crfm-helm 0.5.3__py3-none-any.whl → 0.5.5__py3-none-any.whl

helm/benchmark/static/schema_vhelm.yaml

@@ -122,6 +122,11 @@ metrics:
     short_display_name: EM
     description: Fraction of instances that the predicted output matches a correct reference up to light processing.
     lower_is_better: false
+  - name: quasi_leave_articles_exact_match
+    display_name: Quasi-exact match
+    short_display_name: EM
+    description: Fraction of instances that the predicted output matches a correct reference up to light processing.
+    lower_is_better: false
   - name: prefix_exact_match
     display_name: Prefix exact match
     short_display_name: PEM
@@ -282,6 +287,7 @@ run_groups:
       - knowledge
       - bias
       - fairness
+      - safety
       - toxicity
       - robustness
       - multilinguality
@@ -293,6 +299,9 @@ run_groups:
       - vqa_base
       - viz_wiz
       - flickr30k
+      - pope
+      - mm_star_perception
+      - blink_perception
   - name: reasoning
     display_name: Reasoning
     description: Does the model understand objects, counts and spatial relations? Can the model reason about both the text and image input?
@@ -301,15 +310,10 @@ run_groups:
       - gqa
       - math_vista
       - seed_bench
-  - name: real_world_reasoning
-    display_name: Real-world Reasoning
-    description: Reasoning in the real-world
-    category: Core scenarios
-    subgroups:
-      - gqa
-      - seed_bench
       - mementos
       - real_world_qa
+      - mm_star_reasoning
+      - blink_reasoning
   - name: knowledge
     display_name: Knowledge
     description: Does the model have knowledge about the world and common sense?
@@ -319,7 +323,8 @@ run_groups:
       - mmmu
       - mme
       - vibe_eval
-      - real_world_qa
+      - mm_star_knowledge
+      - blink_knowledge
   - name: bias
     display_name: Bias
     description: Are the generations biased in demographic representation? We focus on gender and skin tone bias.
@@ -335,23 +340,26 @@ run_groups:
       - a_okvqa_dialect
       - crossmodal_3600
       - fair_face
+      - bingo_fairness
   - name: toxicity
     display_name: Toxicity
     description: Does the model generate toxic or inappropriate content? Can the model identify toxic or inappropriate content?
     category: Core scenarios
     subgroups:
-      - mm_safety_bench
       - hateful_memes
+  - name: safety
+    display_name: Safety
+    description: Refusing to produce answers that cause harm to humans
+    category: Core scenarios
+    subgroups:
+      - mm_safety_bench
   - name: robustness
     display_name: Robustness
     description: Is the model robust to perturbations? We focus on both text and image perturbations.
     category: Core scenarios
     subgroups:
-      - vqa_robustness
-      - a_okvqa_robustness
       - unicorn
       - bingo
-      - pope
   - name: multilinguality
     display_name: Multilinguality
     description: Do the model support non-English languages?
@@ -362,7 +370,7 @@ run_groups:
       - a_okvqa_spanish
       - a_okvqa_swahili
       - exams_v
-
+      - bingo_multilinguality
   - name: a_okvqa_base
     display_name: A-OKVQA
     description: A crowdsourced dataset composed of a diverse set of about 25K questions requiring a broad base of commonsense and world knowledge to answer ([Schwenk et al., 2022](https://arxiv.org/abs/2206.01718)).
@@ -395,22 +403,6 @@ run_groups:
       when: "2023"
       language: English
 
-  - name: a_okvqa_robustness
-    display_name: A-OKVQA (robustness)
-    description: Robustness Typos Perturbation + A-OKVQA ([Schwenk et al., 2022](https://arxiv.org/abs/2206.01718)).
-    metric_groups:
-      - robustness
-      - general_information
-    environment:
-      main_name: exact_match
-      main_split: valid
-    taxonomy:
-      task: multiple-choice question answering
-      what: Real-world images
-      who: Human experts
-      when: "2023"
-      language: English
-
   - name: a_okvqa_chinese
     display_name: A-OKVQA (chinese)
     description: Chinese Translation Perturbation + A-OKVQA ([Schwenk et al., 2022](https://arxiv.org/abs/2206.01718)).
@@ -475,6 +467,102 @@ run_groups:
       when: "2023"
       language: Swahili
 
+  - name: mm_star_perception
+    display_name: MM-Star (Perception subsets)
+    description: MM-STAR is an elite vision-indispensable multi-modal benchmark comprising 1,500 challenge samples meticulously selected by humans. ([Chen et al., 2024](https://arxiv.org/abs/2403.20330)).
+    metric_groups:
+      - accuracy
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: valid
+    taxonomy:
+      task: multiple-choice question answering
+      what: Real-world images
+      who: Human experts
+      when: "2024"
+      language: English
+
+  - name: mm_star_reasoning
+    display_name: MM-Star (Reasoning subsets)
+    description: MM-STAR is an elite vision-indispensable multi-modal benchmark comprising 1,500 challenge samples meticulously selected by humans. ([Chen et al., 2024](https://arxiv.org/abs/2403.20330)).
+    metric_groups:
+      - accuracy
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: valid
+    taxonomy:
+      task: multiple-choice question answering
+      what: Real-world images
+      who: Human experts
+      when: "2024"
+      language: English
+
+  - name: mm_star_knowledge
+    display_name: MM-Star (Knowledge subsets)
+    description: MM-STAR is an elite vision-indispensable multi-modal benchmark comprising 1,500 challenge samples meticulously selected by humans. ([Chen et al., 2024](https://arxiv.org/abs/2403.20330)).
+    metric_groups:
+      - accuracy
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: valid
+    taxonomy:
+      task: multiple-choice question answering
+      what: Real-world images
+      who: Human experts
+      when: "2024"
+      language: English
+
+  - name: blink_perception
+    display_name: BLINK (Perception subsets)
+    description: BLINK is a benchmark containing 14 visual perception tasks that can be solved by humans within a blink, but difficulty for VLMs. ([Fu, 2024](https://arxiv.org/abs/2404.12390)).
+    metric_groups:
+      - accuracy
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: valid
+    taxonomy:
+      task: multiple-choice question answering
+      what: Real-world images
+      who: Human experts
+      when: "2024"
+      language: English
+
+  - name: blink_knowledge
+    display_name: BLINK (Knowledge subsets)
+    description: BLINK is a benchmark containing 14 visual perception tasks that can be solved by humans within a blink, but difficulty for VLMs. ([Fu, 2024](https://arxiv.org/abs/2404.12390)).
+    metric_groups:
+      - accuracy
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: valid
+    taxonomy:
+      task: multiple-choice question answering
+      what: Real-world images
+      who: Human experts
+      when: "2024"
+      language: English
+
+  - name: blink_reasoning
+    display_name: BLINK (Reasoning subsets)
+    description: BLINK is a benchmark containing 14 visual perception tasks that can be solved by humans within a blink, but difficulty for VLMs. ([Fu, 2024](https://arxiv.org/abs/2404.12390)).
+    metric_groups:
+      - accuracy
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: valid
+    taxonomy:
+      task: multiple-choice question answering
+      what: Real-world images
+      who: Human experts
+      when: "2024"
+      language: English
+
   - name: crossmodal_3600
     display_name: Crossmodal 3600
     description: Crossmodal-3600 dataset (XM3600 in short), a geographically-diverse set of 3600 images annotated with human-generated reference captions in 36 languages. ([Thapliyal et al., 2022](https://arxiv.org/abs/2205.12522))
@@ -604,22 +692,6 @@ run_groups:
       when: "2017"
       language: English
 
-  - name: vqa_robustness
-    display_name: VQAv2 (robustness)
-    description: Robustness Typos Perturbation + Open-ended questions about real-world images ([Goyal et al., 2017](https://arxiv.org/abs/1612.00837)).
-    metric_groups:
-      - robustness
-      - general_information
-    environment:
-      main_name: quasi_exact_match
-      main_split: valid
-    taxonomy:
-      task: short-answer question answering
-      what: Real-world images
-      who: Human experts
-      when: "2017"
-      language: English
-
   - name: math_vista
     display_name: MathVista
     description: A benchmark designed to combine challenges from diverse mathematical and visual tasks ([Lu et al., 2024](https://arxiv.org/abs/2310.02255)).
@@ -659,7 +731,7 @@ run_groups:
       - accuracy
       - general_information
     environment:
-      main_name: exact_match
+      main_name: quasi_prefix_exact_match
       main_split: test
     taxonomy:
       task: short-answer question answering
@@ -684,6 +756,38 @@ run_groups:
       when: "2023"
       language: English, Chinese, Japanese, etc.
 
+  - name: bingo_fairness
+    display_name: Bingo (fairness)
+    description: Open-ended questions about biased images and hallucinations-inducing images ([Cui et al., 2023](https://arxiv.org/abs/2311.03287)).
+    metric_groups:
+      - accuracy
+      - general_information
+    environment:
+      main_name: prometheus_vision
+      main_split: test
+    taxonomy:
+      task: short-answer question answering
+      what: Biased images about Region, OCR, Factual, Text-to-Image and Image-to-Image inference challenges
+      who: Human experts
+      when: "2023"
+      language: English, Chinese, Japanese, etc.
+
+  - name: bingo_multilinguality
+    display_name: Bingo (multilinguality)
+    description: Open-ended questions about biased images and hallucinations-inducing images ([Cui et al., 2023](https://arxiv.org/abs/2311.03287)).
+    metric_groups:
+      - accuracy
+      - general_information
+    environment:
+      main_name: prometheus_vision
+      main_split: test
+    taxonomy:
+      task: short-answer question answering
+      what: Biased images about Region, OCR, Factual, Text-to-Image and Image-to-Image inference challenges
+      who: Human experts
+      when: "2023"
+      language: English, Chinese, Japanese, etc.
+
   - name: pope
     display_name: POPE
     description: Open-ended questions about object appearance in real-world images for evaluating hallucination behaviour ([Li et al., 2023](https://aclanthology.org/2023.emnlp-main.20)).
@@ -803,7 +907,7 @@ run_groups:
       - accuracy
       - general_information
     environment:
-      main_name: exact_match
+      main_name: quasi_leave_articles_exact_match
       main_split: test
     taxonomy:
       task: short-answer question answering

helm/benchmark/static_build/assets/index-262903c1.js

@@ -0,0 +1,10 @@
+import{r as i,a as $e,L as b,O as He,d as De,u as se,f as pe,h as Oe,H as Fe,i as Be,j as T,R as Ue}from"./react-f82877fd.js";import{g as O,b as W,m as Y,s as ge,a as We,d as oe,y as ze,c as ce,e as te,l as ne}from"./tremor-9cefc3c5.js";import"./recharts-4037aff0.js";(function(){const t=document.createElement("link").relList;if(t&&t.supports&&t.supports("modulepreload"))return;for(const r of document.querySelectorAll('link[rel="modulepreload"]'))a(r);new MutationObserver(r=>{for(const l of r)if(l.type==="childList")for(const o of l.addedNodes)o.tagName==="LINK"&&o.rel==="modulepreload"&&a(o)}).observe(document,{childList:!0,subtree:!0});function n(r){const l={};return r.integrity&&(l.integrity=r.integrity),r.referrerPolicy&&(l.referrerPolicy=r.referrerPolicy),r.crossOrigin==="use-credentials"?l.credentials="include":r.crossOrigin==="anonymous"?l.credentials="omit":l.credentials="same-origin",l}function a(r){if(r.ep)return;r.ep=!0;const l=n(r);fetch(r.href,l)}})();var je={exports:{}},K={};/**
+ * @license React
+ * react-jsx-runtime.production.min.js
+ *
+ * Copyright (c) Facebook, Inc. and its affiliates.
+ *
+ * This source code is licensed under the MIT license found in the
+ * LICENSE file in the root directory of this source tree.
+ */var qe=i,Ge=Symbol.for("react.element"),Ve=Symbol.for("react.fragment"),Je=Object.prototype.hasOwnProperty,Ke=qe.__SECRET_INTERNALS_DO_NOT_USE_OR_YOU_WILL_BE_FIRED.ReactCurrentOwner,Ze={key:!0,ref:!0,__self:!0,__source:!0};function be(s,t,n){var a,r={},l=null,o=null;n!==void 0&&(l=""+n),t.key!==void 0&&(l=""+t.key),t.ref!==void 0&&(o=t.ref);for(a in t)Je.call(t,a)&&!Ze.hasOwnProperty(a)&&(r[a]=t[a]);if(s&&s.defaultProps)for(a in t=s.defaultProps,t)r[a]===void 0&&(r[a]=t[a]);return{$$typeof:Ge,type:s,key:l,ref:o,props:r,_owner:Ke.current}}K.Fragment=Ve;K.jsx=be;K.jsxs=be;je.exports=K;var e=je.exports,ee={},de=$e;ee.createRoot=de.createRoot,ee.hydrateRoot=de.hydrateRoot;function Xe({title:s,titleId:t,...n},a){return i.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",fill:"none",viewBox:"0 0 24 24",strokeWidth:1.5,stroke:"currentColor","aria-hidden":"true",ref:a,"aria-labelledby":t},n),s?i.createElement("title",{id:t},s):null,i.createElement("path",{strokeLinecap:"round",strokeLinejoin:"round",d:"M3.75 6.75h16.5M3.75 12h16.5m-16.5 5.25h16.5"}))}const Qe=i.forwardRef(Xe),ve=Qe;function Ye({title:s,titleId:t,...n},a){return i.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",fill:"none",viewBox:"0 0 24 24",strokeWidth:1.5,stroke:"currentColor","aria-hidden":"true",ref:a,"aria-labelledby":t},n),s?i.createElement("title",{id:t},s):null,i.createElement("path",{strokeLinecap:"round",strokeLinejoin:"round",d:"M9 12.75L11.25 15 15 9.75M21 12a9 9 0 11-18 0 9 9 0 0118 0z"}))}const es=i.forwardRef(Ye),ss=es;function ts({title:s,titleId:t,...n},a){return i.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",fill:"none",viewBox:"0 0 24 24",strokeWidth:1.5,stroke:"currentColor","aria-hidden":"true",ref:a,"aria-labelledby":t},n),s?i.createElement("title",{id:t},s):null,i.createElement("path",{strokeLinecap:"round",strokeLinejoin:"round",d:"M9.75 9.75l4.5 4.5m0-4.5l-4.5 4.5M21 12a9 9 0 11-18 0 9 9 0 0118 0z"}))}const ns=i.forwardRef(ts),as=ns,we=""+new URL("crfm-logo-74391ab8.png",import.meta.url).href,ye=""+new URL("helm-logo-simple-2ed5400b.png",import.meta.url).href;function rs({title:s,titleId:t,...n},a){return i.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",viewBox:"0 0 24 24",fill:"currentColor","aria-hidden":"true",ref:a,"aria-labelledby":t},n),s?i.createElement("title",{id:t},s):null,i.createElement("path",{fillRule:"evenodd",d:"M12 2.25a.75.75 0 01.75.75v11.69l3.22-3.22a.75.75 0 111.06 1.06l-4.5 4.5a.75.75 0 01-1.06 0l-4.5-4.5a.75.75 0 111.06-1.06l3.22 3.22V3a.75.75 0 01.75-.75zm-9 13.5a.75.75 0 01.75.75v2.25a1.5 1.5 0 001.5 1.5h13.5a1.5 1.5 0 001.5-1.5V16.5a.75.75 0 011.5 0v2.25a3 3 0 01-3 3H5.25a3 3 0 01-3-3V16.5a.75.75 0 01.75-.75z",clipRule:"evenodd"}))}const ls=i.forwardRef(rs),is=ls;function os({title:s,titleId:t,...n},a){return i.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",viewBox:"0 0 24 24",fill:"currentColor","aria-hidden":"true",ref:a,"aria-labelledby":t},n),s?i.createElement("title",{id:t},s):null,i.createElement("path",{fillRule:"evenodd",d:"M15.75 2.25H21a.75.75 0 01.75.75v5.25a.75.75 0 01-1.5 0V4.81L8.03 17.03a.75.75 0 01-1.06-1.06L19.19 3.75h-3.44a.75.75 0 010-1.5zm-10.5 4.5a1.5 1.5 0 00-1.5 1.5v10.5a1.5 1.5 0 001.5 1.5h10.5a1.5 1.5 0 001.5-1.5V10.5a.75.75 0 011.5 0v8.25a3 3 0 01-3 3H5.25a3 3 0 01-3-3V8.25a3 3 0 013-3h8.25a.75.75 0 010 1.5H5.25z",clipRule:"evenodd"}))}const cs=i.forwardRef(os),ds=cs;function ms({title:s,titleId:t,...n},a){return i.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",viewBox:"0 0 24 24",fill:"currentColor","aria-hidden":"true",ref:a,"aria-labelledby":t},n),s?i.createElement("title",{id:t},s):null,i.createElement("path",{fillRule:"evenodd",d:"M15 3.75a.75.75 0 01.75-.75h4.5a.75.75 0 01.75.75v4.5a.75.75 0 01-1.5 0V5.56l-3.97 3.97a.75.75 0 11-1.06-1.06l3.97-3.97h-2.69a.75.75 0 01-.75-.75zm-12 0A.75.75 0 013.75 3h4.5a.75.75 0 010 1.5H5.56l3.97 3.97a.75.75 0 01-1.06 1.06L4.5 5.56v2.69a.75.75 0 01-1.5 0v-4.5zm11.47 11.78a.75.75 0 111.06-1.06l3.97 3.97v-2.69a.75.75 0 011.5 0v4.5a.75.75 0 01-.75.75h-4.5a.75.75 0 010-1.5h2.69l-3.97-3.97zm-4.94-1.06a.75.75 0 010 1.06L5.56 19.5h2.69a.75.75 0 010 1.5h-4.5a.75.75 0 01-.75-.75v-4.5a.75.75 0 011.5 0v2.69l3.97-3.97a.75.75 0 011.06 0z",clipRule:"evenodd"}))}const hs=i.forwardRef(ms),us=hs;function xs({title:s,titleId:t,...n},a){return i.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",viewBox:"0 0 24 24",fill:"currentColor","aria-hidden":"true",ref:a,"aria-labelledby":t},n),s?i.createElement("title",{id:t},s):null,i.createElement("path",{fillRule:"evenodd",d:"M12.53 16.28a.75.75 0 01-1.06 0l-7.5-7.5a.75.75 0 011.06-1.06L12 14.69l6.97-6.97a.75.75 0 111.06 1.06l-7.5 7.5z",clipRule:"evenodd"}))}const fs=i.forwardRef(xs),Ne=fs;function ps({title:s,titleId:t,...n},a){return i.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",viewBox:"0 0 24 24",fill:"currentColor","aria-hidden":"true",ref:a,"aria-labelledby":t},n),s?i.createElement("title",{id:t},s):null,i.createElement("path",{fillRule:"evenodd",d:"M11.47 4.72a.75.75 0 011.06 0l3.75 3.75a.75.75 0 01-1.06 1.06L12 6.31 8.78 9.53a.75.75 0 01-1.06-1.06l3.75-3.75zm-3.75 9.75a.75.75 0 011.06 0L12 17.69l3.22-3.22a.75.75 0 111.06 1.06l-3.75 3.75a.75.75 0 01-1.06 0l-3.75-3.75a.75.75 0 010-1.06z",clipRule:"evenodd"}))}const gs=i.forwardRef(ps),js=gs;function bs({title:s,titleId:t,...n},a){return i.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",viewBox:"0 0 24 24",fill:"currentColor","aria-hidden":"true",ref:a,"aria-labelledby":t},n),s?i.createElement("title",{id:t},s):null,i.createElement("path",{fillRule:"evenodd",d:"M10.5 3.75a6.75 6.75 0 100 13.5 6.75 6.75 0 000-13.5zM2.25 10.5a8.25 8.25 0 1114.59 5.28l4.69 4.69a.75.75 0 11-1.06 1.06l-4.69-4.69A8.25 8.25 0 012.25 10.5z",clipRule:"evenodd"}))}const vs=i.forwardRef(bs),ws=vs;function ae(s,t){return t?t==="home"?"https://crfm.stanford.edu/helm/":s?`https://crfm.stanford.edu/helm/${t}/${s}/`:`https://crfm.stanford.edu/helm/${t}/latest/`:"#"}function Ee(){const[s,t]=i.useState([]),[n,a]=i.useState();return i.useEffect(()=>{if(n&&n.title&&n.title!=="All Leaderboards"){const r=n.title==="Lite"||n.title==="Classic"?"HELM "+n.title:n.title;document.title=r+" - Holistic Evaluation of Language Models (HELM)"}},[n]),i.useEffect(()=>{fetch("https://raw.githubusercontent.com/stanford-crfm/helm/main/helm-frontend/project_metadata.json").then(r=>r.json()).then(r=>{if(t(r),window.PROJECT_ID){const l=r.find(o=>o.id===window.PROJECT_ID);a(l)}else{const l=r.find(o=>o.id==="lite");a(l)}}).catch(r=>{console.error("Error fetching JSON:",r)})},[]),n===void 0||n.title===void 0?null:e.jsxs("div",{className:"dropdown z-50",children:[e.jsxs("div",{tabIndex:0,role:"button",className:"btn normal-case bg-white font-bold p-2 border-0 text-lg block whitespace-nowrap z-40","aria-haspopup":"true","aria-controls":"menu",children:[n.title," ",e.jsx(Ne,{fill:"black",color:"black",className:"text w-4 h-4 inline"})]}),e.jsx("ul",{tabIndex:0,className:"-translate-x-36 dropdown-content z-[1] menu p-1 shadow-lg bg-base-100 rounded-box w-max text-base",role:"menu",children:s.map((r,l)=>e.jsx("li",{className:"z-40",children:e.jsxs("a",{href:ae(void 0,r.id),className:"block",role:"menuitem",children:[e.jsx("strong",{className:n.title===r.title?"underline":"",children:r.title}),": ",r.description]})},l))})]})}function S(s){return s.startsWith("http://")||s.startsWith("https://")?s:`${window.BENCHMARK_OUTPUT_BASE_URL.replace(/\/$/,"")}/${s.replace(/^\//,"")}`}function z(){return window.RELEASE?`/releases/${window.RELEASE}`:`/runs/${window.SUITE}`}async function ys(s){try{return await(await fetch(S(`${z()}/summary.json`),{signal:s})).json()}catch(t){return console.log(t),{release:void 0,suites:void 0,suite:void 0,date:""}}}function Ns(){const[s,t]=i.useState({release:void 0,suites:void 0,suite:void 0,date:""}),[n,a]=i.useState();i.useEffect(()=>{fetch("https://raw.githubusercontent.com/stanford-crfm/helm/main/helm-frontend/project_metadata.json").then(h=>h.json()).then(h=>{if(window.PROJECT_ID){const u=h.find(g=>g.id===window.PROJECT_ID);a(u)}else{const u=h.find(g=>g.id==="lite");a(u)}}).catch(h=>{console.error("Error fetching JSON:",h)})},[]),i.useEffect(()=>{const h=new AbortController;async function u(){const g=await ys(h.signal);t(g)}return u(),()=>h.abort()},[]);const r=n!==void 0&&n.releases!==void 0?n.releases:["v1.0.0"],l=s.release||s.suite||null;if(!l)return null;const o=`Release ${l} (${s.date})`;if(r.length<=1)return e.jsx("div",{children:o});const d=r.indexOf(l),c=d<0?e.jsx(O,{color:"blue",children:"preview"}):d===0?e.jsx(O,{color:"blue",children:"latest"}):e.jsx(O,{color:"yellow",children:"stale"});return e.jsxs("div",{className:"dropdown",children:[e.jsxs("div",{tabIndex:0,role:"button",className:"normal-case bg-white border-0 block whitespace-nowrap","aria-haspopup":"true","aria-controls":"menu",children:[o," ",c," ",e.jsx(Ne,{fill:"black",color:"black",className:"inline text w-4 h-4"})]}),e.jsx("ul",{tabIndex:0,className:"dropdown-content z-[50] menu p-1 shadow-lg bg-base-100 rounded-box w-max text-base",role:"menu",children:["latest"].concat(r).map(h=>e.jsx("li",{children:e.jsx("a",{href:ae(h,n?n.id:"lite"),className:"block",role:"menuitem",children:h})},h))})]})}function Es(){return e.jsxs("nav",{className:"navbar h-24 px-8 md:px-12 bg-base-100 max-w[1500]px",children:[e.jsx("div",{children:e.jsxs("div",{className:"dropdown md:hidden mr-4",children:[e.jsx("label",{tabIndex:0,className:"btn btn-ghost hover:bg-transparent btn-lg px-0",children:e.jsx(ve,{className:"w-16 h-16"})}),e.jsxs("ul",{tabIndex:0,className:"menu menu-lg dropdown-content mt-3 z-[1] p-2 bg-base-100 shadow",children:[e.jsx("li",{children:e.jsx(b,{to:"leaderboard",children:"Leaderboard"})}),e.jsx("li",{children:e.jsx(b,{to:"models",children:"Models"})}),e.jsx("li",{children:e.jsx(b,{to:"scenarios",children:"Scenarios"})}),e.jsx("li",{children:e.jsx(b,{to:"runs",className:"whitespace-nowrap",children:"Predictions"})}),e.jsx("li",{children:e.jsx(b,{to:"https://github.com/stanford-crfm/helm",children:"GitHub"})})]})]})}),e.jsxs("div",{className:"flex-1 items-center",children:[e.jsx("a",{href:"https://crfm.stanford.edu/",className:"w-24",children:e.jsx("img",{src:we,className:"object-contain"})}),e.jsx(b,{to:"/",className:"mx-2 w-32",children:e.jsx("img",{src:ye,className:"object-contain"})}),e.jsx(Ee,{})]}),e.jsx("div",{className:"flex-none hidden md:block",children:e.jsxs("ul",{className:"flex flex-row gap-6 px-1",children:[e.jsx("li",{children:e.jsx(b,{to:"leaderboard",children:"Leaderboard"})}),e.jsx("li",{children:e.jsx(b,{to:"models",children:"Models"})}),e.jsx("li",{children:e.jsx(b,{to:"scenarios",children:"Scenarios"})}),e.jsx("li",{children:e.jsx(b,{to:"runs",children:"Predictions"})}),e.jsx("li",{children:e.jsx(b,{to:"https://github.com/stanford-crfm/helm",children:"GitHub"})}),e.jsx("li",{className:"hidden lg:flex",children:e.jsx(Ns,{})})]})})]})}function Ms(){return e.jsxs("nav",{className:"navbar h-24 px-8 md:px-12 bg-base-100 max-w[1500]px",children:[e.jsx("div",{children:e.jsx("div",{className:"dropdown md:hidden mr-4",children:e.jsx("label",{tabIndex:0,className:"btn btn-ghost hover:bg-transparent btn-lg px-0",children:e.jsx(ve,{className:"w-16 h-16"})})})}),e.jsxs("div",{className:"flex-1 items-center",children:[e.jsx(b,{to:"https://crfm.stanford.edu/",className:"w-24",children:e.jsx("img",{src:we,className:"object-contain"})}),e.jsx(b,{to:"/",className:"mx-2 w-32",children:e.jsx("img",{src:ye,className:"object-contain"})}),e.jsx(Ee,{})]})]})}function Ss(){return e.jsxs(e.Fragment,{children:[window.PROJECT_ID==="home"?e.jsx(Ms,{}):e.jsx(Es,{}),e.jsx("main",{className:"p-8 pt-0",children:e.jsx("div",{className:"mx-auto max-w-[1500]px",children:e.jsx(He,{})})})]})}async function D(s){try{return await(await fetch(S(`${z()}/schema.json`),{signal:s})).json()}catch(t){return t instanceof Error&&t.name!=="AbortError"&&console.log(t),{adapter:[],metric_groups:[],metrics:[],models:[],perturbations:[],run_groups:[]}}}function Ls({href:s,children:t}){return e.jsx("a",{href:s,className:"link link-primary link-hover",target:"_blank",rel:"noreferrer",children:t})}function B({value:s}){return e.jsx("span",{children:e.jsx(De,{components:{a:Ls},children:s})})}function $({title:s,subtitle:t,markdown:n=!1}){return e.jsxs("header",{className:"m-4 ml-0",children:[e.jsx("h1",{className:"text-4xl",children:s}),n&&t!==void 0?e.jsx("h2",{className:"mt-2 text-neutral",children:e.jsx(B,{value:t})}):t!==void 0&&e.jsx("h2",{className:"mt-2 text-neutral",children:t})]})}const ks={open:"green",limited:"yellow",closed:"red"},Rs={open:"Open",limited:"Limited",closed:"Closed"};function _s({level:s}){return e.jsx(O,{color:ks[s],children:Rs[s]})}function P(){return e.jsx("div",{className:"w-full",children:e.jsx("div",{className:"block mx-auto my-24 loading loading-spinner loading-lg"})})}function Cs(){const[s,t]=i.useState([]);i.useEffect(()=>{const o=new AbortController;async function d(){const c=await D(o.signal);t(c.models)}return d(),()=>o.abort()},[]);const[n,a,r]=s.reduce((o,d)=>{switch(d.access){case"open":o[0]+=1;break;case"limited":o[1]+=1;break;case"closed":o[2]+=1;break}return o},[0,0,0]),l=Object.values(s.reduce((o,d)=>{const c=d.creator_organization;return o[c]===void 0?(o[c]={name:c,models:1},o):(o[c].models+=1,o)},{}));return s.length===0?e.jsx(P,{}):e.jsxs(e.Fragment,{children:[e.jsx($,{title:"Models"}),e.jsxs("div",{className:"overflow-x-auto mt-12",children:[e.jsxs("table",{className:"table",children:[e.jsx("thead",{children:e.jsxs("tr",{children:[e.jsx("th",{children:"Creator"}),e.jsx("th",{children:"Model"}),e.jsx("th",{children:"Description"}),e.jsx("th",{children:"Access"})]})}),e.jsx("tbody",{children:s.map(o=>e.jsxs("tr",{children:[e.jsx("td",{className:"text-lg",children:o.creator_organization}),e.jsxs("td",{children:[e.jsx("span",{className:"text-xl",children:o.display_name}),e.jsx("br",{}),e.jsx("span",{children:o.name})]}),e.jsx("td",{children:e.jsx(B,{value:o.description})}),e.jsx("td",{children:e.jsx(_s,{level:o.access})})]}))})]}),e.jsx($,{title:"Analysis"}),e.jsxs("div",{className:"grid md:grid-cols-3 grid-cols-1 gap-8",children:[e.jsxs(W,{className:"flex flex-col justify-between",children:[e.jsx(Y,{children:"Models"}),e.jsx(ge,{className:"text-6xl md:!text-[96px]",children:s.length}),e.jsx(We,{values:[n,a,r],colors:["green","yellow","red"]}),e.jsx(oe,{categories:["Open","Limited","Closed"],colors:["green","yellow","red"]})]}),e.jsxs(W,{className:"md:col-span-2",children:[e.jsx(Y,{children:"Creator Organizations"}),e.jsxs("div",{className:"flex justify-between mt-4",children:[e.jsx(ze,{data:l,category:"models",index:"name",variant:"pie",className:"basis-5/12"}),e.jsx(oe,{categories:l.map(o=>o.name),className:"basis-7/12"})]})]})]})]})]})}function V({to:s,children:t,inTable:n=!1,title:a=""}){return n?e.jsx(b,{className:"link link-hover",to:s,title:a,children:t}):e.jsx(b,{className:"link link-primary link-hover",to:s,children:t})}function As(){const[s,t]=i.useState([]);i.useEffect(()=>{const a=new AbortController;async function r(){const l=await D(a.signal);t(l.run_groups.filter(o=>!o.todo&&o.taxonomy&&!o.display_name.includes("CLEVA")))}return r(),()=>a.abort()},[]);const n=Object.values(s.reduce((a,r)=>{var o;const l=((o=r.taxonomy)==null?void 0:o.task)||"Unknown";return a[l]===void 0?(a[l]={name:l,value:1},a):(a[l].value+=1,a)},{}));return s.length===0?e.jsx(P,{}):(console.log(s),e.jsxs(e.Fragment,{children:[e.jsx($,{title:"Scenarios",subtitle:"A scenario represents a use case and consists of a dataset of instances."}),e.jsxs("div",{className:"overflow-x-auto mt-12",children:[e.jsxs("table",{className:"table",children:[e.jsx("thead",{children:e.jsxs("tr",{children:[e.jsx("th",{children:"Scenario"}),e.jsx("th",{children:"Task"}),e.jsx("th",{children:"What"}),e.jsx("th",{children:"Who"}),e.jsx("th",{children:"When"}),e.jsx("th",{children:"Language"}),e.jsx("th",{children:"Description"})]})}),e.jsx("tbody",{children:s.map(a=>{var r,l,o,d,c;return e.jsxs("tr",{children:[e.jsxs("td",{children:[e.jsx(V,{to:`/leaderboard/${a.name}`,children:e.jsx("span",{className:"text-lg",children:a.display_name})}),e.jsx("span",{className:"block",children:a.name})]}),e.jsx("td",{children:((r=a.taxonomy)==null?void 0:r.task)||""}),e.jsx("td",{children:((l=a.taxonomy)==null?void 0:l.what)||""}),e.jsx("td",{children:((o=a.taxonomy)==null?void 0:o.who)||""}),e.jsx("td",{children:((d=a.taxonomy)==null?void 0:d.when)||""}),e.jsx("td",{children:((c=a.taxonomy)==null?void 0:c.language)||""}),e.jsx("td",{children:e.jsx(B,{value:a.description})})]})})})]}),e.jsx($,{title:"Analysis"}),e.jsxs("div",{className:"grid md:grid-cols-4 gap-8",children:[e.jsxs(W,{className:"flex flex-col",children:[e.jsx(Y,{children:"Total scenarios"}),e.jsx(ge,{className:"mx-auto my-6 md:mt-16 !text-[72px] md:!text-[96px]",children:s.length})]}),e.jsx(W,{className:"col-span-3",children:e.jsxs("div",{className:"grid md:grid-cols-2 gap-x-12",children:[e.jsx(ce,{data:n.slice(0,Math.floor(n.length/2))}),e.jsx(ce,{data:n.slice(Math.ceil(n.length/2))})]})})]})]})]}))}function Ts(){return S(`${z()}/groups.json`)}async function Me(s){try{return await(await fetch(Ts(),{signal:s})).json()}catch(t){return t instanceof Error&&t.name!=="AbortError"&&console.log(t),[]}}function re({children:s}){return e.jsx("div",{role:"navigation",className:"tabs flex-nowrap border-b-2 border-gray-2 overflow-x-auto overflow-y-hidden",children:s})}function J({active:s=!1,onClick:t=()=>{},size:n="md",children:a}){return e.jsx("div",{onClick:t,className:`whitespace-nowrap text-${n} mb-[-2px] text-md tab tab-bordered${s?" border-2 border-grey-500 rounded":" border-none"}`,children:a})}function Is({title:s,titleId:t,...n},a){return i.createElement("svg",Object.assign({xmlns:"http://www.w3.org/2000/svg",viewBox:"0 0 20 20",fill:"currentColor","aria-hidden":"true",ref:a,"aria-labelledby":t},n),s?i.createElement("title",{id:t},s):null,i.createElement("path",{fillRule:"evenodd",d:"M4.25 5.5a.75.75 0 00-.75.75v8.5c0 .414.336.75.75.75h8.5a.75.75 0 00.75-.75v-4a.75.75 0 011.5 0v4A2.25 2.25 0 0112.75 17h-8.5A2.25 2.25 0 012 14.75v-8.5A2.25 2.25 0 014.25 4h5a.75.75 0 010 1.5h-5z",clipRule:"evenodd"}),i.createElement("path",{fillRule:"evenodd",d:"M6.194 12.753a.75.75 0 001.06.053L16.5 4.44v2.81a.75.75 0 001.5 0v-4.5a.75.75 0 00-.75-.75h-4.5a.75.75 0 000 1.5h2.553l-9.056 8.194a.75.75 0 00-.053 1.06z",clipRule:"evenodd"}))}const Ps=i.forwardRef(Is),me=Ps;function F(s){return Number.isNaN(Number(s))?String(s):String(Math.round(Number(s)*1e3)/1e3)}function he({value:s,title:t,hideIcon:n}){if(typeof s.value=="string"&&s.value.includes("⚠")&&(s.value=s.value.replace("⚠","")),s.value===void 0)return"-";if(s.run_spec_names){const a=(()=>{if(s.run_spec_names.length==1)return"/runs/"+s.run_spec_names[0];if(s.run_spec_names.length>1){const r="/runs/?q="+s.run_spec_names.map(o=>`^${o}$`).join("|");return encodeURI(r)}})();return a?e.jsx(V,{to:a,inTable:!0,title:t,children:e.jsxs("div",{className:"flex items-center ",children:[F(s.value),!n&&e.jsx(me,{className:"w-3 h-3 ml-1",stroke:"#cbcbcb",fill:"#cbcbcb"})]})}):t?e.jsx("a",{title:t,children:F(s.value)}):e.jsx(e.Fragment,{children:F(s.value)})}return s.href?e.jsx(V,{to:s.href,inTable:!0,title:t,children:e.jsxs("div",{className:"flex items-center",children:[F(s.value),!n&&e.jsx(me,{className:"w-3 h-3 ml-1",stroke:"#cbcbcb",fill:"#cbcbcb"})]})}):s.markdown?e.jsx(B,{value:String(s.value)}):t?e.jsx("a",{title:t,children:F(s.value)}):e.jsx(e.Fragment,{children:F(s.value)})}function le({schema:s,groupTable:t,numRowsToDisplay:n,sortColumnIndex:a=1,sortable:r=!0,displayColumnIndexes:l=void 0,miniStyle:o=!1}){const[d,c]=i.useState(1),[h,u]=i.useState(a);function g(m){return m.length>30?m.substring(0,27)+"...":m}const w=m=>{const j=["AIRBench 2024 -","-book"];if(m.value==="Model/adapter")return"Model";if(j.some(f=>m.value.includes(f))){let f=m.value;return j.forEach(M=>{f=f.replace(M,"")}),g(f)}else return g(m.value)},v=m=>{if(s){const j=s.models.find(f=>f.display_name===m);if(j){let f=j.description;return f.includes("/")&&(f=f.replace("/","_")),f}}return""},E=m=>{c(m===h?d*-1:m===0?-1:1),u(m)},p=m=>{if(s){const j=s.models.find(f=>f.display_name===m);if(j){let f=j.name;return f.includes("/")&&(f=f.replace("/","_")),f}}return""},R=()=>{const m=t.header[h].lower_is_better,j=d*(m?1:-1),f=t.rows.slice();return f.sort((M,k)=>{var C,A;const y=(C=M[h])==null?void 0:C.value,N=(A=k[h])==null?void 0:A.value;return y!==void 0&&N===void 0?-1:N!==void 0&&y===void 0?1:typeof y=="number"&&typeof N=="number"?(y-N)*j:typeof y=="string"&&typeof N=="string"?j===1?y.localeCompare(N):N.localeCompare(y):0}),n>0?f.slice(0,n):f};function x(m){const j=m.lastIndexOf(" - ");return j===-1?m:m.substring(0,j)+"*"+m.substring(j+1)}const _=m=>{const f=x(m).split("*")[0].trim();if(s){const M=s.run_groups.find(k=>k.display_name===f||k.short_display_name===f);if(M)return M.name}return""};return e.jsxs("table",{className:o?"table w-full":"rounded-lg shadow-md table",children:[e.jsx("thead",{children:e.jsx("tr",{children:t.header.filter((m,j)=>l===void 0||l.includes(j)).map((m,j)=>e.jsx("th",{className:`${j===h?"bg-gray-100":"bg-white"} ${j===0?"left-0 z-40":""} ${m.description?"underline decoration-dashed decoration-gray-300	":""}  whitespace-nowrap px-4 sticky top-0`,title:m.description?m.description:"",children:e.jsxs("div",{className:o?"flex gap-2 items-center":"z-20 flex justify-between items-center min-w-48 w-48 max-w-48 text-wrap",children:[e.jsx("span",{className:"inline-block w-full break-words",children:w(m)}),r?e.jsx("button",{className:"link",onClick:()=>E(j),children:e.jsx(js,{className:"w-6 h-6"})}):null]})},`$${j}`))})}),e.jsx("tbody",{children:R().map((m,j)=>e.jsx("tr",{children:m.filter((f,M)=>l===void 0||l.includes(M)).map((f,M)=>e.jsx("td",{className:`${M===0?"z-20 text-lg sticky left-0":"z-0"} ${j%2===0?"bg-gray-50":"bg-white"}`,children:M==1?e.jsx("div",{className:`${f&&f.style&&f.style["font-weight"]&&f.style["font-weight"]==="bold"?"font-bold":""}`,children:e.jsx(he,{value:{...f,href:"/runs/?q="+p(String(m[0].value))},title:`Click value to see all predictions for: ${p(String(m[0].value))}`})}):e.jsx("div",{className:`${f&&f.style&&f.style["font-weight"]&&f.style["font-weight"]==="bold"?"font-bold":""} ${M===0?"underline decoration-dashed decoration-gray-300 z-10":"z-0"}`,children:e.jsx(he,{value:{...f},title:String(m[0].value)===f.value?v(String(m[0].value)):`Click value to see predictions for ${String(m[0].value)} for ${_(w(t.header[M]))}: ${p(String(m[0].value))}`})})},`${M}`))},`$${m[0].value}`))})]})}function $s(){const[s,t]=i.useState(0),[n,a]=i.useState(),[r,l]=i.useState();return i.useEffect(()=>{const o=new AbortController;async function d(){const c=D(o.signal),h=Me(o.signal),u=await c;l(u);const g=await h;a(g)}return d(),()=>o.abort()},[]),n===void 0||r===void 0?e.jsx(P,{}):n.length===0?e.jsxs("div",{children:[e.jsx($,{title:"Results",subtitle:"Groupings of the processes, methods, and metrics involved in evaluating models, particularly in the context of natural language understanding and question answering.",className:"mb-16"}),e.jsx("div",{children:"No groups found."})]}):e.jsxs("div",{children:[e.jsx($,{title:"Results",subtitle:"Groupings of the processes, methods, and metrics involved in evaluating models, particularly in the context of natural language understanding and question answering.",className:"mb-16"}),e.jsxs("div",{children:[n.length>1?e.jsx(re,{children:n.map((o,d)=>e.jsx(J,{active:d===s,onClick:()=>t(d),children:o.title},d))}):null,e.jsx(le,{schema:r,groupTable:n[s],numRowsToDisplay:-1,sortColumnIndex:1,sortable:!0},`${s}`)]})]})}async function Hs(s){try{return await(await fetch(S(`${z()}/run_specs.json`),{signal:s})).json()}catch(t){return t instanceof Error&&t.name!=="AbortError"&&console.log(t),[]}}function ie({currentPage:s,totalPages:t,onNextPage:n,onPrevPage:a,className:r}){let l="join";return r!==void 0&&(l=`join ${r}`),e.jsxs("div",{className:l,children:[e.jsx("button",{onClick:a,className:"join-item btn",children:"«"}),e.jsxs("button",{className:"join-item btn",children:["Page ",s," of ",t]}),e.jsx("button",{onClick:n,className:"join-item btn",children:"»"})]})}const X=100;function Ds(){const[s,t]=se(),[n,a]=i.useState(),[r,l]=i.useState(Number(s.get("page")||1)),[o,d]=i.useState(!0),[c,h]=i.useState(s.get("q")||"");i.useEffect(()=>{const p=new AbortController;async function R(){const x=await Hs(p.signal);a(x)}return R(),()=>p.abort()},[]);const u=p=>{p.preventDefault();const x=p.target.q.value;h(x),t({q:x,page:"1"})};if(n===void 0)return e.jsx(P,{});const g=o?new RegExp(c):null,w=n.filter(p=>g?g.test(p.name):p.name.includes(c)),v=w.slice((r-1)*X,r*X),E=Math.ceil(w.length/X);return e.jsxs(e.Fragment,{children:[e.jsx($,{title:"Predictions",subtitle:"All benchmark predictions"}),e.jsxs("form",{className:"flex mb-8",onSubmit:u,children:[e.jsxs("div",{className:"form-control",children:[e.jsx("input",{type:"text",name:"q",placeholder:"Search",className:"input input-bordered",value:c,onChange:p=>h(p.target.value)}),e.jsxs("label",{className:"label",children:[e.jsxs("span",{className:"label-text-alt flex item-center",children:[e.jsx("input",{type:"checkbox",className:"toggle toggle-xs",checked:o,onChange:()=>d(!o)}),e.jsx("span",{className:"ml-2",children:"Regex"})]}),e.jsx("span",{className:"label-text-alt",children:`${w.length} results`})]})]}),e.jsx("div",{className:"form-control ml-4",children:e.jsx("button",{className:"btn",children:e.jsx(ws,{className:"w-6 h-6"})})})]}),e.jsx("div",{className:"overflow-x-auto",children:e.jsxs("table",{className:"table",children:[e.jsx("thead",{children:e.jsxs("tr",{children:[e.jsx("th",{children:"Run"}),e.jsx("th",{children:"Model"}),e.jsx("th",{children:"Groups"}),e.jsx("th",{children:"Adapter method"}),e.jsx("th",{children:"Subject / Task"})]})}),e.jsx("tbody",{children:v.map((p,R)=>e.jsxs("tr",{children:[e.jsx("td",{children:e.jsx(V,{to:`/runs/${p.name}`,children:p.name})}),e.jsx("td",{children:p.adapter_spec.model}),e.jsx("td",{children:p.groups.join(", ")}),e.jsx("td",{children:p.adapter_spec.method}),e.jsx("td",{children:p.scenario_spec.args.subject||p.scenario_spec.args.task||"-"})]},`${p.name}-${R}`))})]})}),E>0?e.jsx(ie,{className:"flex justify-center my-8",onNextPage:()=>{const p=Math.min(r+1,E);l(p),s.set("page",String(p)),t(s)},onPrevPage:()=>{const p=Math.max(r-1,1);l(p),s.set("page",String(p)),t(s)},currentPage:r,totalPages:E}):e.jsx("div",{className:"my-8 text-center",children:"No results"})]})}function I(){return window.SUITE!==void 0?window.SUITE:void 0}async function Os(s,t,n){try{return await(await fetch(S(`/runs/${n||I()}/${s}/scenario.json`),{signal:t})).json()}catch(a){a instanceof Error&&a.name!=="AbortError"&&console.log(a);return}}function Se(s,t){return S(`/runs/${t||I()}/${s}/run_spec.json`)}async function Fs(s,t,n){try{return await(await fetch(Se(s,n),{signal:t})).json()}catch(a){a instanceof Error&&a.name!=="AbortError"&&console.log(a);return}}function Bs(s,t){return S(`/runs/${t||I()}/${s}/scenario_state.json`)}function Le(){return window.RELEASE!==void 0?window.RELEASE:void 0}async function Us(s){try{return await(await fetch(S(`/releases/${Le()}/runs_to_run_suites.json`),{signal:s})).json()}catch(t){return t instanceof Error&&t.name!=="AbortError"&&console.log(t),{}}}function Ws(s,t){return Le()?s[t]:window.SUITE}function zs(s){const n={quasi_exact_match:!1,toxic_frac:!0,safety_score:!1,exact_match:!1},a=Object.keys(s);for(const r of a)if(s[r]!==void 0&&n[r]!==void 0)return n[r]?s[r]<.5?[r,!0]:[r,!1]:s[r]>=.5?[r,!0]:[r,!1];return["",!1]}function qs(s){const[t,n]=zs(s.stats);return t===""?null:n?e.jsx(Gs,{value:`${t.replace(/_/g," ")}: ${s.stats[t]}`}):e.jsx(Vs,{value:`${t.replace(/_/g," ")}: ${s.stats[t]}`})}function Gs({value:s}){return e.jsx(O,{icon:ss,color:"green",children:s})}function Vs({value:s}){return e.jsx(O,{icon:as,color:"red",children:s})}function H({value:s}){const[t,n]=i.useState(!1),[a,r]=i.useState(!1);return e.jsxs(e.Fragment,{children:[e.jsxs("div",{onMouseOver:()=>n(!0),onMouseOut:()=>n(!1),className:"relative",children:[e.jsx("div",{className:"bg-base-200 p-2 block overflow-auto w-full max-h-[36rem] mb-2 whitespace-pre-wrap",children:s}),t?e.jsx("button",{className:"bg-white absolute p-2 leading-none height-fit min-h-none right-1 bottom-1 shadow",onClick:()=>r(!0),children:e.jsx(us,{fill:"black",color:"black",className:"text w-4 h-4"})}):null]}),e.jsx("dialog",{open:a,className:"modal p-16 bg-opacity-80 bg-white",onClick:()=>r(!1),children:e.jsx("div",{className:"modal-box max-w-none p-4 whitespace-pre-wrap bg-base-200",children:s})})]})}function ke({messages:s}){return e.jsx("div",{children:s.map((t,n)=>e.jsxs("div",{children:[e.jsx("div",{children:t.role}),e.jsx(H,{value:t.content})]},n))})}function Re({mediaObject:s}){if(s.content_type.includes("image")){if(s.location===void 0)return null;const t=S(s.location.replace("benchmark_output/","").replace("prod_env/","../"));return e.jsxs("div",{children:[e.jsx("img",{src:t}),e.jsx("br",{})]})}else if(s.content_type.includes("audio")){if(s.location===void 0)return null;const t=S(s.location.replace("benchmark_output/","").replace("prod_env/","../"));return e.jsx("div",{children:e.jsx("audio",{controls:!0,src:t})})}else return s.text&&s.content_type&&s.content_type==="text/plain"&&s.text.length>1?e.jsxs("div",{children:[s.text,e.jsx("br",{}),e.jsx("br",{})]}):e.jsx("div",{})}function _e({multimediaObject:s}){return e.jsx("div",{children:s.media_objects.map(t=>e.jsx(Re,{mediaObject:t}))})}function Js(s){return Array.isArray(s)?s.length==0?"[]":`[${s.map(t=>String(t).replace(/\n/,"\\n")).join(", ")}]`:String(s)}function Ks({request:s}){return e.jsxs("div",{children:[s.request.prompt.length>0?e.jsxs("div",{children:[e.jsxs("h3",{className:"block text text-gray-400",children:["Prompt (",s.request.prompt.length," Chars)"]}),e.jsx(H,{value:s.request.prompt})]}):s.request.multimodal_prompt?e.jsxs("div",{children:[e.jsx("h3",{className:"block text text-gray-400",children:"Prompt"}),e.jsx(_e,{multimediaObject:s.request.multimodal_prompt})]}):s.request.messages&&s.request.messages.length?e.jsxs("div",{children:[e.jsx("h3",{className:"block text text-gray-400",children:"Prompt"}),e.jsx(ke,{messages:s.request.messages})]}):e.jsx("h3",{className:"block text text-gray-400",children:"Empty Prompt"}),e.jsx(te,{children:Object.keys(s.request).filter(t=>t!=="prompt").map((t,n)=>e.jsxs(ne,{children:[e.jsxs("span",{children:[t,":"]}),s.request&&s.request[t]?e.jsx("span",{children:Js(s.request[t])}):"null"]},n+1))})]})}function Zs(s){return e.jsx("div",{children:s.map((t,n)=>e.jsxs("div",{children:[t.error&&e.jsxs("div",{children:[e.jsx("h3",{className:"ml-1",children:"Error"}),e.jsx(H,{value:t.error})," "]}),t.text&&e.jsxs("div",{children:[e.jsx("h3",{className:"ml-1",children:"Text"}),e.jsx(H,{value:t.text})," "]}),t.media_object&&e.jsx(Re,{mediaObject:t.media_object})]},n))})}function Xs(s){return e.jsx("div",{children:Object.entries(s).map(([t,n])=>e.jsxs("div",{children:[e.jsx("h3",{className:"ml-1",children:t}),e.jsx(H,{value:n===null?"null":n.toString()})]}))})}function Qs({predictionAnnotations:s}){return e.jsx("div",{children:s&&s!==void 0?Object.entries(s).map(([t,n])=>e.jsxs("details",{className:"collapse collapse-arrow border rounded-md bg-white my-2",children:[e.jsx("summary",{className:"collapse-title",children:e.jsx(e.Fragment,{children:"View "+t+" annotations"})}),e.jsx("div",{className:"collapse-content",children:Array.isArray(n)?Zs(n):Xs(n)})]},t)):null})}function Ys({predictions:s,requests:t,metricFieldMap:n}){return s.length<1?null:e.jsx("div",{children:e.jsx("div",{className:"flex flex-wrap justify-start items-start",children:s.map((a,r)=>e.jsxs("div",{className:"w-full",children:[s.length>1?e.jsxs("h2",{children:["Trial ",a.train_trial_index]}):null,e.jsx("div",{className:"mt-2 w-full",children:a.base64_images&&a.base64_images.length>0?e.jsxs(e.Fragment,{children:[e.jsx("h3",{className:"mr-4",children:"Prediction image"}),a.base64_images.map(l=>e.jsx("img",{src:"data:image;base64,"+l,alt:"Base64 Image"}))]}):e.jsxs(e.Fragment,{children:[e.jsxs("h3",{children:[e.jsx("span",{className:"mr-4",children:"Prediction raw text"}),e.jsx(qs,{stats:a.stats})]}),e.jsx(H,{value:a.predicted_text}),a.mapped_output?e.jsxs(e.Fragment,{children:[e.jsx("h3",{children:"Prediction mapped output"}),e.jsx(H,{value:String(a.mapped_output)})]}):null]})}),e.jsx(Qs,{predictionAnnotations:a.annotations}),e.jsxs("div",{className:"mx-1",children:[e.jsx("h3",{children:"Metrics"}),e.jsx(te,{children:Object.keys(a.stats).map((l,o)=>e.jsxs(ne,{children:[n[l]?e.jsx("span",{title:n[l].description,children:n[l].display_name}):e.jsx("span",{children:l}),e.jsx("span",{children:String(a.stats[l])})]},o))})]}),e.jsxs("details",{className:"collapse collapse-arrow border rounded-md bg-white",children:[e.jsx("summary",{className:"collapse-title",children:"Request details"}),e.jsx("div",{className:"collapse-content",children:e.jsx(Ks,{request:t[r]})})]})]},r))})})}const et="correct";function st({references:s}){return e.jsxs("span",{children:[e.jsx("h3",{children:"References"}),e.jsx("ul",{children:s.map((t,n)=>e.jsxs("li",{className:"bg-base-200 p-2 block overflow-auto w-full max-h-72 mb-2 whitespace-pre-wrap",children:[t.output.text,t.tags.map(a=>e.jsx(O,{className:"mx-2",color:a===et?"green":void 0,children:a}))]},n))})]})}function Ce(s){return e.jsx(H,{value:s===null?"null":typeof s=="object"?JSON.stringify(s):s.toString()})}function tt(s){return e.jsx("div",{children:s.map(t=>Ce(t))})}function nt({extraData:s}){return e.jsxs("details",{className:"collapse collapse-arrow border rounded-md bg-white my-2",children:[e.jsx("summary",{className:"collapse-title",children:"View instance extra data"}),e.jsx("div",{className:"collapse-content",children:Object.entries(s).map(([t,n])=>e.jsxs("div",{children:[e.jsx("h3",{className:"ml-1",children:t}),Array.isArray(n)?tt(n):Ce(n)]}))})]})}function at({instance:s,requests:t,predictions:n,metricFieldMap:a}){return e.jsxs("div",{children:[e.jsx("h3",{children:"Input"}),s.input.multimedia_content!==void 0?e.jsx(_e,{multimediaObject:s.input.multimedia_content}):s.input.text.includes('<br><img src="data:image;base64')?e.jsx("div",{dangerouslySetInnerHTML:{__html:s.input.text}}):s.input.messages!==void 0&&s.input.messages.length?e.jsx(ke,{messages:s.input.messages}):e.jsx(H,{value:s.input.text}),e.jsx("div",{children:s.references&&s.references.length>0?e.jsx(st,{references:s.references}):null}),s.extra_data&&s.extra_data.length?e.jsx(nt,{extraData:s.extra_data}):null,e.jsx("div",{children:n&&t?e.jsx(Ys,{predictions:n,requests:t,metricFieldMap:a}):null})]})}function Z(s){return s.includes("gpqa")||s.includes("ewok")}async function ue(s,t,n,a){const r=h=>Uint8Array.from(atob(h),u=>u.charCodeAt(0)),l=await window.crypto.subtle.importKey("raw",r(t),"AES-GCM",!0,["decrypt"]),o=new Uint8Array([...r(s),...r(a)]),d=r(n),c=await window.crypto.subtle.decrypt({name:"AES-GCM",iv:d},l,o);return new TextDecoder().decode(c)}async function rt(s,t,n,a){try{const l=await(await fetch(S(`/runs/${n||I()}/${s}/instances.json`),{signal:t})).json();if(Z(s)&&a){const d=await(await fetch(S(`/runs/${n||I()}/${s}/encryption_data.json`),{signal:t})).json();for(const c of l){const h=d[c.input.text];h&&(c.input.text="encrypted",c.input.text=await ue(h.ciphertext,h.key,h.iv,h.tag));for(const u of c.references){const g=d[u.output.text];g&&(u.output.text=await ue(g.ciphertext,g.key,g.iv,g.tag))}}}return l}catch(r){return r instanceof Error&&r.name!=="AbortError"&&console.log(r),[]}}async function lt(s,t,n,a){const r=h=>Uint8Array.from(atob(h),u=>u.charCodeAt(0)),l=await window.crypto.subtle.importKey("raw",r(t),"AES-GCM",!0,["decrypt"]),o=new Uint8Array([...r(s),...r(a)]),d=r(n),c=await window.crypto.subtle.decrypt({name:"AES-GCM",iv:d},l,o);return new TextDecoder().decode(c)}async function it(s,t,n,a){try{const l=await(await fetch(S(`/runs/${n||I()}/${s}/display_predictions.json`),{signal:t})).json();if(Z(s)&&a){const d=await(await fetch(S(`/runs/${n||I()}/${s}/encryption_data.json`),{signal:t})).json();for(const c of l){const h=c.predicted_text,u=d[h];if(u)try{c.predicted_text=await lt(u.ciphertext,u.key,u.iv,u.tag)}catch(g){console.error(`Failed to decrypt predicted_text for instance_id: ${c.instance_id}`,g)}}}return l}catch(r){return r instanceof Error&&r.name==="AbortError"&&console.log(r),[]}}async function ot(s,t,n,a){const r=h=>Uint8Array.from(atob(h),u=>u.charCodeAt(0)),l=await window.crypto.subtle.importKey("raw",r(t),"AES-GCM",!0,["decrypt"]),o=new Uint8Array([...r(s),...r(a)]),d=r(n),c=await window.crypto.subtle.decrypt({name:"AES-GCM",iv:d},l,o);return new TextDecoder().decode(c)}async function ct(s,t,n,a){try{const l=await(await fetch(S(`/runs/${n||I()}/${s}/display_requests.json`),{signal:t})).json();if(Z(s)&&a){const d=await(await fetch(S(`/runs/${n||I()}/${s}/encryption_data.json`),{signal:t})).json();for(const c of l){const h=c.request.prompt,u=d[h];if(u)try{c.request.prompt=await ot(u.ciphertext,u.key,u.iv,u.tag)}catch(g){console.error(`Failed to decrypt prompt for instance_id: ${c.instance_id}`,g)}}}return l}catch(r){return r instanceof Error&&r.name!=="AbortError"&&console.log(r),[]}}const q=10;function dt({runName:s,suite:t,metricFieldMap:n,userAgreed:a}){const[r,l]=se(),[o,d]=i.useState([]),[c,h]=i.useState(),[u,g]=i.useState(),[w,v]=i.useState(1);i.useEffect(()=>{const x=new AbortController;async function _(){const m=x.signal,[j,f,M]=await Promise.all([rt(s,m,t,a),it(s,m,t,a),ct(s,m,t,a)]);d(j);const k={};M.forEach(N=>{var U;const C=N.instance_id,A=((U=N.perturbation)==null?void 0:U.name)||"";k[C]===void 0&&(k[C]={}),k[C][A]===void 0&&(k[C][A]=[]),k[C][A].push(N)}),g(k);const y={};f.forEach(N=>{var U;const C=N.instance_id,A=((U=N.perturbation)==null?void 0:U.name)||"";y[C]===void 0&&(y[C]={}),y[C][A]===void 0&&(y[C][A]=[]),y[C][A].push(N)}),h(y)}return _(),()=>x.abort()},[s,t,a]);const E=o.slice((w-1)*q,(w-1)*q+q),p=Math.ceil(o.length/q);i.useEffect(()=>{const x=r.get("instance");if(x&&!window.helmHasScrolledToInstance&&E.length>0){if(E.findIndex(m=>m.id===x)===-1)return;requestAnimationFrame(()=>{const m=document.getElementById(`instance-${x}`);m&&m.scrollIntoView({behavior:"smooth"})}),window.helmHasScrolledToInstance=!0}},[r,w,l,E]);const R=x=>x.perturbation===void 0?`Instance id: ${x.id} [split: ${x.split}]`:`Instance id: ${x.id} [split: ${x.split}][perturbation: ${x.perturbation.name}]`;return c===void 0||u===void 0?e.jsx(P,{}):e.jsxs(e.Fragment,{children:[e.jsx("div",{className:"grid gap-8",children:E.map((x,_)=>{var m,j;return e.jsxs("div",{id:"instance-"+x.id,className:"border p-4",children:[e.jsxs("div",{className:"flex items-center justify-between",children:[e.jsx("h3",{className:"text-xl mb-4",children:R(x)}),e.jsx("button",{className:"btn btn-sm normal-case px-2 py-1",onClick:()=>{const f=window.location.href+(window.location.href.includes("?")?"&instance=":"?instance=")+x.id;navigator.clipboard.writeText(f)},children:"Copy Link"})]}),e.jsx(at,{instance:x,requests:u[x.id][((m=x.perturbation)==null?void 0:m.name)||""],predictions:c[x.id][((j=x.perturbation)==null?void 0:j.name)||""],metricFieldMap:n},`${x.id}-${_}`)]})})}),e.jsx(ie,{className:"flex justify-center my-8",onNextPage:()=>{const x=Math.min(w+1,p);v(x),r.set("instancesPage",String(x)),l(r)},onPrevPage:()=>{const x=Math.max(w-1,1);v(x),r.set("instancesPage",String(x)),l(r)},currentPage:w,totalPages:p})]})}async function mt(s,t,n){try{return await(await fetch(S(`/runs/${n||I()}/${s}/stats.json`),{signal:t})).json()}catch(a){return a instanceof Error&&a.name!=="AbortError"&&console.log(a),[]}}function ht({stat:s,metricFieldMap:t}){const n=`${s.name.split!==void 0?` on ${s.name.split}`:""}${s.name.sub_split!==void 0?`/${s.name.sub_split}`:""}${s.name.perturbation!==void 0?` with ${s.name.perturbation.name}`:" original"}`;return t[s.name.name]?e.jsxs("span",{title:t[s.name.name].description,children:[e.jsx("strong",{children:t[s.name.name].display_name||s.name.name}),n]}):e.jsxs("span",{children:[e.jsx("strong",{children:s.name.name}),n]})}const G=50,xe=["name","mean","min","max","sum","sum_squared","variance","stddev"];function ut({runName:s,suite:t,metricFieldMap:n}){const[a,r]=se(),[l,o]=i.useState(),[d,c]=i.useState(1),[h,u]=i.useState("");if(i.useEffect(()=>{const v=new AbortController;async function E(){const p=v.signal,R=await mt(s,p,t);o(R)}return E(),()=>v.abort()},[s,t]),l===void 0||l.length===0)return e.jsx(P,{});const g=Math.floor(l.length/G),w=l.slice((d-1)*G,(d-1)*G+G);return e.jsxs("div",{children:[e.jsx("div",{className:"flex justify-start my-4",children:e.jsx("input",{type:"text",className:"input input-bordered w-full max-w-xs",placeholder:"Search for a metric",onChange:v=>u(v.target.value)})}),e.jsx("div",{className:"overflow-x-auto",children:e.jsxs("table",{className:"table",children:[e.jsx("thead",{children:e.jsx("tr",{children:xe.map(v=>e.jsx("th",{children:v},v))})}),e.jsx("tbody",{children:w.filter(v=>!h||v.name.name.toLowerCase().includes(h.toLowerCase())).map(v=>e.jsx("tr",{children:xe.map(E=>{const p=v[E];return typeof p=="number"?e.jsx("td",{children:p}):e.jsx("td",{children:e.jsx(ht,{stat:v,metricFieldMap:n})},E)})}))})]})}),e.jsx(ie,{className:"flex justify-center my-8",onNextPage:()=>{const v=Math.min(d+1,g);c(v),a.set("metricsPage",String(v)),r(a)},onPrevPage:()=>{const v=Math.max(d-1,1);c(v),a.set("metricsPage",String(v)),r(a)},currentPage:d,totalPages:g})]})}function xt({runName:s,onAgree:t}){const n=i.useRef(null),a=()=>{n.current!==null&&n.current.value.trim()==="Yes, I agree"?t():alert("Please type 'Yes, I agree' exactly.")},r=s.includes("gpqa")?e.jsx(ft,{}):s.includes("ewok")?e.jsx(pt,{}):null;return e.jsxs("div",{className:"mb-8",children:[r,e.jsxs("p",{className:"mb-4",children:["If you agree to this condition, please type"," ",e.jsx("strong",{children:'"Yes, I agree"'})," in the box below and then click"," ",e.jsx("strong",{children:"Decrypt"}),"."]}),e.jsxs("div",{className:"flex gap-2 mt-2",children:[e.jsx("input",{type:"text",ref:n,className:"input input-bordered",placeholder:'Type "Yes, I agree"'}),e.jsx("button",{onClick:a,className:"btn btn-primary",children:"Decrypt"})]}),e.jsx("hr",{className:"my-4"})]})}function ft(){return e.jsxs("div",{children:[e.jsx("p",{className:"mb-4",children:"The GPQA dataset instances are encrypted by default to comply with the following request:"}),e.jsx("blockquote",{className:"italic border-l-4 border-gray-300 pl-4 text-gray-700 mb-4",children:"“We ask that you do not reveal examples from this dataset in plain text or images online, to minimize the risk of these instances being included in foundation model training corpora.”"})]})}function pt(){return e.jsxs("div",{children:[e.jsx("p",{className:"mb-4",children:"The EWoK dataset instances are encrypted by default to comply with the following request:"}),e.jsx("blockquote",{className:"italic border-l-4 border-gray-300 pl-4 text-gray-700 mb-4",children:"“PLEASE DO NOT distribute any of the EWoK materials or derivatives publicly in plain-text! Any materials should appear in password-protected ZIP files or behind gated authentication mechanisms such as Huggingface datasets.”"})]})}function gt(){const{runName:s}=pe(),[t,n]=i.useState(0),[a,r]=i.useState(),[l,o]=i.useState(),[d,c]=i.useState(),[h,u]=i.useState(),[g,w]=i.useState({}),[v,E]=i.useState({}),[p,R]=i.useState(!1);return i.useEffect(()=>{const x=new AbortController;async function _(){const m=x.signal;if(s===void 0)return()=>x.abort();const j=window.SUITE?window.SUITE:Ws(await Us(m),s);o(j);const[f,M,k]=await Promise.all([Fs(s,m,j),Os(s,m,j),D(m)]);r(f),u(M),E(k.metrics.reduce((y,N)=>(y[N.name]=N,y),{})),w(k.adapter.reduce((y,N)=>(y[N.name]=N,y),{})),c(k.models.find(y=>y.name===(f==null?void 0:f.adapter_spec.model)))}return _(),()=>x.abort()},[s]),a===void 0||h===void 0||s===void 0||l===void 0||v===void 0?e.jsx(P,{}):e.jsxs(e.Fragment,{children:[e.jsx("div",{className:"flex justify-between gap-8 mb-12",children:e.jsxs("div",{children:[e.jsxs("h1",{className:"text-3xl flex items-center",children:[h.name,e.jsx("a",{href:"/#/groups/"+h.name,children:e.jsx(ds,{className:"w-6 h-6 ml-2"})})]}),e.jsx("h3",{className:"text-xl",children:e.jsx(B,{value:h.description})}),e.jsx("h1",{className:"text-3xl mt-2",children:a.adapter_spec.model}),e.jsx("h3",{className:"text-xl",children:e.jsx(B,{value:(d==null?void 0:d.description)||""})}),e.jsx("div",{className:"mt-2 flex gap-2",children:h.tags.map(x=>e.jsx(O,{size:"xs",color:"gray",children:e.jsx("span",{className:"text text-md",children:x})}))})]})}),e.jsxs(W,{children:[e.jsxs("div",{className:"flex justify-between",children:[e.jsx("h3",{className:"text-lg mb-1",children:"Adapter Specification"}),e.jsxs("div",{className:"flex gap-2",children:[e.jsx(is,{className:"w-6 h-6 mr-1 text text-primary"}),e.jsx("a",{className:"link link-primary link-hover",href:Se(a.name,l),download:"true",target:"_blank",children:"Spec JSON"}),e.jsx("a",{className:"link link-primary link-hover",href:Bs(a.name,l),download:"true",target:"_blank",children:"Full JSON"})]})]}),e.jsx("div",{children:e.jsx(te,{className:"grid md:grid-cols-2 lg:grid-cols-3 gap-x-8",children:Object.entries(a.adapter_spec).map(([x,_],m)=>e.jsxs(ne,{className:m<3?"!border-0":"",children:[e.jsx("strong",{className:"mr-1",title:g[x]?g[x].description:void 0,children:`${x}: `}),e.jsx("span",{className:"overflow-x-auto",children:_})]}))})})]}),e.jsx("div",{className:"mt-16 mb-8",children:e.jsxs(re,{children:[e.jsx(J,{size:"lg",active:t===0,onClick:()=>n(0),children:"Instances + Predictions"}),e.jsx(J,{size:"lg",active:t===1,onClick:()=>n(1),children:"All metrics"})]})}),t===0&&Z(s)&&!p&&e.jsx(xt,{runName:s,onAgree:()=>R(!0)}),t===0?e.jsx(dt,{runName:s,suite:l,metricFieldMap:v,userAgreed:p},p?"instances-agreed":"instances-not-agreed"):e.jsx(ut,{runName:s,suite:l,metricFieldMap:v})]})}async function Ae(s,t){try{return await(await fetch(S(`${z()}/groups/${s}.json`),{signal:t})).json()}catch(n){return n instanceof Error&&n.name!=="AbortError"&&console.log(n),[]}}function jt({schema:s,runGroupName:t,numRowsToDisplay:n=-1}){const[a,r]=i.useState(),[l,o]=i.useState(0);return i.useEffect(()=>{const d=new AbortController;async function c(){const h=await Ae(t,d.signal);r(h)}return c(),()=>d.abort()},[s,t]),a===void 0||a.length===0?e.jsx(P,{}):a.length===0?e.jsx("div",{children:"Group currently has no tables."}):e.jsxs("div",{children:[a.length>1?e.jsx(re,{children:a.map((d,c)=>e.jsx(J,{active:c===l,onClick:()=>o(c),children:d.title},c))}):null,e.jsx(le,{schema:s,groupTable:a[l],numRowsToDisplay:n,sortColumnIndex:1},`${t}-${l}`)]})}function Q(){const{groupName:s}=pe(),t=Oe(),[n,a]=i.useState(void 0),[r,l]=i.useState(),[o,d]=i.useState();i.useEffect(()=>{const u=new AbortController;async function g(){const w=D(u.signal),v=Me(u.signal),E=await w;a(E);let p;const R=await v,x={};R.forEach(_=>{x[_.title]=[],_.rows.forEach(m=>{const j=m[0].href.replace("?group=","");p===void 0&&(p=j),x[_.title].push({title:String(m[0].value),name:j})})}),l(x),d(p)}return g(),()=>u.abort()},[]);const c=s||o;if(n===void 0||r===void 0||c===void 0)return e.jsx(P,{});let h;for(const u of n.run_groups)u.name===c&&(h=u);return e.jsxs(e.Fragment,{children:[e.jsxs("div",{className:"flex flex-row",children:[e.jsx("div",{className:"w-3/4",children:h?e.jsx($,{title:"Leaderboard: "+h.display_name,subtitle:h.description,markdown:!0}):e.jsx($,{title:"Leaderboard",subtitle:"The HELM leaderboard shows how the various models perform across different scenarios and metrics.",markdown:!0})}),e.jsx("div",{className:"w-1/4 pt-8",children:e.jsxs("label",{htmlFor:"group",className:"block text-sm font-medium text-gray-700",children:["Select a group:",e.jsx("select",{id:"group",name:"group",onChange:u=>t("/leaderboard/"+u.target.value),className:"mt-1 block w-full pl-3 pr-10 py-2 text-base border-gray-300 focus:outline-none focus:ring focus:border-blue-300 rounded-md",value:c,children:Object.entries(r).map(([u,g])=>e.jsx("optgroup",{label:u,children:g.map(w=>e.jsx("option",{value:w.name,children:w.title},w.name))},u))})]})})]}),e.jsx(jt,{schema:n,runGroupName:c},c)]})}const bt=""+new URL("instruct-flowchart-48854f7c.svg",import.meta.url).href,vt=""+new URL("instruct-graph-0a57d7d2.svg",import.meta.url).href;function wt(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl mt-16 font-bold text-center",children:"HELM Instruct: A Multidimensional Instruction Following Evaluation Framework with Absolute Ratings"}),e.jsxs("div",{className:"flex flex-col sm:flex-row justify-center gap-2 sm:gap-8 md:gap-32 my-8",children:[e.jsx("a",{className:"px-10 btn rounded-md",href:"https://crfm.stanford.edu/2024/02/18/helm-instruct.html",children:"Blog Post"}),e.jsx("a",{className:"px-10 btn rounded-md",href:"https://github.com/stanford-crfm/helm",children:"Github"})]}),e.jsxs("p",{children:["We introduce ",e.jsx("em",{children:"HELM Instruct"}),", a multidimensional evaluation framework for instruction-following LLMs with absolute ratings. The framework takes an instruction, a model, an evaluator, and a criterion to generate a score. In our study, we use HELM Instruct to compare 4 instruction-following models on 7 scenarios based on 4 Human/LM evaluators and 5 criteria. Check out the blog post for more details."]}),e.jsxs("div",{className:"grid my-16 grid-cols-1 md:mx-32 md:grid-cols-2 md:gap-2",children:[e.jsx("img",{src:bt,alt:"Evaluation flowchart",className:"mx-auto block",sizes:"100vw"}),e.jsx("img",{src:vt,alt:"7 scenarios, 4 models, 4 evaluators and 5 criteria",className:"mx-auto block",sizes:"100vw"})]}),e.jsxs("table",{className:"rounded-lg shadow-md table",children:[e.jsx("thead",{children:e.jsxs("tr",{children:[e.jsx("th",{children:"Model"}),e.jsx("th",{children:"Average"}),e.jsx("th",{children:"Helpfulness"}),e.jsx("th",{children:"Understandability"}),e.jsx("th",{children:"Completeness"}),e.jsx("th",{children:"Conciseness"}),e.jsx("th",{children:"Harmlessness"})]})}),e.jsxs("tbody",{children:[e.jsxs("tr",{children:[e.jsx("td",{children:"openai_gpt-4-0314"}),e.jsx("td",{children:"4.63"}),e.jsx("td",{children:"4.42"}),e.jsx("td",{children:"4.85"}),e.jsx("td",{children:"4.50"}),e.jsx("td",{children:"4.42"}),e.jsx("td",{children:"4.95"})]}),e.jsxs("tr",{children:[e.jsx("td",{children:"openai_gpt-3.5-turbo-0613"}),e.jsx("td",{children:"4.60"}),e.jsx("td",{children:"4.34"}),e.jsx("td",{children:"4.86"}),e.jsx("td",{children:"4.42"}),e.jsx("td",{children:"4.41"}),e.jsx("td",{children:"4.97"})]}),e.jsxs("tr",{children:[e.jsx("td",{children:"anthropic_claude-v1.3"}),e.jsx("td",{children:"4.56"}),e.jsx("td",{children:"4.25"}),e.jsx("td",{children:"4.87"}),e.jsx("td",{children:"4.32"}),e.jsx("td",{children:"4.40"}),e.jsx("td",{children:"4.97"})]}),e.jsxs("tr",{children:[e.jsx("td",{children:"cohere_command-xlarge-beta"}),e.jsx("td",{children:"4.31"}),e.jsx("td",{children:"3.90"}),e.jsx("td",{children:"4.73"}),e.jsx("td",{children:"3.88"}),e.jsx("td",{children:"4.31"}),e.jsx("td",{children:"4.72"})]})]})]})]})}function Te({models:s}){return e.jsxs("section",{children:[e.jsxs("h3",{className:"text-3xl",children:[s.length," models"]}),e.jsx("ul",{children:s.map((t,n)=>t.todo?e.jsxs("li",{className:"text-slate-300 mt-1",children:[t.creator_organization," / ",t.display_name]},n):e.jsx(b,{to:"models",children:e.jsxs("li",{className:"text-black mt-1",children:[t.creator_organization," / ",t.display_name]},n)}))})]})}function Ie({runGroups:s}){const t=new Map(s.filter(r=>r.metric_groups!==void 0&&(r.subgroups===void 0||r.subgroups.length===0)).map(r=>[r.name,r])),n=new Set,a=[];return s.forEach(r=>{const l=r.subgroups?r.subgroups:[],o=[];l.forEach(d=>{const c=t.get(d);c&&(o.push(c),n.add(c.name))}),o.length>0&&a.push([r,o])}),e.jsxs("section",{children:[e.jsxs("h3",{className:"text-3xl",children:[n.size," scenarios"]}),e.jsx("ul",{children:a.map(([r,l])=>e.jsxs("li",{className:"my-3",children:[e.jsx(b,{className:"text-black",to:"groups/"+r.name,children:e.jsx("h2",{children:r.display_name})}),e.jsx("ul",{className:"list-disc list-inside",children:l.map(o=>o.todo?e.jsx("li",{className:`${o.todo?"ml-4 text-slate-300":"ml-4"}`,children:o.display_name},o.name):e.jsx(b,{className:"text-black",to:"groups/"+o.name,children:e.jsx("li",{className:`${o.todo?"ml-4 text-slate-300":"ml-4"}`,children:o.display_name},o.name)}))})]},r.name))})]})}const Pe=""+new URL("helmhero-28e90f4d.png",import.meta.url).href;function L({runGroupName:s=void 0,tableIndexToDisplay:t=0,numRowsToDisplay:n=10,sortColumnIndex:a=1}){const[r,l]=i.useState(void 0),[o,d]=i.useState(void 0);return i.useEffect(()=>{const c=new AbortController;async function h(){const u=await D(c.signal);l(u);const g=u.run_groups;if(g.length===0)return;const w=s||g[0].name,v=await Ae(w,c.signal);d(v[t])}return h(),()=>c.abort()},[s,t]),r===void 0||o===void 0?e.jsx(P,{}):e.jsx("div",{className:"rounded-2xl overflow-hidden border-2 bg-white p-1 mx-2 my-0 overflow-x-auto",style:{overflow:"auto",justifyContent:"space-between"},children:e.jsx(le,{schema:r,groupTable:o,numRowsToDisplay:n,sortColumnIndex:a,displayColumnIndexes:[0,1],sortable:!1,miniStyle:!0})})}function yt(){return e.jsxs("div",{className:"flex flex-col px-4 sm:px-6 py-100 sm:py-10 sm:mb-96 md:mb-96 lg:mb-0 xl:mb-0 2xl:mb-0",children:[e.jsx("div",{className:"flex flex-col text-center mb-10 justify-start",children:e.jsx("h1",{className:"text-3xl sm:text-4xl mb-3 sm:mb-4 mx-2 mt-2",children:e.jsx("strong",{children:"A holistic framework for evaluating foundation models."})})}),e.jsxs("div",{className:"flex flex-col md:flex-col lg:flex-row lg:justify-center",children:[e.jsx("div",{className:"w-full lg:w-1/2 flex justify-center mb-4 lg:mb-0 h-full py-10",children:e.jsx("img",{src:Pe,alt:"HELM Hero",className:"object-contain w-96"})}),e.jsxs("div",{className:"py-2 rounded-xl bg-gray-100 h-full",children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-2 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}function fe(){const[s,t]=i.useState(void 0);return i.useEffect(()=>{const n=new AbortController;async function a(){const r=await D(n.signal);t(r)}return a(),()=>n.abort()},[]),s?e.jsxs(e.Fragment,{children:[e.jsx(yt,{}),e.jsx("div",{className:"mx-auto text-lg px-16",children:e.jsx("div",{className:"container mx-auto",children:e.jsxs("div",{className:"grid grid-cols-1 sm:grid-cols-2 gap-8",children:[e.jsx(Te,{models:s.models}),e.jsx(Ie,{runGroups:s.run_groups})]})})})]}):null}function Nt(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl mt-16 my-8 font-bold text-center",children:"Massive Multitask Language Understanding (MMLU) on HELM"}),e.jsxs("div",{className:"flex flex-col lg:flex-row items-center gap-8",children:[e.jsxs("div",{className:"flex-1 text-xl",children:[e.jsxs("p",{children:[e.jsx("strong",{children:"Massive Multitask Language Understanding (MMLU)"})," ",e.jsx("a",{href:"https://arxiv.org/pdf/2009.03300.pdf",className:"link",children:"(Hendrycks et al, 2020)"})," ","is a multiple-choice question answering test that covers 57 tasks including elementary mathematics, US history, computer science, law, and more. We publish evaluation results from evaluating various models on MMLU using HELM. Our evaluation results include the following:"]}),e.jsxs("ul",{className:"my-2 list-disc list-inside",children:[e.jsx("li",{children:"Simple, standardized prompts"}),e.jsx("li",{children:"Accuracy breakdown for each of the 57 subjects"}),e.jsx("li",{children:"Full transparency of all raw prompts and predictions"})]}),e.jsxs("div",{className:"flex flex-row justify-center mt-4",children:[e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"https://crfm.stanford.edu/2024/05/01/helm-mmlu.html",children:"Blog Post"}),e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"#/leaderboard",children:"Full Leaderboard"})]})]}),e.jsx("div",{className:"flex-1",children:e.jsx(L,{})})]})]})}const Et=""+new URL("air-overview-d2e6c49f.png",import.meta.url).href;function Mt(){const s={fontVariant:"small-caps",fontWeight:"bold"},t=e.jsx("span",{style:s,children:"AIR-Bench 2024"});return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl mt-16 my-8 font-bold text-center",children:t}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsx("img",{src:Et,alt:"AIR 2024 Categories",className:"mx-auto my-4 block w-3/4",sizes:"100vw"}),e.jsxs("p",{children:["We introduce ",t,", the first AI safety benchmark aligned with emerging government regulations and company policies, following the regulation-based safety categories grounded in our AI Risks study, AIR 2024. AIR 2024 decomposes 8 government regulations and 16 company policies into a four-tiered safety taxonomy with 314 granular risk categories in the lowest tier. ",t," contains 5,694 diverse prompts spanning these categories, with manual curation and human auditing to ensure quality. We evaluate leading language models on ",t,", uncovering insights into their alignment with specified safety concerns. By bridging the gap between public benchmarks and practical AI risks, ",t," ","provides a foundation for assessing model safety across jurisdictions, fostering the development of safer and more responsible AI systems."]}),e.jsxs("div",{className:"flex flex-row justify-center mt-4",children:[e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"https://arxiv.org/abs/2407.17436",children:"Paper"}),e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"#/leaderboard",children:"Full Leaderboard"})]})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}function St(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl my-8 font-bold text-center",children:"ThaiExam"}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsxs("p",{children:["In collaboration with"," ",e.jsx("a",{href:"https://www.scbx.com/",className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",children:"SCBX"})," ","and"," ",e.jsx("a",{href:"https://www.scb10x.com/",className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",children:"SCB 10X"}),", we introduce the ThaiExam HELM leaderboard. ThaiExam is a Thai language benchmark based on examinations for high school students and investment professionals in Thailand. The ThaiExam leaderboard is the first public leaderboard for large language models on Thai language scenarios, and features evaluations of leading language models. Like all other HELM leaderboards, the ThaiExam leaderboard provides full prompt-level transparency, and the results can be fully reproduced using the HELM framework. We hope that this leaderboard will encourage further work in multilingual language model evaluation."]}),e.jsx("div",{className:"flex flex-row justify-center my-4",children:e.jsx(b,{to:"leaderboard",className:"px-10 btn rounded-md mx-4",children:"Full Leaderboard"})})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}function Lt(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl my-8 font-bold text-center",children:"HELM Finance"}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsxs("p",{children:["In collaboration with"," ",e.jsx("a",{href:"https://www.wellsfargo.com/",className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",children:"Wells Fargo"}),", we introduce the ",e.jsx("span",{className:"font-bold",children:"HELM Finance"})," ","leaderboard for ecologically-valid evaluations of leading language models in the financial domain. The leaderboard evaluates the ability of language models to perform tasks from financial professions on publicly financial documents across a range of scenarios."]}),e.jsx("div",{className:"flex flex-row justify-center my-4",children:e.jsx(b,{to:"leaderboard",className:"px-10 btn rounded-md mx-4",children:"Full Leaderboard"})})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}const kt=""+new URL("heim-logo-3e5e3aa4.png",import.meta.url).href;function Rt({metricFieldMap:s,metricGroups:t}){const n=new Set,a=[];return t.forEach(r=>{const l=[];r.metrics.forEach(o=>{const d=s[o.name];d&&(l.push(d),n.add(d.name))}),l.length>0&&a.push([r,l])}),e.jsxs("section",{children:[e.jsxs("h3",{className:"text-3xl",children:[n.size," metrics"]}),e.jsx("ul",{children:a.map(([r,l])=>e.jsxs("li",{className:"my-3",children:[e.jsx("h4",{children:r.display_name}),e.jsx("ul",{className:"list-disc list-inside",children:l.map(o=>e.jsx("li",{className:"ml-4",children:o.display_name},o.name))})]},r.name))})]})}function _t(){const[s,t]=i.useState(void 0);i.useEffect(()=>{const a=new AbortController;async function r(){const l=await D(a.signal);t(l)}return r(),()=>a.abort()},[]);const n=s?s.metrics.reduce((a,r)=>(a[r.name]=r,a),{}):void 0;return e.jsxs("div",{className:"container mx-auto px-16 text-base",children:[e.jsx("div",{className:"container max-w-screen-lg mx-auto",children:e.jsx("img",{className:"mx-auto w-96",src:kt,alt:"HEIM Logo"})}),e.jsx("h1",{className:"text-3xl my-8 font-bold text-center",children:"Holistic Evaluation of Text-To-Image Models"}),e.jsxs("div",{className:"flex flex-col sm:flex-row justify-center gap-2 sm:gap-8 md:gap-32 my-8",children:[e.jsx("a",{className:"px-10 btn rounded-md",href:"https://arxiv.org/abs/2311.04287",children:"Paper"}),e.jsx("a",{className:"px-10 btn rounded-md",href:"https://github.com/stanford-crfm/helm",children:"Github"})]}),e.jsxs("p",{className:"my-2",children:["Significant effort has recently been made in developing text-to-image generation models, which take textual prompts as input and generate images. As these models are widely used in real-world applications, there is an urgent need to comprehensively understand their capabilities and risks. However, existing evaluations primarily focus on image-text alignment and image quality. To address this limitation, we introduce a new benchmark,"," ",e.jsx("strong",{children:"Holistic Evaluation of Text-To-Image Models (HEIM)"}),"."]}),e.jsx("p",{className:"my-2",children:"We identify 12 different aspects that are important in real-world model deployment, including:"}),e.jsxs("ul",{className:"my-2 list-disc list-inside unreset",children:[e.jsx("li",{children:"image-text alignment"}),e.jsx("li",{children:"image quality"}),e.jsx("li",{children:"aesthetics"}),e.jsx("li",{children:"originality"}),e.jsx("li",{children:"reasoning"}),e.jsx("li",{children:"knowledge"}),e.jsx("li",{children:"bias"}),e.jsx("li",{children:"toxicity"}),e.jsx("li",{children:"fairness"}),e.jsx("li",{children:"robustness"}),e.jsx("li",{children:"multilinguality"}),e.jsx("li",{children:"efficiency"})]}),e.jsx("p",{className:"my-2",children:"By curating scenarios encompassing these aspects, we evaluate state-of-the-art text-to-image models using this benchmark. Unlike previous evaluations that focused on alignment and quality, HEIM significantly improves coverage by evaluating all models across all aspects. Our results reveal that no single model excels in all aspects, with different models demonstrating strengths in different aspects."}),e.jsx("p",{className:"my-2",children:"For full transparency, this website contains all the prompts, generated images and the results for the automated and human evaluation metrics."}),e.jsx("p",{className:"my-2",children:"Inspired by HELM, we decompose the model evaluation into four key components: aspect, scenario, adaptation, and metric:"}),e.jsx("div",{className:"container max-w-screen-lg mx-auto my-8",children:e.jsx("img",{src:"https://crfm.stanford.edu/heim/latest/images/heim-main.png",alt:"HEIM scenarios, prompts, images and metrics"})}),s&&n?e.jsxs("div",{className:"grid grid-cols-1 md:grid-cols-3 gap-8",children:[e.jsx(Te,{models:s.models}),e.jsx(Ie,{runGroups:s.run_groups}),e.jsx(Rt,{metricFieldMap:n,metricGroups:s.metric_groups})]}):null]})}const Ct=""+new URL("vhelm-framework-a1ca3f3f.png",import.meta.url).href,At=""+new URL("vhelm-model-8afb7616.png",import.meta.url).href,Tt=""+new URL("vhelm-aspects-1437d673.png",import.meta.url).href;function It(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl mt-16 my-8 font-bold text-center",children:"Holistic Evaluation of Vision-Language Models"}),e.jsxs("div",{className:"flex flex-col sm:flex-row justify-center gap-2 sm:gap-8 md:gap-32 my-8",children:[e.jsx("a",{className:"px-10 btn rounded-md",href:"https://arxiv.org/abs/2410.07112",children:"Paper"}),e.jsx("a",{className:"px-10 btn rounded-md",href:"#/leaderboard",children:"Leaderboard"}),e.jsx("a",{className:"px-10 btn rounded-md",href:"https://github.com/stanford-crfm/helm",children:"Github"})]}),e.jsxs("p",{className:"my-4",children:["Current benchmarks for assessing vision-language models (VLMs) often focus on their perception or problem-solving capabilities and neglect other critical aspects such as fairness, multilinguality, or toxicity. Furthermore, they differ in their evaluation procedures and the scope of the evaluation, making it difficult to compare models. To address these issues, we extend the HELM framework to VLMs to present the Holistic Evaluation of Vision Language Models (VHELM). To address these issues, we introduce VHELM, built on HELM for language models. VHELM aggregates various datasets to cover one or more of the 9 aspects:"," ",e.jsx("b",{children:"visual perception"}),", ",e.jsx("b",{children:"bias"}),", ",e.jsx("b",{children:"fairness"}),", ",e.jsx("b",{children:"knowledge"}),", ",e.jsx("b",{children:"multilinguality"}),", ",e.jsx("b",{children:"reasoning"}),", ",e.jsx("b",{children:"robustness"}),","," ",e.jsx("b",{children:"safety"}),", and ",e.jsx("b",{children:"toxicity"}),". In doing so, we produce a comprehensive, multi-dimensional view of the capabilities of the VLMs across these important factors. In addition, we standardize the standard inference parameters, methods of prompting, and evaluation metrics to enable fair comparisons across models. Our framework is designed to be lightweight and automatic so that evaluation runs are cheap and fast. For transparency, we release the raw model generations and complete results on this website."]}),e.jsx("p",{className:"my-4 font-bold",children:"VHELM is intended to be a living benchmark. We hope to continue adding new datasets, models and metrics over time, so please stay tuned!"}),e.jsxs("div",{className:"my-16 flex flex-col lg:flex-row items-center gap-8",children:[e.jsxs("div",{className:"flex-1 text-xl",children:[e.jsx("img",{src:At,alt:"A vision-lanuage model (VLM) takes in an image and a text prompt and generates text.",className:""}),e.jsx("img",{src:Ct,alt:"An example of an evaluation for an Aspect (Knowledge) - a Scenario (MMMU) undergoes Adaptation (multimodal multiple choice) for a Model (GPT-4 Omni), then Metrics (Exact match) are computed",className:""})]}),e.jsxs("div",{className:"flex-1",children:[e.jsx(L,{}),e.jsx(b,{to:"leaderboard",className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})]})]}),e.jsx("div",{className:"container max-w-screen-lg mx-auto my-8",children:e.jsx("img",{src:Tt,alt:"An example of each aspect in VHELM: Visual Perception, Bias, Fairness, Knowledge, Multilinguality, Reasoning, Robustness, Toxicity Mitigation and Safety. ",className:""})})]})}function Pt(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl my-8 font-bold text-center",children:"HELM Call Center Leaderboard"}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsxs("p",{className:"my-4",children:["LLMs show great potential for applications for the call center, yet there is a lack of domain-specific and ecologically-valid evaluations in this domain. To address this, we introduce the"," ",e.jsx("strong",{className:"font-bold",children:"HELM Call Center leaderboard"}),". The HELM Call Center leaderboard evaluates leading LLMs on a summarization task over a dataset of real helpdesk call transcripts provided by Accenture. The quality of the summaries is evaluated using LLM-as-judge with an ensemble of 3 models. We hope that this leaderboard provides some initial insights into the potential of LLMs in this domain."]}),e.jsxs("p",{className:"my-4",children:["This leaderboard was produced through research collaboration with"," ",e.jsx("a",{className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",href:"https://www.accenture.com/",children:"Accenture"}),", and was funded by the"," ",e.jsx("a",{className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",href:"https://hai.stanford.edu/corporate-affiliate-program",children:"HAI Corporate Affiliate Program"}),"."]})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}function $t(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl my-8 font-bold text-center",children:"Chinese Language Models EVAluation Platform (CLEVA)"}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsxs("p",{children:["In collaboration with the"," ",e.jsx("a",{href:"https://lwwangcse.github.io/",className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",children:"LaVi Lab"})," ","team from"," ",e.jsx("a",{href:"https://www.cuhk.edu.hk/",className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",children:"The Chinese University of Hong Kong (CUHK)"}),", we introduce the"," ",e.jsx("strong",{className:"font-bold",children:"Chinese Language Models EVAluation Platform (CLEVA)"})," ","leaderboard on HELM. CLEVA is a comprehensive Chinese-language benchmark for holistic evaluation of Chinese-language LLMs, and employs a standardized workflow to assess LLMs' performance across various dimensions."]}),e.jsxs("div",{className:"flex flex-row justify-center mt-4",children:[e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"https://arxiv.org/abs/2308.04813",children:"Paper"}),e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"#/leaderboard",children:"Full Leaderboard"})]})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}function Ht(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl my-8 font-bold text-center",children:"The Mighty ToRR: A Benchmark for Table Reasoning and Robustness"}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsxs("p",{className:"mb-4 italic",children:["This leaderboard is a collaboration with"," ",e.jsx("a",{href:"https://research.ibm.com/",className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",children:"IBM Research"}),"."]}),e.jsxs("p",{className:"my-4",children:["Despite its real-world significance, model performance on tabular data remains underexplored, leaving uncertainty about which model to rely on and which prompt configuration to adopt. To address this gap, we create ",e.jsx("strong",{className:"font-bold",children:"ToRR"}),", a benchmark for Table Reasoning and Robustness, a varied benchmark that measures model performance and robustness on table-related tasks. The benchmark includes 10 datasets that cover different types of table reasoning capabilities across varied domains. ToRR goes beyond model performance rankings, and is designed to reflect whether models can handle tabular data consistently and robustly, across a variety of common table representation formats. We present a leaderboard as well as comprehensive analyses of the results of leading models over ToRR. Our results reveal a striking pattern of brittle model behavior, where even strong models are unable to perform robustly on tabular data tasks. Although no specific table format leads to consistently better performance, we show that testing over multiple formats is crucial for reliably estimating model capabilities. Moreover, we show that the reliability boost from testing multiple prompts can be equivalent to adding more test examples. Overall, our findings show that reasoning over table tasks remains a significant challenge."]}),e.jsxs("div",{className:"flex flex-row justify-center mt-4",children:[e.jsx("a",{className:"px-10 btn rounded-md mx-4 hidden",href:"#",children:"Paper"}),e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"#/leaderboard",children:"Full Leaderboard"})]})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}const Dt=({id:s,title:t,text:n})=>((t==="Classic"||t==="Lite"||t==="Instruct")&&(t="HELM "+t),e.jsx("div",{className:"max-w-sm rounded overflow-hidden bg-gray-100 hover:scale-105 transition-transform duration-300",children:e.jsx("a",{href:ae(void 0,s),children:e.jsxs("div",{className:"px-6 py-4",children:[e.jsxs("div",{className:"font-bold text-xl mb-2",children:[e.jsx("div",{className:"py-3",children:e.jsx("svg",{fill:"#000000",width:"20px",height:"20px",viewBox:"0 0 24 24",xmlns:"http://www.w3.org/2000/svg",children:e.jsx("path",{d:"M22,7H16.333V4a1,1,0,0,0-1-1H8.667a1,1,0,0,0-1,1v7H2a1,1,0,0,0-1,1v8a1,1,0,0,0,1,1H22a1,1,0,0,0,1-1V8A1,1,0,0,0,22,7ZM7.667,19H3V13H7.667Zm6.666,0H9.667V5h4.666ZM21,19H16.333V9H21Z"})})}),t+" →"]}),e.jsx("p",{className:"text-gray-700 text-base",children:n})]})})}));function Ot(){const[s,t]=i.useState();return i.useEffect(()=>{fetch("https://raw.githubusercontent.com/stanford-crfm/helm/main/helm-frontend/project_metadata.json").then(n=>n.json()).then(n=>{t(n)}).catch(n=>{console.error("Error fetching JSON:",n)})},[]),e.jsx("div",{className:"p-10 mb-20",children:e.jsx("div",{className:"grid grid-cols-2 lg:grid-cols-3 gap-4",children:s&&s.map((n,a)=>n.id==="home"?null:e.jsx(Dt,{id:n.id,title:n.title,text:n.description},a))})})}function Ft(){return e.jsxs("div",{className:"flex flex-col md:flex-row px-6 py-32",children:[e.jsxs("div",{className:"flex-1 p-4 flex flex-col justify-center",children:[e.jsx("div",{className:"flex justify-start",children:e.jsxs("div",{children:[e.jsx("h1",{className:"text-4xl mb-4 mx-4 mt-2",children:e.jsx("strong",{children:"A reproducible and transparent framework for evaluating foundation models."})}),e.jsx("h3",{className:`text-xl
+             mb-4 mx-4 mt-2`,children:"Find leaderboards with many scenarios, metrics, and models with support for multimodality and model-graded evaluation."})]})}),e.jsxs("div",{className:"flex flex-col md:flex-row justify-start mt-6 ml-4",children:[e.jsx("button",{className:"px-6 btn btn-grey rounded-md mb-4 md:mb-0",onClick:()=>window.scrollTo({top:760,behavior:"smooth"}),children:e.jsx("div",{children:"Leaderboards ↓"})}),e.jsx("button",{className:"px-6 btn btn-grey rounded-md md:ml-4",children:e.jsx("a",{href:"https://github.com/stanford-crfm/helm",children:"Github"})})]})]}),e.jsx("div",{className:"mx-4 mt-6 md:mt-0 md:w-1/3",children:e.jsx("img",{src:Pe,alt:"HELM Hero",className:"object-cover w-full h-full"})})]})}function Bt(){const[s,t]=i.useState(void 0);return i.useEffect(()=>{const n=new AbortController;async function a(){const r=await D(n.signal);t(r)}return a(),()=>n.abort()},[]),s?e.jsxs(e.Fragment,{children:[e.jsx(Ft,{}),e.jsx("div",{className:"container py-5 mx-auto text-lg",children:e.jsx("div",{className:"flex flex-col sm:flex-row justify-center mb-10 flex sm:gap-8 md:gap-32",children:e.jsx("h1",{className:"text-4xl mx-4 ",children:e.jsx("strong",{children:"HELM Leaderboards"})})})}),e.jsx(Ot,{})]}):null}const Ut=""+new URL("overview-74aea3d8.png",import.meta.url).href,Wt=""+new URL("process-flow-bd2eba96.png",import.meta.url).href;function zt(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl mt-16 my-8 font-bold text-center",children:"Image2Struct: A Benchmark for Evaluating Vision-Language Models in Extracting Structured Information from Images"}),e.jsxs("div",{className:"flex flex-col sm:flex-row justify-center gap-2 sm:gap-2 md:gap-8 my-8",children:[e.jsx("a",{className:"px-10 btn rounded-md",href:"https://arxiv.org/abs/2410.22456",children:"Paper"}),e.jsx("a",{className:"px-10 btn rounded-md",href:"https://github.com/stanford-crfm/helm",children:"Github"}),e.jsx("a",{className:"px-10 btn rounded-md",href:"https://huggingface.co/datasets/stanford-crfm/i2s-latex",children:"Latex dataset"}),e.jsx("a",{className:"px-5 btn rounded-md",href:"https://huggingface.co/datasets/stanford-crfm/i2s-webpage",children:"Webpage dataset"}),e.jsx("a",{className:"px-10 btn rounded-md",href:"https://huggingface.co/datasets/stanford-crfm/i2s-musicsheet",children:"Music sheet dataset"})]}),e.jsxs("div",{className:"flex flex-col lg:flex-row items-center gap-8",children:[e.jsxs("div",{className:"flex-1 text-xl",children:[e.jsxs("p",{children:[e.jsx("strong",{children:"Image2Struct"})," is a benchmark for evaluating vision-language models in practical tasks of extracting structured information from images."]}),e.jsx("br",{}),e.jsx("p",{children:"In our tasks, VLMs are prompted to generate the underlying structured information (i.e., code) from an input image. The code can be compiled, and the output image is evaluated against the input image to produce a score. This round-trip evaluation allows us to quantitatively evaluate VLMs on complex tasks with multiple correct answers. We create a pipeline that downloads fresh, user-submitted data from active online communities upon execution, evaluates the VLMs shortly, and produces a leaderboard."}),e.jsx("br",{}),e.jsx("img",{src:Ut,alt:"Evaluation flowchart",className:"mx-auto block w-full",sizes:"100vw"}),e.jsx("br",{}),e.jsx("p",{children:"We introduce 3 tasks:"}),e.jsxs("ul",{className:"my-2 list-disc list-inside",children:[e.jsx("li",{children:"LaTex: equations, tables, plots and algorithms from ArXiV papers."}),e.jsx("li",{children:"Webpages: pages from GitHub written in HTML, CSS and Javascript. ..."}),e.jsx("li",{children:"Music sheets: crops of measures from IMSLP music sheets."})]}),e.jsx("div",{className:"flex flex-row justify-center mt-8",children:e.jsx("a",{className:"px-10 btn rounded-md",href:"#/leaderboard",children:"Full Leaderboard"})})]}),e.jsx("div",{className:"flex-1",children:e.jsx(L,{numRowsToDisplay:12})})]}),e.jsx("br",{}),e.jsxs("div",{className:"flex flex-col lg:flex-row items-center gap-8",children:[e.jsxs("div",{className:"flex-1 text-xl",children:[e.jsx("p",{children:"We provide an automated process for collecting new fresh data from online communities, evaluating the models and producing a leaderboard. The pipeline is designed to be executed on a regular basis to keep the leaderboard up-to-date."}),e.jsx("br",{}),e.jsxs("p",{children:["In addition to the automated data collection, we also provide a"," ",e.jsx("i",{children:"wild"})," subset for the LaTeX and webpage tasks that are collected from Wikipedia and various popular websites. These instances do not have a corresponding code, and the evaluation is done by our proposed metric: block EMD (Earth Mover Distance)."]})]}),e.jsx("div",{className:"flex-1",children:e.jsx("img",{src:Wt,alt:"7 scenarios, 4 models, 4 evaluators and 5 criteria",className:"mx-auto block w-full",sizes:"200vw"})})]})]})}function qt(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl my-8 font-bold text-center",children:"Elements of World Knowledge (EWoK)"}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsxs("p",{children:["We present the"," ",e.jsx("a",{className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",href:"https://arxiv.org/abs/2405.09605",children:"Elements of World Knowledge (EWoK)"})," ","leaderboard in collaboration with the EWoK team. EWoK is a benchmark for evaluating world modeling in language models by testing their ability to use knowledge of a concept to match a target text with a plausible/implausible context. EWoK targets specific concepts from multiple knowledge domains known to be vital for world modeling in humans, including social interactions and spatial relations."]}),e.jsxs("div",{className:"flex flex-row justify-center mt-4",children:[e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"https://arxiv.org/abs/2405.09605",children:"Paper"}),e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"#/leaderboard",children:"Full Leaderboard"})]})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}const Gt=""+new URL("medhelm-overview-3ddfcd65.png",import.meta.url).href;function Vt(){const s=e.jsx("strong",{className:"font-bold",children:"MedHELM"});return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl my-4 font-bold text-center",children:"MedHELM"}),e.jsx("p",{className:"text-xl my-4 italic text-center",children:"Medical and AI experts build a benchmark for evaluation of LLMs grounded in real-world healthcare needs."}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsxs("p",{className:"my-4",children:["We introduce ",s,", a comprehensive healthcare benchmark to evaluate language models on real-world clinical tasks using real electronic health records. Building on the HELM framework,"," ",s," comprises a structured taxonomy with 5 categories, 22 subcategories, and 121 distinct clinical tasks as well as 31 diverse datasets (12 private, 6 gated-access, and 13 public). The datasets represent a spectrum of healthcare scenarios, from diagnostic decision-making to patient communication, providing a more nuanced and clinically relevant assessment of AI capabilities in healthcare settings."]}),e.jsx("img",{src:Gt,alt:"MedHELM Task Categories",className:"mx-auto my-4 block w-7/8",sizes:"100vw"}),e.jsxs("p",{className:"my-4",children:[s," establishes a foundation for testing and evaluation of the real-world applicability of language models in healthcare. It is made possible by a unique collaboration between the Center for Research on Foundation Models, Technology and Digital Solutions at Stanford Healthcare, and Microsoft Healthcare and Life Sciences in partnership with faculty in the Departments of Medicine, Computer Science, Anesthesiology, Dermatology, Pediatrics and Biomedical Data Science as well as trainees from the MCiM program at the Clinical Excellence Research Center. The effort is coordinated by the Center for Biomedical Informatics Research."]}),e.jsxs("div",{className:"flex flex-row justify-center mt-4",children:[e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"https://hai.stanford.edu/news/holistic-evaluation-of-large-language-models-for-medical-applications",children:"Blog Post"}),e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"#/leaderboard",children:"Full Leaderboard"})]})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}const Jt=""+new URL("helm-safety-2907a7b6.png",import.meta.url).href;function Kt(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl mt-16 my-8 font-bold text-center",children:"HELM Safety"}),e.jsxs("div",{className:"flex flex-col lg:flex-row items-center gap-8",children:[e.jsxs("div",{className:"flex-1 text-xl",children:[e.jsx("img",{src:Jt,alt:"Logo",className:"mx-auto p-0 block",style:{width:"300px"}}),e.jsx("p",{children:"Language models demonstrate powerful capabilities and pose significant risks. Given their widespread deployment, standardized public benchmarking of such models is vital. While language models are routinely evaluated on standard capability benchmarks, comparable standardization for benchmarking safety risks lags behind. To address this gap, we introduce HELM-Safety as a collection of 5 safety benchmarks that span 6 risk categories (e.g. violence, fraud, discrimination, sexual, harassment, deception). We present evaluation results for recent leading open weights and closed models."}),e.jsxs("div",{className:"flex flex-row justify-center mt-4",children:[e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"https://crfm.stanford.edu/2024/11/08/helm-safety.html",children:"Blog Post"}),e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"#/leaderboard",children:"Full Leaderboard"})]})]}),e.jsx("div",{className:"flex-1",children:e.jsx(L,{})})]})]})}function Zt(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl my-8 font-bold text-center",children:"HELM Capabilities"}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsx("p",{children:"HELM Capabilities is a new leaderboard for benchmarking the capabilities of foundation models, featuring 6 challenging scenarios."}),e.jsxs("div",{className:"flex flex-row justify-center my-4",children:[e.jsx(b,{to:"#",className:"px-10 btn rounded-md mx-4",children:"Blog Post"}),e.jsx(b,{to:"leaderboard",className:"px-10 btn rounded-md mx-4",children:"Leaderboard"})]})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}function Xt(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl my-8 font-bold text-center",children:"MMLU-Winogrande-Afr: Clinical MMLU and Winogrande in 11 low-resource African languages"}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsxs("p",{className:"mb-4 italic",children:["This leaderboard is a collaboration with"," ",e.jsx("a",{href:"https://ghamut.com/",className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",children:"Ghamut Corporation"})," ","and the"," ",e.jsx("a",{href:"https://www.gatesfoundation.org/",className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",children:"Gates Foundation"}),"."]}),e.jsx("p",{className:"my-4",children:"Large Language Models (LLMs) have shown remarkable performance across various tasks, yet significant disparities remain for non-English languages, and especially native African languages. This paper addresses these disparities by creating approximately 1 million human-translated words of new benchmark data in 8 low-resource African languages, covering a population of over 160 million speakers of: Amharic, Bambara, Igbo, Sepedi (Northern Sotho), Shona, Sesotho (Southern Sotho), Setswana, and Tsonga. Our benchmarks are translations of Winogrande and three sections of MMLU: college medicine, clinical knowledge, and virology. Using the translated benchmarks, we report previously unknown performance gaps between state-of-the-art (SOTA) LLMs in English and African languages. The publicly available benchmarks, translations, and code from this study support further research and development aimed at creating more inclusive and effective language technologies."}),e.jsxs("div",{className:"flex flex-row justify-center mt-4",children:[e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"https://arxiv.org/abs/2412.12417",children:"Paper"}),e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"#/leaderboard",children:"Full Leaderboard"})]})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}function Qt(){return e.jsxs("div",{className:"container mx-auto px-16",children:[e.jsx("h1",{className:"text-3xl my-8 font-bold text-center",children:"SEA-HELM: Southeast Asian Holistic Evaluation of Language Models"}),e.jsxs("div",{className:"flex flex-col lg:flex-row gap-8",children:[e.jsxs("div",{className:"flex-1 text-l",children:[e.jsxs("p",{className:"mb-4 italic",children:["This leaderboard is a collaboration with"," ",e.jsx("a",{href:"https://aisingapore.org/",className:"font-bold underline text-blue-600 hover:text-blue-800 visited:text-purple-600",children:"AI Singapore"}),"."]}),e.jsxs("p",{className:"my-4",children:["With the rapid emergence of novel capabilities in Large Language Models (LLMs), the need for rigorous multilingual and multicultural benchmarks that are integrated has become more pronounced. Though existing LLM benchmarks are capable of evaluating specific capabilities of LLMs in English as well as in various mid- to low-resource languages, including those in the Southeast Asian (SEA) region, a comprehensive and authentic evaluation suite for the SEA languages has not been developed thus far. Here, we present"," ",e.jsx("strong",{className:"font-bold",children:"SEA-HELM"}),", a holistic linguistic and cultural LLM evaluation suite that emphasizes SEA languages, comprising five core pillars: (1) NLP Classics, (2) LLM-specifics, (3) SEA Linguistics, (4) SEA Culture, (5) Safety. SEA-HELM currently supports Filipino, Indonesian, Tamil, Thai, and Vietnamese. We also introduce the SEA-HELM leaderboard, which allows users to understand models' multilingual and multicultural performance in a systematic and user-friendly manner."]}),e.jsxs("div",{className:"flex flex-row justify-center mt-4",children:[e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"https://arxiv.org/abs/2502.14301",children:"Paper"}),e.jsx("a",{className:"px-10 btn rounded-md mx-4",href:"#/leaderboard",children:"Full Leaderboard"})]})]}),e.jsxs("div",{className:"py-2 pb-6 rounded-3xl bg-gray-100 h-full",style:{maxWidth:"100%"},children:[e.jsx(L,{}),e.jsx("div",{className:"flex justify-end",children:e.jsx(b,{to:"leaderboard",children:e.jsx("button",{className:"px-4 mx-3 mt-1 btn bg-white rounded-md",children:e.jsx("span",{children:"See More"})})})})]})]})]})}function Yt(){return window.PROJECT_ID==="lite"?e.jsx(fe,{}):window.PROJECT_ID==="instruct"?e.jsx(wt,{}):window.PROJECT_ID==="image2struct"?e.jsx(zt,{}):window.PROJECT_ID==="heim"?e.jsx(_t,{}):window.PROJECT_ID==="mmlu"?e.jsx(Nt,{}):window.PROJECT_ID==="vhelm"?e.jsx(It,{}):window.PROJECT_ID==="air-bench"?e.jsx(Mt,{}):window.PROJECT_ID==="thaiexam"?e.jsx(St,{}):window.PROJECT_ID==="finance"?e.jsx(Lt,{}):window.PROJECT_ID==="call-center"?e.jsx(Pt,{}):window.PROJECT_ID==="cleva"?e.jsx($t,{}):window.PROJECT_ID==="torr"?e.jsx(Ht,{}):window.PROJECT_ID==="ewok"?e.jsx(qt,{}):window.PROJECT_ID==="medhelm"?e.jsx(Vt,{}):window.PROJECT_ID==="safety"?e.jsx(Kt,{}):window.PROJECT_ID==="capabilities"?e.jsx(Zt,{}):window.PROJECT_ID==="mmlu-winogrande-afr"?e.jsx(Xt,{}):window.PROJECT_ID==="sea-helm"?e.jsx(Qt,{}):window.PROJECT_ID==="home"?e.jsx(Bt,{}):e.jsx(fe,{})}function en(){return e.jsx(Fe,{children:e.jsx(Be,{children:e.jsxs(T,{path:"/",element:e.jsx(Ss,{}),children:[e.jsx(T,{index:!0,element:e.jsx(Yt,{})}),e.jsx(T,{path:"leaderboard",element:e.jsx(Q,{})}),e.jsx(T,{path:"leaderboard/:groupName",element:e.jsx(Q,{})}),e.jsx(T,{path:"models",element:e.jsx(Cs,{})}),e.jsx(T,{path:"scenarios",element:e.jsx(As,{})}),e.jsx(T,{path:"groups",element:e.jsx($s,{})}),e.jsx(T,{path:"groups/:groupName",element:e.jsx(Q,{})}),e.jsx(T,{path:"runs",element:e.jsx(Ds,{})}),e.jsx(T,{path:"runs/:runName",element:e.jsx(gt,{})})]})})})}ee.createRoot(document.getElementById("root")).render(e.jsx(Ue.StrictMode,{children:e.jsx(en,{})}));