApi(output);
-  const tokensUsed = {
-    total: (expectedEmbedding.tokenUsage?.total || 0) + (outputEmbedding.tokenUsage?.total || 0),
-    prompt: (expectedEmbedding.tokenUsage?.prompt || 0) + (outputEmbedding.tokenUsage?.prompt || 0),
-    completion:
-      (expectedEmbedding.tokenUsage?.completion || 0) +
-      (outputEmbedding.tokenUsage?.completion || 0),
-  };
-  if (expectedEmbedding.error || outputEmbedding.error) {
-    return {
-      pass: false,
-      score: 0,
-      reason:
-        expectedEmbedding.error || outputEmbedding.error || 'Unknown error fetching embeddings',
-      tokensUsed,
-    };
-  }
-  if (!expectedEmbedding.embedding || !outputEmbedding.embedding) {
-    return {
-      pass: false,
-      score: 0,
-      reason: 'Embedding not found',
-      tokensUsed,
-    };
-  }
-  const similarity = cosineSimilarity(expectedEmbedding.embedding, outputEmbedding.embedding);
-  const pass = inverse ? similarity <= threshold : similarity >= threshold;
-  const greaterThanReason = `Similarity ${similarity} is greater than threshold ${threshold}`;
-  const lessThanReason = `Similarity ${similarity} is less than threshold ${threshold}`;
-  if (pass) {
-    return {
-      pass: true,
-      score: inverse ? 1 - similarity : similarity,
-      reason: inverse ? lessThanReason : greaterThanReason,
-      tokensUsed,
-    };
-  }
-  return {
-    pass: false,
-    score: inverse ? 1 - similarity : similarity,
-    reason: inverse ? greaterThanReason : lessThanReason,
-    tokensUsed,
-  };
-}
-export async function matchesLlmRubric(
-  expected: string,
-  output: string,
-  options?: GradingConfig,
-): Promise<Omit<GradingResult, 'assertion'>> {
-  if (!options) {
-    throw new Error(
-      'Cannot grade output without grading config. Specify --grader option or grading config.',
-    );
-  }
-  const prompt = nunjucks.renderString(options.rubricPrompt || DEFAULT_GRADING_PROMPT, {
-    output: output.replace(/\n/g, '\\n').replace(/"/g, '\\"'),
-    rubric: expected.replace(/\n/g, '\\n').replace(/"/g, '\\"'),
-  });
-  let provider = options.provider || DefaultGradingProvider;
-  if (typeof provider === 'string') {
-    provider = await loadApiProvider(provider);
-  }
-  const resp = await provider.callApi(prompt);
-  if (resp.error || !resp.output) {
-    return {
-      pass: false,
-      score: 0,
-      reason: resp.error || 'No output',
-      tokensUsed: {
-        total: resp.tokenUsage?.total || 0,
-        prompt: resp.tokenUsage?.prompt || 0,
-        completion: resp.tokenUsage?.completion || 0,
-      },
-    };
-  }
-  try {
-    const parsed = JSON.parse(resp.output) as Omit<GradingResult, 'score'>;
-    parsed.tokensUsed = {
-      total: resp.tokenUsage?.total || 0,
-      prompt: resp.tokenUsage?.prompt || 0,
-      completion: resp.tokenUsage?.completion || 0,
-    };
-    return { ...parsed, score: parsed.pass ? 1 : 0 };
-  } catch (err) {
-    return {
-      pass: false,
-      score: 0,
-      reason: `Output is not valid JSON: ${resp.output}`,
-      tokensUsed: {
-        total: resp.tokenUsage?.total || 0,
-        prompt: resp.tokenUsage?.prompt || 0,
-        completion: resp.tokenUsage?.completion || 0,
-      },
-    };
-  }
-}
 export function assertionFromString(expected: string): Assertion {
   // Legacy options
   if (expected.startsWith('fn:') || expected.startsWith('eval:')) {
@@ -621,11 +530,11 @@ export function assertionFromString(expected: string): Assertion {
   // New options
   const assertionRegex =
-    /^(not-)?(equals|contains-any|contains-all|contains-json|is-json|regex|icontains|contains|webhook|rouge-n|similar|starts-with)(?::|\((\d+(\.\d+)?)\):)?(.*)$/;
+    /^(not-)?(equals|contains-any|contains-all|contains-json|is-json|regex|icontains|contains|webhook|rouge-n|similar|starts-with|levenshtein)(?:\((\d+(?:\.\d+)?)\))?(?::(.*))?$/;
   const regexMatch = expected.match(assertionRegex);
   if (regexMatch) {
-    const [_, notPrefix, type, __, thresholdStr, value] = regexMatch;
+    const [_, notPrefix, type, thresholdStr, value] = regexMatch;
     const fullType = notPrefix ? `not-${type}` : type;
     const threshold = parseFloat(thresholdStr);
@@ -638,7 +547,12 @@ export function assertionFromString(expected: string): Assertion {
       return {
         type: fullType as AssertionType,
       };
-    } else if (type === 'rouge-n' || type === 'similar' || type === 'starts-with') {
+    } else if (
+      type === 'rouge-n' ||
+      type === 'similar' ||
+      type === 'starts-with' ||
+      type === 'levenshtein'
+    ) {
       return {
         type: fullType as AssertionType,
         value,
@@ -659,6 +573,7 @@ export function assertionFromString(expected: string): Assertion {
   };
 }
+// These exports are used by the node.js package (index.ts)
 export default {
   matchesSimilarity,
   matchesLlmRubric,

package/src/evaluator.ts CHANGED Viewed

@@ -294,6 +294,10 @@ class Evaluator {
                 ...data.vars,
                 ...test.vars,
               },
+              options: {
+                ...testSuite.defaultTest?.options,
+                ...test.options,
+              },
             };
           });
           // Add scenario tests to tests
@@ -356,7 +360,7 @@ class Evaluator {
                   ...prompt,
                   raw: prependToPrompt + prompt.raw + appendToPrompt,
                 },
-                test: { ...testCase, vars },
+                test: { ...testCase, vars, options: testCase.options },
                 includeProviderId: testSuite.providers.length > 1,
                 rowIndex,
                 colIndex,

package/src/main.ts CHANGED Viewed

@@ -101,7 +101,7 @@ async function main() {
     const packageJson = JSON.parse(
       readFileSync(pathJoin(getDirectory(), '../package.json'), 'utf8'),
     );
-    console.log(packageJson.version);
+    logger.info(packageJson.version);
     process.exit(0);
   });
@@ -259,7 +259,11 @@ async function main() {
     .option('--no-progress-bar', 'Do not show progress bar')
     .option('--no-table', 'Do not output table in CLI', defaultConfig?.commandLineOptions?.table)
     .option('--share', 'Create a shareable URL', defaultConfig?.commandLineOptions?.share)
-    .option('--grader', 'Model that will grade outputs', defaultConfig?.commandLineOptions?.grader)
+    .option(
+      '--grader <provider>',
+      'Model that will grade outputs',
+      defaultConfig?.commandLineOptions?.grader,
+    )
     .option('--verbose', 'Show debug logs', defaultConfig?.commandLineOptions?.verbose)
     .option('--view [port]', 'View in browser ui')
     .action(async (cmdObj: CommandLineOptions & Command) => {

package/src/matchers.ts ADDED Viewed

@@ -0,0 +1,120 @@
+import { DefaultEmbeddingProvider, DefaultGradingProvider } from './providers/openai';
+import { cosineSimilarity, getNunjucksEngine } from './util';
+import { loadApiProvider } from './providers';
+import { DEFAULT_GRADING_PROMPT } from './prompts';
+import type { GradingConfig, GradingResult } from './types';
+const nunjucks = getNunjucksEngine();
+export async function matchesSimilarity(
+  expected: string,
+  output: string,
+  threshold: number,
+  inverse: boolean = false,
+): Promise<Omit<GradingResult, 'assertion'>> {
+  const expectedEmbedding = await DefaultEmbeddingProvider.callEmbeddingApi(expected);
+  const outputEmbedding = await DefaultEmbeddingProvider.callEmbeddingApi(output);
+  const tokensUsed = {
+    total: (expectedEmbedding.tokenUsage?.total || 0) + (outputEmbedding.tokenUsage?.total || 0),
+    prompt: (expectedEmbedding.tokenUsage?.prompt || 0) + (outputEmbedding.tokenUsage?.prompt || 0),
+    completion:
+      (expectedEmbedding.tokenUsage?.completion || 0) +
+      (outputEmbedding.tokenUsage?.completion || 0),
+  };
+  if (expectedEmbedding.error || outputEmbedding.error) {
+    return {
+      pass: false,
+      score: 0,
+      reason:
+        expectedEmbedding.error || outputEmbedding.error || 'Unknown error fetching embeddings',
+      tokensUsed,
+    };
+  }
+  if (!expectedEmbedding.embedding || !outputEmbedding.embedding) {
+    return {
+      pass: false,
+      score: 0,
+      reason: 'Embedding not found',
+      tokensUsed,
+    };
+  }
+  const similarity = cosineSimilarity(expectedEmbedding.embedding, outputEmbedding.embedding);
+  const pass = inverse ? similarity <= threshold : similarity >= threshold;
+  const greaterThanReason = `Similarity ${similarity} is greater than threshold ${threshold}`;
+  const lessThanReason = `Similarity ${similarity} is less than threshold ${threshold}`;
+  if (pass) {
+    return {
+      pass: true,
+      score: inverse ? 1 - similarity : similarity,
+      reason: inverse ? lessThanReason : greaterThanReason,
+      tokensUsed,
+    };
+  }
+  return {
+    pass: false,
+    score: inverse ? 1 - similarity : similarity,
+    reason: inverse ? greaterThanReason : lessThanReason,
+    tokensUsed,
+  };
+}
+export async function matchesLlmRubric(
+  expected: string,
+  output: string,
+  grading?: GradingConfig,
+): Promise<Omit<GradingResult, 'assertion'>> {
+  if (!grading) {
+    throw new Error(
+      'Cannot grade output without grading config. Specify --grader option or grading config.',
+    );
+  }
+  const prompt = nunjucks.renderString(grading.rubricPrompt || DEFAULT_GRADING_PROMPT, {
+    output: output.replace(/\n/g, '\\n').replace(/"/g, '\\"'),
+    rubric: expected.replace(/\n/g, '\\n').replace(/"/g, '\\"'),
+  });
+  let provider = grading.provider || DefaultGradingProvider;
+  if (typeof provider === 'string') {
+    provider = await loadApiProvider(provider);
+  }
+  const resp = await provider.callApi(prompt);
+  if (resp.error || !resp.output) {
+    return {
+      pass: false,
+      score: 0,
+      reason: resp.error || 'No output',
+      tokensUsed: {
+        total: resp.tokenUsage?.total || 0,
+        prompt: resp.tokenUsage?.prompt || 0,
+        completion: resp.tokenUsage?.completion || 0,
+      },
+    };
+  }
+  try {
+    const parsed = JSON.parse(resp.output) as Omit<GradingResult, 'score'>;
+    parsed.tokensUsed = {
+      total: resp.tokenUsage?.total || 0,
+      prompt: resp.tokenUsage?.prompt || 0,
+      completion: resp.tokenUsage?.completion || 0,
+    };
+    return { ...parsed, score: parsed.pass ? 1 : 0 };
+  } catch (err) {
+    return {
+      pass: false,
+      score: 0,
+      reason: `Output is not valid JSON: ${resp.output}`,
+      tokensUsed: {
+        total: resp.tokenUsage?.total || 0,
+        prompt: resp.tokenUsage?.prompt || 0,
+        completion: resp.tokenUsage?.completion || 0,
+      },
+    };
+  }
+}