promptfoo 0.22.0 → 0.23.0
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 +1 -1
- package/dist/src/assertions.d.ts.map +1 -1
- package/dist/src/assertions.js +8 -7
- package/dist/src/assertions.js.map +1 -1
- package/dist/src/evaluator.d.ts.map +1 -1
- package/dist/src/evaluator.js +3 -0
- package/dist/src/evaluator.js.map +1 -1
- package/dist/src/index.d.ts.map +1 -1
- package/dist/src/index.js +6 -0
- package/dist/src/index.js.map +1 -1
- package/dist/src/main.js +9 -4
- package/dist/src/main.js.map +1 -1
- package/dist/src/matchers.d.ts +3 -3
- package/dist/src/matchers.d.ts.map +1 -1
- package/dist/src/matchers.js +22 -4
- package/dist/src/matchers.js.map +1 -1
- package/dist/src/prompts.js +2 -2
- package/dist/src/prompts.js.map +1 -1
- package/dist/src/providers/azureopenai.js +1 -1
- package/dist/src/providers/azureopenai.js.map +1 -1
- package/dist/src/providers/localai.js +1 -1
- package/dist/src/providers/localai.js.map +1 -1
- package/dist/src/providers/openai.js +1 -1
- package/dist/src/providers/openai.js.map +1 -1
- package/dist/src/providers/palm.d.ts +40 -0
- package/dist/src/providers/palm.d.ts.map +1 -0
- package/dist/src/providers/palm.js +93 -0
- package/dist/src/providers/palm.js.map +1 -0
- package/dist/src/providers/shared.d.ts +1 -5
- package/dist/src/providers/shared.d.ts.map +1 -1
- package/dist/src/providers/shared.js +4 -4
- package/dist/src/providers/shared.js.map +1 -1
- package/dist/src/providers/vertex.d.ts +51 -0
- package/dist/src/providers/vertex.d.ts.map +1 -0
- package/dist/src/providers/vertex.js +132 -0
- package/dist/src/providers/vertex.js.map +1 -0
- package/dist/src/providers.d.ts.map +1 -1
- package/dist/src/providers.js +10 -0
- package/dist/src/providers.js.map +1 -1
- package/dist/src/types.d.ts +11 -2
- package/dist/src/types.d.ts.map +1 -1
- package/dist/src/util.d.ts +3 -0
- package/dist/src/util.d.ts.map +1 -1
- package/dist/src/util.js +52 -21
- 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/chunks/613-b8dbbee738eefa17.js +1 -1
- package/dist/src/web/nextui/_next/static/chunks/app/eval/[id]/not-found-14c00147c1eba0de.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/eval/[id]/{page-319d2ee38d37574e.js → page-f49f9dde02e0609a.js} +1 -1
- package/dist/src/web/nextui/_next/static/chunks/app/eval/page-e7a68ae4a30dd34a.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/layout-d60ac3764e5cf889.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/app/{page-1ae60660130041b2.js → page-3d77d57947154b0c.js} +1 -1
- package/dist/src/web/nextui/_next/static/chunks/app/setup/page-6ffcc85cc2fddd4f.js +1 -0
- package/dist/src/web/nextui/_next/static/chunks/{main-app-581ccf0003955b21.js → main-app-a18517c2b68322ff.js} +1 -1
- package/dist/src/web/nextui/_next/static/chunks/webpack-dcf86e6c452ed420.js +1 -0
- package/dist/src/web/nextui/_next/static/css/16c1dd82fc87c9d7.css +1 -0
- package/dist/src/web/nextui/_next/static/css/{408b930062eabefb.css → ff2293947f885320.css} +1 -1
- 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 +2 -2
- package/dist/src/web/nextui/setup/index.txt +9 -9
- package/package.json +1 -1
- package/dist/src/web/nextui/_next/static/chunks/app/eval/[id]/not-found-366629541fd598e9.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/eval/page-a6b1ff91723b7beb.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/layout-19c92c911a7607b5.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/app/setup/page-4b0585deaf7eff9e.js +0 -1
- package/dist/src/web/nextui/_next/static/chunks/webpack-c173e301236a84c2.js +0 -1
- package/dist/src/web/nextui/_next/static/css/4d399fceacd06992.css +0 -1
- /package/dist/src/web/nextui/_next/static/{RdnGIHHt12YXTL4egAcOk → DnQDUETrvmnRwmEIY4Deh}/_buildManifest.js +0 -0
- /package/dist/src/web/nextui/_next/static/{RdnGIHHt12YXTL4egAcOk → DnQDUETrvmnRwmEIY4Deh}/_ssgManifest.js +0 -0
|
@@ -1 +0,0 @@
|
|
|
1
|
-
(self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[413],{3515:function(e,t,n){Promise.resolve().then(n.bind(n,2378))},2378:function(e,t,n){"use strict";n.r(t),n.d(t,{default:function(){return ec}});var r=n(7437),a=n(2265),i=n(5551),l=n(8938),s=n(3226),o=n(6507),c=n(3457),d=n(9394),p=n(2834),u=n(6337),h=n(4173),x=n(1797),m=n(4740),j=n(4033),f=n(6882),g=n(1938),v=n(279),Z=()=>{let e=(0,j.useRouter)(),{env:t,description:n,providers:l,prompts:s,testCases:o}=(0,g.o)(),[c,d]=(0,a.useState)(!1),[p,u]=(0,a.useState)(0),h=async()=>{d(!0);try{let r=await fetch("".concat(v.eA,"/api/eval/"),{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({env:t,description:n,providers:l,prompts:s,tests:o})});if(!r.ok)throw Error("HTTP error! status: ".concat(r.status));let a=await r.json(),i=setInterval(async()=>{let t=await fetch("".concat(v.eA,"/api/eval/").concat(a.id,"/"));if(!t.ok)throw clearInterval(i),Error("HTTP error! status: ".concat(t.status));let n=await t.json();if("complete"===n.status)clearInterval(i),d(!1),v.T8?e.push("/eval"):e.push("/eval/remote:".concat(encodeURIComponent(a.id)));else if("failed"===n.status)throw clearInterval(i),d(!1),Error("Job failed");else{let e=0===n.total?0:Math.round(n.progress/n.total*100);u(e)}},1e3)}catch(e){console.error(e),d(!1),alert("An error occurred: ".concat(e.message))}};return(0,r.jsx)(i.Z,{variant:"contained",color:"primary",onClick:h,disabled:c,children:c?(0,r.jsxs)(r.Fragment,{children:[(0,r.jsx)(f.Z,{size:24,sx:{marginRight:2}}),p.toFixed(0),"% complete"]}):"Run Evaluation"})},b=n(2057),y=n(8768),C=n(5873),k=n(1975),A=n(3295),_=()=>{let{env:e,setEnv:t}=(0,g.o)(),[n,l]=(0,a.useState)(!1),[s,o]=(0,a.useState)(e),c=()=>{l(!1)};return(0,r.jsxs)(r.Fragment,{children:[(0,r.jsx)(i.Z,{variant:"outlined",startIcon:(0,r.jsx)(A.Z,{}),onClick:()=>{l(!0)},children:"API keys"}),(0,r.jsxs)(d.Z,{open:n,onClose:c,fullWidth:!0,maxWidth:"md",children:[(0,r.jsx)(x.Z,{children:"Provider settings"}),(0,r.jsxs)(u.Z,{children:[(0,r.jsxs)(b.Z,{defaultExpanded:!0,children:[(0,r.jsx)(C.Z,{children:"OpenAI"}),(0,r.jsxs)(y.Z,{children:[(0,r.jsx)(k.Z,{label:"OpenAI API key",fullWidth:!0,margin:"normal",value:s.OPENAI_API_KEY,onChange:e=>o({...s,OPENAI_API_KEY:e.target.value})}),(0,r.jsx)(k.Z,{label:"OpenAI API host",fullWidth:!0,margin:"normal",value:s.OPENAI_API_HOST,onChange:e=>o({...s,OPENAI_API_HOST:e.target.value})}),(0,r.jsx)(k.Z,{label:"OpenAI organization",fullWidth:!0,margin:"normal",value:s.OPENAI_ORGANIZATION,onChange:e=>o({...s,OPENAI_ORGANIZATION:e.target.value})})]})]}),(0,r.jsxs)(b.Z,{children:[(0,r.jsx)(C.Z,{children:"Azure"}),(0,r.jsx)(y.Z,{children:(0,r.jsx)(k.Z,{label:"Azure API key",fullWidth:!0,margin:"normal",value:s.AZURE_OPENAI_API_KEY,onChange:e=>o({...s,AZURE_OPENAI_API_KEY:e.target.value})})})]}),(0,r.jsxs)(b.Z,{children:[(0,r.jsx)(C.Z,{children:"Anthropic"}),(0,r.jsx)(y.Z,{children:(0,r.jsx)(k.Z,{label:"Anthropic API key",fullWidth:!0,margin:"normal",value:s.ANTHROPIC_API_KEY,onChange:e=>o({...s,ANTHROPIC_API_KEY:e.target.value})})})]}),(0,r.jsxs)(b.Z,{children:[(0,r.jsx)(C.Z,{children:"Replicate"}),(0,r.jsx)(y.Z,{children:(0,r.jsx)(k.Z,{label:"Replicate API key",fullWidth:!0,margin:"normal",value:s.REPLICATE_API_KEY,onChange:e=>o({...s,REPLICATE_API_KEY:e.target.value})})})]})]}),(0,r.jsxs)(p.Z,{children:[(0,r.jsx)(i.Z,{onClick:c,color:"primary",children:"Cancel"}),(0,r.jsx)(i.Z,{onClick:()=>{t(s),c()},color:"primary",variant:"contained",children:"Save"})]})]})]})},I=n(2653),S=n(3701),E=n(9279),P=n(666),w=n(5795),T=n(8489),O=n(8276),N=n(3391),R=n(6446),z=n(1280),W=n(9329),F=e=>{let{open:t,prompt:n,index:l,onAdd:s,onCancel:o}=e,[c,h]=a.useState(n),m=a.useRef(null);a.useEffect(()=>{h(n)},[n]);let j=e=>{s(c),h(""),e?o():m.current&&m.current.focus()};return(0,r.jsxs)(d.Z,{open:t,onClose:o,fullWidth:!0,maxWidth:"md",children:[(0,r.jsx)(x.Z,{children:"Edit Prompt ".concat(l+1)}),(0,r.jsx)(u.Z,{children:(0,r.jsx)(k.Z,{value:c,onChange:e=>h(e.target.value),fullWidth:!0,margin:"normal",multiline:!0,placeholder:"The quick brown {{animal1}} jumps over the lazy {{animal2}}.",helperText:"Tip: use the {{varname}} syntax to add variables to your prompt.",inputRef:m})}),(0,r.jsxs)(p.Z,{children:[(0,r.jsx)(i.Z,{onClick:j.bind(null,!0),color:"primary",variant:"contained",disabled:!c.length,children:"Add"}),(0,r.jsx)(i.Z,{onClick:j.bind(null,!1),color:"primary",variant:"contained",disabled:!c.length,children:"Add Another"}),(0,r.jsx)(i.Z,{onClick:o,color:"secondary",children:"Cancel"})]})]})},L=()=>{let[e,t]=(0,a.useState)(!1),[n,l]=(0,a.useState)(null),{prompts:o,setPrompts:d}=(0,g.o)(),p=(0,a.useRef)(null);(0,a.useEffect)(()=>{null!==n&&n>0&&p.current&&p.current.focus()},[n]);let u=e=>{l(e),t(!0)},h=(e,t)=>{e.stopPropagation();let n=o[t];d([...o,n])},x=(e,t)=>{d(o.map((n,r)=>r===e?t:n))},m=(e,t)=>{e.stopPropagation(),confirm("Are you sure you want to remove this prompt?")&&d(o.filter((e,n)=>n!==t))};return(0,r.jsxs)("div",{children:[(0,r.jsxs)(c.Z,{direction:"row",spacing:2,justifyContent:"space-between",children:[(0,r.jsx)(s.Z,{variant:"h5",children:"Prompts"}),(0,r.jsxs)("div",{children:[(0,r.jsx)("label",{htmlFor:"file-input-add-prompt",children:(0,r.jsx)(O.Z,{title:"Upload prompt from file",children:(0,r.jsxs)("span",{children:[(0,r.jsx)(I.Z,{component:"span",children:(0,r.jsx)(z.Z,{})}),(0,r.jsx)("input",{id:"file-input-add-prompt",type:"file",accept:".txt,.md",onChange:e=>{var t;e.stopPropagation(),e.preventDefault();let n=null===(t=e.target.files)||void 0===t?void 0:t[0];if(n){let e=new FileReader;e.onload=e=>{var t,n;let r=null===(n=e.target)||void 0===n?void 0:null===(t=n.result)||void 0===t?void 0:t.toString();r&&d([...o,r])},e.readAsText(n)}},style:{display:"none"}})]})})}),(0,r.jsx)(i.Z,{color:"primary",onClick:()=>{t(!0)},variant:"contained",children:"Add Prompt"})]})]}),(0,r.jsx)(w.Z,{children:(0,r.jsx)(S.Z,{children:(0,r.jsx)(E.Z,{children:0===o.length?(0,r.jsx)(T.Z,{children:(0,r.jsx)(P.Z,{colSpan:2,align:"center",children:"No prompts added yet."})}):o.map((e,t)=>(0,r.jsxs)(T.Z,{sx:{"&:hover":{backgroundColor:"rgba(0, 0, 0, 0.04)",cursor:"pointer"}},onClick:()=>u(t),children:[(0,r.jsx)(P.Z,{children:(0,r.jsxs)(s.Z,{variant:"body2",children:["Prompt #".concat(t+1,": "),(e.length>250?e.slice(0,250)+" ...":e).split(/({{\w+}})/g).map((e,t)=>/{{\w+}}/g.test(e)?(0,r.jsx)("span",{style:{backgroundColor:"linen",padding:"0.25rem",borderRadius:"4px"},children:e},t):e)]})}),(0,r.jsxs)(P.Z,{align:"right",sx:{minWidth:150},children:[(0,r.jsx)(I.Z,{onClick:()=>u(t),size:"small",children:(0,r.jsx)(N.Z,{})}),(0,r.jsx)(I.Z,{onClick:e=>h(e,t),size:"small",children:(0,r.jsx)(W.Z,{})}),(0,r.jsx)(I.Z,{onClick:e=>m(e,t),size:"small",children:(0,r.jsx)(R.Z,{})})]})]},t))})})}),(0,r.jsx)(F,{open:e,prompt:null!==n?o[n]:"",index:null!==n?n:0,onAdd:e=>{null!==n?x(n,e):d([...o,e]),l(null)},onCancel:()=>{l(null),t(!1)}})]})},V=n(6988),U=e=>{let{onAdd:t,varsList:n,initialValues:i}=e,[l,d]=a.useState(i||{});return(0,a.useEffect)(()=>{let e={};n.forEach(t=>{e[t]=(null==i?void 0:i[t])||""}),d(e)},[n,i]),(0,r.jsxs)(o.Z,{my:2,children:[(0,r.jsx)(s.Z,{variant:"h6",mb:2,children:"Vars"}),n.length>0?(0,r.jsx)(c.Z,{direction:"row",spacing:2,alignItems:"center",children:Object.keys(l).map((e,n)=>(0,r.jsx)(c.Z,{direction:"row",spacing:2,alignItems:"center",children:(0,r.jsx)(k.Z,{placeholder:e,label:e,value:l[e],fullWidth:!0,onChange:n=>{let r=n.target.value,a={...l,[e]:r};d(a),t(a)}})},n))}):(0,r.jsxs)(s.Z,{variant:"subtitle1",gutterBottom:!0,children:["Add variables to your prompt using the ","{{varname}}"," syntax."]})]})},Y=n(8440);let J=["equals","contains","icontains","contains-all","contains-any","starts-with","regex","is-json","contains-json","javascript","python","similar","llm-rubric","webhook","rouge-n","rouge-s","rouge-l","not-equals","not-contains","not-icontains","not-contains-all","not-contains-any","not-starts-with","not-regex","not-is-json","not-contains-json","not-javascript","not-python","not-similar","not-llm-rubric","not-webhook","not-rouge-n","not-rouge-s","not-rouge-l"];var B=e=>{let{onAdd:t,initialValues:n}=e,[l,d]=(0,a.useState)(n||[]),p=e=>{let n=l.filter((t,n)=>n!==e);d(n),t(n)};return(0,r.jsxs)(r.Fragment,{children:[(0,r.jsx)(s.Z,{variant:"h6",children:"Asserts"}),(0,r.jsx)(o.Z,{my:l.length>0?2:0,children:(0,r.jsx)(c.Z,{direction:"column",spacing:2,children:l.map((e,n)=>(0,r.jsxs)(c.Z,{direction:"row",spacing:2,alignItems:"center",children:[(0,r.jsx)(Y.Z,{value:e.type,options:J,sx:{minWidth:200},onChange:(e,r)=>{let a=l.map((e,t)=>t===n?{...e,type:r}:e);d(a),t(a)},renderInput:e=>(0,r.jsx)(k.Z,{...e,label:"Type"})}),(0,r.jsx)(k.Z,{label:"Value",value:e.value,fullWidth:!0,onChange:e=>{let r=e.target.value,a=l.map((e,t)=>t===n?{...e,value:r}:e);d(a),t(a)}}),(0,r.jsx)(I.Z,{onClick:()=>p(n),size:"small",children:(0,r.jsx)(R.Z,{})})]},n))})}),(0,r.jsx)(i.Z,{color:"primary",onClick:()=>{let e=[...l,{type:"equals",value:""}];d(e),t(e)},children:"Add Assert"})]})},H=e=>{let{open:t,onAdd:n,varsList:l,initialValues:s,onCancel:c}=e,[h,m]=(0,a.useState)((null==s?void 0:s.description)||""),[j,f]=(0,a.useState)((null==s?void 0:s.vars)||{}),[g,v]=(0,a.useState)((null==s?void 0:s.assert)||[]),[Z,b]=(0,a.useState)(0);a.useEffect(()=>{s?(m(s.description||""),f(s.vars||{}),v(s.assert||[])):(m(""),f({}),v([]))},[s]);let y=e=>{n({description:h,vars:j,assert:g},e),e&&c(),m(""),f({}),v([]),b(e=>e+1)};return(0,r.jsxs)(d.Z,{open:t,onClose:c,fullWidth:!0,maxWidth:"md",children:[(0,r.jsx)(x.Z,{children:s?"Edit Test Case":"Add Test Case"}),(0,r.jsx)(u.Z,{children:(0,r.jsxs)(o.Z,{children:[(0,r.jsx)(U,{onAdd:e=>f(e),varsList:l,initialValues:null==s?void 0:s.vars}),(0,r.jsx)(B,{onAdd:e=>v(e),initialValues:(null==s?void 0:s.assert)||[]},Z)]})}),(0,r.jsxs)(p.Z,{children:[(0,r.jsx)(i.Z,{onClick:y.bind(void 0,!0),color:"primary",variant:"contained",children:s?"Update Test Case":"Add Test Case"}),!s&&(0,r.jsx)(i.Z,{onClick:y.bind(void 0,!1),color:"primary",variant:"contained",children:"Add Another"}),(0,r.jsx)(i.Z,{onClick:c,color:"secondary",children:"Cancel"})]})]})},K=e=>{let{varsList:t}=e,{testCases:n,setTestCases:l}=(0,g.o)(),[o,d]=a.useState(null),[p,u]=a.useState(!1),h=(e,t)=>{e.stopPropagation(),confirm("Are you sure you want to delete this test case?")&&l(n.filter((e,n)=>n!==t))},x=(e,t)=>{e.stopPropagation();let r=JSON.parse(JSON.stringify(n[t]));l([...n,r])};return(0,r.jsxs)(r.Fragment,{children:[(0,r.jsxs)(c.Z,{direction:"row",spacing:2,justifyContent:"space-between",children:[(0,r.jsx)(s.Z,{variant:"h5",children:"Test Cases"}),(0,r.jsx)(i.Z,{color:"primary",onClick:()=>u(!0),variant:"contained",children:"Add Test Case"})]}),(0,r.jsx)(w.Z,{children:(0,r.jsxs)(S.Z,{children:[(0,r.jsx)(V.Z,{children:(0,r.jsxs)(T.Z,{children:[(0,r.jsx)(P.Z,{children:"Description"}),(0,r.jsx)(P.Z,{children:"Assertions"}),(0,r.jsx)(P.Z,{children:"Variables"}),(0,r.jsx)(P.Z,{align:"right"})]})}),(0,r.jsx)(E.Z,{children:0===n.length?(0,r.jsx)(T.Z,{children:(0,r.jsx)(P.Z,{colSpan:4,align:"center",children:"No test cases added yet."})}):n.map((e,t)=>{var n;return(0,r.jsxs)(T.Z,{sx:{"&:hover":{backgroundColor:"rgba(0, 0, 0, 0.04)",cursor:"pointer"}},onClick:()=>{d(t),u(!0)},children:[(0,r.jsx)(P.Z,{children:(0,r.jsx)(s.Z,{variant:"body2",children:e.description||"Test Case #".concat(t+1)})}),(0,r.jsxs)(P.Z,{children:[(null===(n=e.assert)||void 0===n?void 0:n.length)||0," assertions"]}),(0,r.jsx)(P.Z,{children:Object.entries(e.vars||{}).map(e=>{let[t,n]=e;return t+"="+n}).join(", ")}),(0,r.jsxs)(P.Z,{align:"right",sx:{minWidth:150},children:[(0,r.jsx)(I.Z,{onClick:()=>{d(t),u(!0)},size:"small",children:(0,r.jsx)(N.Z,{})}),(0,r.jsx)(I.Z,{onClick:e=>x(e,t),size:"small",children:(0,r.jsx)(W.Z,{})}),(0,r.jsx)(I.Z,{onClick:e=>h(e,t),size:"small",children:(0,r.jsx)(R.Z,{})})]})]},t)})})]})}),(0,r.jsx)(H,{open:p,onAdd:(e,t)=>{if(null===o)l([...n,e]);else{let t=n.map((t,n)=>n===o?e:t);l(t),d(null)}t&&u(!1)},varsList:t,initialValues:null!==o?n[o]:void 0,onCancel:()=>{d(null),u(!1)}})]})},q=n(7827),M=e=>{let{onChange:t,...n}=e,[i,l]=a.useState(""),[s,o]=a.useState(!1);return(0,r.jsx)(k.Z,{...n,error:s,helperText:s?"Invalid JSON":"",value:i,onChange:e=>{let n=e.target.value;try{let e=JSON.parse(n);l(n),o(!1),t&&t(e)}catch(e){l(n),o(!0)}}})},D=e=>{let{open:t,providerId:n,config:l,onClose:s,onSave:c}=e,[h,m]=a.useState(l);return a.useEffect(()=>{m(l)},[l]),(0,r.jsxs)(d.Z,{open:t,onClose:s,fullWidth:!0,maxWidth:"md",children:[(0,r.jsxs)(x.Z,{children:["Edit ",n.length>50?n.slice(0,50)+"...":n]}),(0,r.jsx)(u.Z,{children:Object.keys(h).map(e=>{let t;let n=h[e];return"number"==typeof n||"boolean"==typeof n||"string"==typeof n?(t="number"==typeof n?t=>m({...h,[e]:parseFloat(t.target.value)}):"boolean"==typeof n?t=>m({...h,[e]:"true"===t.target.value}):t=>{let n=t.target.value.trim();if(n.startsWith("{")||n.startsWith("["))try{m({...h,[e]:JSON.parse(n)})}catch(t){m({...h,[e]:n})}else"null"===n?m({...h,[e]:null}):"undefined"===n?m({...h,[e]:void 0}):m({...h,[e]:n})},(0,r.jsx)(o.Z,{my:2,children:(0,r.jsx)(k.Z,{label:e,value:n,onChange:t,fullWidth:!0,InputLabelProps:{shrink:!0},type:"number"==typeof n?"number":"text"})},e)):(0,r.jsx)(o.Z,{my:2,children:(0,r.jsx)(M,{label:e,defaultValue:JSON.stringify(n),onChange:t=>{m({...h,[e]:t})},fullWidth:!0,multiline:!0,minRows:2,InputLabelProps:{shrink:!0}})},e)})}),(0,r.jsxs)(p.Z,{children:[(0,r.jsx)(i.Z,{onClick:s,children:"Cancel"}),(0,r.jsx)(i.Z,{onClick:()=>{c(h)},children:"Save"})]})]})};let G=[].concat(["replicate:replicate/flan-t5-small:69716ad8c34274043bf4a135b7315c7c569ec931d8f23d6826e249e1c142a264"].map(e=>({id:e,config:{temperature:.5,max_length:1024,repetition_penality:1}}))).concat(["replicate:replicate/codellama-7b-instruct:0103579e86fc75ba0d65912890fa19ef03c84a68554635319accf2e0ba93d3ae","replicate:replicate/codellama-13b-instruct:da5676342de1a5a335b848383af297f592b816b950a43d251a0a9edd0113604b","replicate:replicate/llama-2-70b-chat:2796ee9483c3fd7aa2e171d38f4ca12251a30609463dcfd4cd76703f22e96cdf"].map(e=>({id:e,config:{system_prompt:"",temperature:.75,top_p:.9,top_k:50,max_new_tokens:128,min_new_tokens:-1}}))).concat(["replicate:replicate/codellama-7b:6880b103613a9cd23950c5fd6c140197e519905bd0dd00e448c4858bdd06090a","replicate:replicate/codellama-13b-python:09b87c02dfa403e0c3289166dece62286b3bce49bae39a9c9204713cf94b8b7d","replicate:replicate/codellama-13b:1c914d844307b0588599b8393480a3ba917b660c7e9dfae681542b5325f228db","replicate:replicate/codellama-34b-python:9048743d22a7b19cd0abb018066809ea6af4f2b4717bef9aad3c5ae21ceac00d","replicate:replicate/codellama-34b:0666717e5ead8557dff55ee8f11924b5c0309f5f1ca52f64bb8eec405fdb38a7"].map(e=>({id:e,config:{temperature:.75,top_p:.9,top_k:50,max_new_tokens:128,min_new_tokens:-1}}))).concat(["replicate:a16z-infra/llama-2-7b-chat:7b0bfc9aff140d5b75bacbed23e91fd3c34b01a1e958d32132de6e0a19796e2c","replicate:a16z-infra/llama-2-13b-chat:2a7f981751ec7fdf87b5b91ad4db53683a98082e9ff7bfd12c8cd5ea85980a52"].map(e=>({id:e,config:{temperature:.95,top_p:.95,top_k:250,max_new_tokens:500,min_new_tokens:-1,repetition_penality:1,system_prompt:""}}))).concat(["anthropic:claude-1","anthropic:claude-1-100k","anthropic:claude-instant-1","anthropic:claude-instant-1-100k"].map(e=>({id:e,config:{max_tokens_to_sample:256,temperature:.5}}))).concat(["openai:gpt-3.5-turbo","openai:gpt-3.5-turbo-0301","openai:gpt-3.5-turbo-0613","openai:gpt-3.5-turbo-16k","openai:gpt-3.5-turbo-16k-0613","openai:gpt-4","openai:gpt-4-0314","openai:gpt-4-0613","openai:gpt-4-32k","openai:gpt-4-32k-0314"].map(e=>({id:e,config:{organization:"",apiHost:"api.openai.com",temperature:.5,max_tokens:1024,top_p:1,frequency_penalty:0,presence_penalty:0,function_call:void 0,functions:void 0,stop:void 0}}))).concat(["azureopenai:gpt-3.5-turbo","azureopenai:gpt-3.5-turbo-0301","azureopenai:gpt-3.5-turbo-0613","azureopenai:gpt-3.5-turbo-16k","azureopenai:gpt-3.5-turbo-16k-0613","azureopenai:gpt-4","azureopenai:gpt-4-0314","azureopenai:gpt-4-0613","azureopenai:gpt-4-32k","azureopenai:gpt-4-32k-0314"].map(e=>({id:e,config:{temperature:.5,max_tokens:1024,top_p:1,frequency_penalty:0,presence_penalty:0,function_call:void 0,functions:void 0,stop:void 0}}))).sort((e,t)=>e.id.localeCompare(t.id)),X={anthropic:"Anthropic",azureopenai:"Azure",openai:"OpenAI",replicate:"Replicate"};var Q=e=>{let{providers:t,onChange:n}=e,[i,l]=a.useState(null),s=e=>"string"==typeof e?e:e.id||"Unknown provider",c=(e,t)=>"string"==typeof e?e:e.id||t,d=e=>{"string"==typeof e?alert("Cannot edit custom providers"):e.config?l(e):alert("There is no config for this provider")};return(0,r.jsxs)(o.Z,{mt:2,children:[(0,r.jsx)(Y.Z,{multiple:!0,freeSolo:!0,options:G,value:t,groupBy:e=>(function(e){if(!e)return"Other";let t=e.split(":")[0];return X[t]||t})(e.id),onChange:(e,t)=>{n(t.map(e=>"string"==typeof e?{id:e}:e))},getOptionLabel:e=>{if(!e)return"";let t="";"string"==typeof e&&(t=e),e.id&&"string"==typeof e.id&&(t=e.id);let n=t.split(":");return n.length>1?n[1]:"Unknown provider"},renderTags:(e,t)=>e.map((e,n)=>{let r=s(e),i=c(e,n);return(0,a.createElement)(q.Z,{variant:"outlined",label:r,...t({index:n}),key:i,onClick:()=>d(e)})}),renderInput:e=>(0,r.jsx)(k.Z,{...e,variant:"outlined",placeholder:"Select LLM providers",helperText:t.length>0?"Click a provider to configure its settings.":null})}),i&&i.id&&(0,r.jsx)(D,{open:!!i,providerId:i.id,config:i.config,onClose:()=>l(null),onSave:e=>{if(i){let r=t.map(t=>t.id===i.id?{...t,config:e}:t);n(r),l(null)}}})]})},$=n(3425),ee=n.n($),et=n(4759);n(5341),n(7555);var en=n(8339),er=n(9963),ea=n(1396),ei=n.n(ea);n(695);var el=()=>{let{env:e,setEnv:t,description:n,setDescription:l,providers:c,setProviders:d,prompts:p,setPrompts:u,testCases:h,setTestCases:x}=(0,g.o)(),[m,j]=a.useState(""),[f,v]=a.useState(!0);a.useEffect(()=>{j(en.default.dump({env:e,description:n,providers:c,prompts:p,tests:h}))},[e,n,c,p,h]);let Z=e=>{t(e.env||{}),l(e.description||""),d(e.providers||[]),u(e.prompts||[]),x(e.tests||[])};return(0,r.jsxs)(o.Z,{mt:4,children:[(0,r.jsx)(s.Z,{variant:"h5",gutterBottom:!0,children:"Configuration"}),(0,r.jsxs)(s.Z,{variant:"body1",gutterBottom:!0,children:["This is the YAML config that defines the evaluation and is processed by promptfoo. See"," ",(0,r.jsx)(ei(),{target:"_blank",href:"https://promptfoo.dev/docs/configuration/guide",children:"configuration docs"})," ","to learn more."]}),(0,r.jsx)(i.Z,{variant:"text",color:"primary",startIcon:f?(0,r.jsx)(N.Z,{}):(0,r.jsx)(er.Z,{}),onClick:()=>{if(!f)try{let e=en.default.load(m,{json:!0});Z(e)}catch(e){}v(!f)},children:f?"Edit YAML":"Save"}),(0,r.jsx)(ee(),{autoCapitalize:"off",value:m,onValueChange:e=>{f||j(e)},highlight:e=>(0,et.highlight)(e,et.languages.yaml),padding:10,style:{fontFamily:'"Fira code", "Fira Mono", monospace',fontSize:12},disabled:f,className:f?"":"glowing-border"})]})};n(284);var es=n(2601);function eo(e){let{error:t,resetErrorBoundary:n}=e;return(0,r.jsxs)("div",{role:"alert",children:[(0,r.jsx)("p",{children:"Something went wrong:"}),(0,r.jsx)("pre",{children:t.message}),(0,r.jsx)("button",{onClick:n,children:"Try again"})]})}var ec=()=>{let[e,t]=(0,a.useState)(!1),{description:n,setDescription:j,providers:f,setProviders:v,prompts:b,setPrompts:y,testCases:C,setTestCases:k}=(0,g.o)();if((0,a.useEffect)(()=>{g.o.persist.rehydrate()},[]),es.env.NEXT_PUBLIC_NO_BROWSING)return null;let A=(e=>{let t=/{{(\w+)}}/g,n=new Set;return e.forEach(e=>{let r;for(;null!==(r=t.exec(e));)n.add(r[1])}),Array.from(n)})(b);return(0,r.jsxs)(l.Z,{maxWidth:"lg",sx:{marginTop:"2rem"},children:[(0,r.jsxs)(c.Z,{direction:"row",spacing:2,justifyContent:"space-between",children:[(0,r.jsx)(s.Z,{variant:"h4",children:"Set up an evaluation"}),(0,r.jsxs)(c.Z,{direction:"row",spacing:2,children:[(0,r.jsx)(Z,{}),(0,r.jsx)(_,{}),(0,r.jsx)(i.Z,{variant:"outlined",color:"primary",onClick:()=>t(!0),children:"Reset"})]})]}),(0,r.jsx)(o.Z,{mt:4}),(0,r.jsx)(o.Z,{mt:2,children:(0,r.jsx)(m.SV,{FallbackComponent:eo,onReset:()=>{v([])},children:(0,r.jsxs)(c.Z,{direction:"column",spacing:2,justifyContent:"space-between",children:[(0,r.jsx)(s.Z,{variant:"h5",children:"Providers"}),(0,r.jsx)(Q,{providers:f,onChange:v})]})})}),(0,r.jsx)(o.Z,{mt:4}),(0,r.jsx)(m.SV,{FallbackComponent:eo,onReset:()=>{y([])},children:(0,r.jsx)(L,{})}),(0,r.jsx)(o.Z,{mt:6}),(0,r.jsx)(m.SV,{FallbackComponent:eo,onReset:()=>{k([])},children:(0,r.jsx)(K,{varsList:A})}),(0,r.jsx)(el,{}),(0,r.jsxs)(d.Z,{open:e,onClose:()=>t(!1),"aria-labelledby":"alert-dialog-title","aria-describedby":"alert-dialog-description",children:[(0,r.jsx)(x.Z,{id:"alert-dialog-title",children:"Confirm Reset"}),(0,r.jsx)(u.Z,{children:(0,r.jsx)(h.Z,{id:"alert-dialog-description",children:"Are you sure you want to reset all the fields? This action cannot be undone."})}),(0,r.jsxs)(p.Z,{children:[(0,r.jsx)(i.Z,{onClick:()=>t(!1),children:"Cancel"}),(0,r.jsx)(i.Z,{onClick:()=>{j(""),v([]),y([]),k([]),t(!1)},autoFocus:!0,children:"Reset"})]})]})]})}},279:function(e,t,n){"use strict";n.d(t,{CT:function(){return i},T8:function(){return a},eA:function(){return l}});var r=n(2601);let a=!r.env.NEXT_PUBLIC_PROMPTFOO_BUILD_STANDALONE_SERVER,i="http://localhost:15500",l=""},1938:function(e,t,n){"use strict";n.d(t,{o:function(){return i}});var r=n(4660),a=n(4810);let i=(0,r.Ue)()((0,a.tJ)((e,t)=>({env:{},testCases:[],description:"",providers:[],prompts:[],setEnv:t=>e({env:t}),setTestCases:t=>e({testCases:t}),setDescription:t=>e({description:t}),setProviders:t=>e({providers:t}),setPrompts:t=>e({prompts:t}),setStateFromConfig:t=>{let n={};t.description&&(n.description=t.description||""),t.tests&&(n.testCases=t.tests),t.providers&&(n.providers=t.providers),t.prompts&&("string"==typeof t.prompts?n.prompts=[t.prompts]:Array.isArray(t.prompts)?n.prompts=t.prompts.filter(e=>!e.endsWith(".txt")&&!e.endsWith(".json")&&!e.endsWith(".yaml")):console.warn("Invalid prompts config",t.prompts)),e(n)},getTestSuite:()=>{let{description:e,testCases:n,providers:r,prompts:a,env:i}=t();return{env:i,description:e,providers:r,prompts:a,tests:n}}}),{name:"promptfoo",skipHydration:!0}))},695:function(){},284:function(){}},function(e){e.O(0,[706,249,339,396,753,971,596,744],function(){return e(e.s=3515)}),_N_E=e.O()}]);
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
!function(){"use strict";var e,t,n,r,o,u,i,c,f,a={},d={};function l(e){var t=d[e];if(void 0!==t)return t.exports;var n=d[e]={exports:{}},r=!0;try{a[e].call(n.exports,n,n.exports,l),r=!1}finally{r&&delete d[e]}return n.exports}l.m=a,e=[],l.O=function(t,n,r,o){if(n){o=o||0;for(var u=e.length;u>0&&e[u-1][2]>o;u--)e[u]=e[u-1];e[u]=[n,r,o];return}for(var i=1/0,u=0;u<e.length;u++){for(var n=e[u][0],r=e[u][1],o=e[u][2],c=!0,f=0;f<n.length;f++)i>=o&&Object.keys(l.O).every(function(e){return l.O[e](n[f])})?n.splice(f--,1):(c=!1,o<i&&(i=o));if(c){e.splice(u--,1);var a=r();void 0!==a&&(t=a)}}return t},l.n=function(e){var t=e&&e.__esModule?function(){return e.default}:function(){return e};return l.d(t,{a:t}),t},n=Object.getPrototypeOf?function(e){return Object.getPrototypeOf(e)}:function(e){return e.__proto__},l.t=function(e,r){if(1&r&&(e=this(e)),8&r||"object"==typeof e&&e&&(4&r&&e.__esModule||16&r&&"function"==typeof e.then))return e;var o=Object.create(null);l.r(o);var u={};t=t||[null,n({}),n([]),n(n)];for(var i=2&r&&e;"object"==typeof i&&!~t.indexOf(i);i=n(i))Object.getOwnPropertyNames(i).forEach(function(t){u[t]=function(){return e[t]}});return u.default=function(){return e},l.d(o,u),o},l.d=function(e,t){for(var n in t)l.o(t,n)&&!l.o(e,n)&&Object.defineProperty(e,n,{enumerable:!0,get:t[n]})},l.f={},l.e=function(e){return Promise.all(Object.keys(l.f).reduce(function(t,n){return l.f[n](e,t),t},[]))},l.u=function(e){return"static/chunks/"+e+"-501c32916b785ef1.js"},l.miniCssF=function(e){return"static/css/"+({11:"fc460b8a7cadb952",185:"408b930062eabefb",413:"4d399fceacd06992",421:"7265c36d84346934",466:"48d388184a2f4ce3",931:"8119d8bd13a8adab"})[e]+".css"},l.g=function(){if("object"==typeof globalThis)return globalThis;try{return this||Function("return this")()}catch(e){if("object"==typeof window)return window}}(),l.o=function(e,t){return Object.prototype.hasOwnProperty.call(e,t)},r={},o="_N_E:",l.l=function(e,t,n,u){if(r[e]){r[e].push(t);return}if(void 0!==n)for(var i,c,f=document.getElementsByTagName("script"),a=0;a<f.length;a++){var d=f[a];if(d.getAttribute("src")==e||d.getAttribute("data-webpack")==o+n){i=d;break}}i||(c=!0,(i=document.createElement("script")).charset="utf-8",i.timeout=120,l.nc&&i.setAttribute("nonce",l.nc),i.setAttribute("data-webpack",o+n),i.src=l.tu(e)),r[e]=[t];var s=function(t,n){i.onerror=i.onload=null,clearTimeout(p);var o=r[e];if(delete r[e],i.parentNode&&i.parentNode.removeChild(i),o&&o.forEach(function(e){return e(n)}),t)return t(n)},p=setTimeout(s.bind(null,void 0,{type:"timeout",target:i}),12e4);i.onerror=s.bind(null,i.onerror),i.onload=s.bind(null,i.onload),c&&document.head.appendChild(i)},l.r=function(e){"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})},l.tt=function(){return void 0===u&&(u={createScriptURL:function(e){return e}},"undefined"!=typeof trustedTypes&&trustedTypes.createPolicy&&(u=trustedTypes.createPolicy("nextjs#bundler",u))),u},l.tu=function(e){return l.tt().createScriptURL(e)},l.p="/_next/",i={272:0},l.f.j=function(e,t){var n=l.o(i,e)?i[e]:void 0;if(0!==n){if(n)t.push(n[2]);else if(272!=e){var r=new Promise(function(t,r){n=i[e]=[t,r]});t.push(n[2]=r);var o=l.p+l.u(e),u=Error();l.l(o,function(t){if(l.o(i,e)&&(0!==(n=i[e])&&(i[e]=void 0),n)){var r=t&&("load"===t.type?"missing":t.type),o=t&&t.target&&t.target.src;u.message="Loading chunk "+e+" failed.\n("+r+": "+o+")",u.name="ChunkLoadError",u.type=r,u.request=o,n[1](u)}},"chunk-"+e,e)}else i[e]=0}},l.O.j=function(e){return 0===i[e]},c=function(e,t){var n,r,o=t[0],u=t[1],c=t[2],f=0;if(o.some(function(e){return 0!==i[e]})){for(n in u)l.o(u,n)&&(l.m[n]=u[n]);if(c)var a=c(l)}for(e&&e(t);f<o.length;f++)r=o[f],l.o(i,r)&&i[r]&&i[r][0](),i[r]=0;return l.O(a)},(f=self.webpackChunk_N_E=self.webpackChunk_N_E||[]).forEach(c.bind(null,0)),f.push=c.bind(null,f.push.bind(f))}();
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
code[class*=language-],pre[class*=language-]{color:#000;background:none;text-shadow:0 1px #fff;font-family:Consolas,Monaco,Andale Mono,Ubuntu Mono,monospace;font-size:1em;text-align:left;white-space:pre;word-spacing:normal;word-break:normal;word-wrap:normal;line-height:1.5;-moz-tab-size:4;tab-size:4;-webkit-hyphens:none;hyphens:none}code[class*=language-] ::selection,code[class*=language-]::selection,pre[class*=language-] ::selection,pre[class*=language-]::selection{text-shadow:none;background:#b3d4fc}@media print{code[class*=language-],pre[class*=language-]{text-shadow:none}}pre[class*=language-]{padding:1em;margin:.5em 0;overflow:auto}:not(pre)>code[class*=language-],pre[class*=language-]{background:#f5f2f0}:not(pre)>code[class*=language-]{padding:.1em;border-radius:.3em;white-space:normal}.token.cdata,.token.comment,.token.doctype,.token.prolog{color:#708090}.token.punctuation{color:#999}.token.namespace{opacity:.7}.token.boolean,.token.constant,.token.deleted,.token.number,.token.property,.token.symbol,.token.tag{color:#905}.token.attr-name,.token.builtin,.token.char,.token.inserted,.token.selector,.token.string{color:#690}.language-css .token.string,.style .token.string,.token.entity,.token.operator,.token.url{color:#9a6e3a;background:hsla(0,0%,100%,.5)}.token.atrule,.token.attr-value,.token.keyword{color:#07a}.token.class-name,.token.function{color:#dd4a68}.token.important,.token.regex,.token.variable{color:#e90}.token.bold,.token.important{font-weight:700}.token.italic{font-style:italic}.token.entity{cursor:help}.glowing-border{border:1px solid #5cb3ff;box-shadow:0 0 8px 2px #5cb3ff}.yaml-config{font-size:10px}
|
|
File without changes
|
|
File without changes
|