Spaces:

fair-forward
/

evals-for-every-language

Running

App Files Files Community

David Pomerenke commited on Feb 16

Commit

ada20e2

2 Parent(s): 790c5f2 c527cda

Merge branch 'feature-observable'

Browse files

Files changed (17) hide show

.gitignore +6 -0
README.md +1 -1
index.html +0 -254
observablehq.config.js +34 -0
package-lock.json +0 -0
package.json +22 -0
results.json +0 -0
src/.gitignore +1 -0
src/compare-ai-models.md +65 -0
src/compare-languages.md +20 -0
src/components/language-chart.js +68 -0
data.txt → src/data/data.txt +1 -1
languagebench.py → src/data/languagebench.json.py +17 -16
languages.rq → src/data/languages.rq +0 -0
languages.tsv → src/data/languages.tsv +0 -0
src/index.md +53 -0
src/methodology.md +12 -0

.gitignore CHANGED Viewed

@@ -4,6 +4,12 @@ ScriptCodes.csv
 .cache
 .env
 # Python-generated files
 __pycache__/
 *.py[oc]

 .cache
 .env
+# Observable
+.DS_Store
+/dist/
+node_modules/
+yarn-error.log
 # Python-generated files
 __pycache__/
 *.py[oc]

README.md CHANGED Viewed

@@ -1,6 +1,6 @@
 ---
 title: AI Language Monitor
-emoji: 👀
 colorFrom: purple
 colorTo: pink
 sdk: static

 ---
 title: AI Language Monitor
+emoji: 🌍
 colorFrom: purple
 colorTo: pink
 sdk: static

index.html DELETED Viewed

@@ -1,254 +0,0 @@
-<!DOCTYPE html>
-<html>
-<head>
-    <title>AI Language Monitor</title>
-    <script src="https://cdn.tailwindcss.com"></script>
-    <style>
-        body {
-            margin: 0 auto;
-            padding: 20px;
-            font-family: sans-serif;
-        }
-        .language-header {
-            margin-bottom: 10px;
-        }
-        .speaker-count {
-            font-size: 0.8em;
-            color: #666;
-            font-weight: normal;
-            margin: 0;
-        }
-    </style>
-    <link rel="icon"
-        href="data:image/svg+xml,<svg xmlns=%22http://www.w3.org/2000/svg%22 viewBox=%220 0 100 100%22><text y=%22.9em%22 font-size=%2290%22 fill=%22black%22>🌍</text></svg>">
-</head>
-<body>
-    <nav>
-        <div class="max-w-7xl mx-auto px-4 sm:px-6 lg:px-8">
-            <!-- Mobile menu button -->
-            <div class="sm:hidden absolute left-4 top-4">
-                <button onclick="toggleMobileMenu()" class="text-gray-500 hover:text-gray-700 focus:outline-none">
-                    <svg class="h-6 w-6" fill="none" viewBox="0 0 24 24" stroke="currentColor">
-                        <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M4 6h16M4 12h16M4 18h16" />
-                    </svg>
-                </button>
-            </div>
-            <!-- Mobile menu (hidden by default) -->
-            <div id="mobileMenu" class="hidden sm:hidden absolute left-0 top-16 bg-white shadow-lg py-4 mx-4 rounded-lg border border-gray-200">
-                <div class="flex flex-col">
-                    <h3 class="px-6 py-2 text-gray-400 text-sm font-medium">Navigation</h3>
-                    <a href="#" onclick="showSection('coverage'); toggleMobileMenu()" class="nav-link px-6 py-3 text-gray-600 hover:bg-gray-50">
-                        Language Coverage
-                    </a>
-                    <a href="#" onclick="showSection('comparison'); toggleMobileMenu()" class="nav-link px-6 py-3 text-gray-600 hover:bg-gray-50">
-                        LLM Comparison
-                    </a>
-                    <a href="#" onclick="showSection('results'); toggleMobileMenu()" class="nav-link px-6 py-3 text-gray-600 hover:bg-gray-50">
-                        Results by Language
-                    </a>
-                </div>
-            </div>
-            <!-- Desktop menu -->
-            <div class="hidden sm:flex justify-center h-16 border-b border-gray-200">
-                <div class="flex">
-                    <div class="flex space-x-8">
-                        <a href="#" onclick="showSection('coverage')" class="nav-link active inline-flex items-center px-1 pt-1 border-b-2 border-indigo-500 text-sm font-medium text-gray-900">
-                            Language Coverage
-                        </a>
-                        <a href="#" onclick="showSection('comparison')" class="nav-link inline-flex items-center px-1 pt-1 border-b-2 border-transparent text-sm font-medium text-gray-500 hover:border-gray-300 hover:text-gray-700">
-                            LLM Comparison
-                        </a>
-                        <a href="#" onclick="showSection('results')" class="nav-link inline-flex items-center px-1 pt-1 border-b-2 border-transparent text-sm font-medium text-gray-500 hover:border-gray-300 hover:text-gray-700">
-                            Results by Language
-                        </a>
-                    </div>
-                </div>
-            </div>
-        </div>
-    </nav>
-    <div class="p-6">
-        <section id="coverage" class="section">
-            <div id="summary-chart"></div>
-        </section>
-        <section id="comparison" class="section hidden">
-            <p class="text-gray-600">Coming soon...</p>
-            <!--
-            - Leaderboard
-                - Filters
-                    - commercial vs open source
-                - Eval results per task (across all languages)
-            - Timeline
-            -->
-        </section>
-        <section id="results" class="section hidden">
-            <div id="language-list"></div>
-            <!--
-            - Filters
-                - free-text search
-                - by continent, by language family
-                - sort by: population ><, performance ><, datasets ><
-            - Language list with details
-                - Eval results for each task and model
-                - Available datasets
-                - Form field to submit more datasets and custom models
-            -->
-        </section>
-    </div>
-    <script type="module">
-        // Import Plot using ESM
-        import * as Plot from "https://cdn.jsdelivr.net/npm/@observablehq/[email protected]/+esm";
-        function showSection(sectionId) {
-            // Update nav links
-            document.querySelectorAll('.nav-link').forEach(link => {
-                link.classList.remove('border-indigo-500', 'text-gray-900');
-                link.classList.add('border-transparent', 'text-gray-500');
-            });
-            const activeLink = document.querySelector(`[onclick="showSection('${sectionId}')"]`);
-            activeLink.classList.remove('border-transparent', 'text-gray-500');
-            activeLink.classList.add('border-indigo-500', 'text-gray-900');
-            // Show/hide sections
-            document.querySelectorAll('.section').forEach(section => {
-                section.classList.add('hidden');
-            });
-            document.getElementById(sectionId).classList.remove('hidden');
-        }
-        window.showSection = showSection;
-        function toggleMobileMenu() {
-            const mobileMenu = document.getElementById('mobileMenu');
-            mobileMenu.classList.toggle('hidden');
-        }
-        window.toggleMobileMenu = toggleMobileMenu;
-        async function init() {
-            const scoreKey = "bleu"
-            const scoreName = "BLEU Score"
-            const summaryChartDiv = document.getElementById('summary-chart');
-            const languageListDiv = document.getElementById('language-list');
-            const response = await fetch('results.json');
-            const data = await response.json();
-            // Format captions
-            const formatScore = (score) => score > 0 ? score.toFixed(2) : "No benchmark available!"
-            const formatTitle = d => (d.language_name + "\n" + parseInt(d.speakers / 1_000_00) / 10 + "M speakers\n" + scoreName + ": " + formatScore(d[scoreKey]))
-            // Create summary plot
-            const summaryPlot = Plot.plot({
-                width: summaryChartDiv.clientWidth,
-                height: 400,
-                marginBottom: 100,
-                x: { label: "Number of speakers", axis: null },
-                y: { label: `${scoreName} (average across models)` },
-                // color: { scheme: "BrBG" },
-                marks: [
-                    Plot.rectY(data, Plot.stackX({
-                        x: "speakers",
-                        order: scoreKey,
-                        reverse: true,
-                        y2: scoreKey, // y2 to avoid stacking by y
-                        title: formatTitle,
-                        tip: true,
-                        fill: d => d[scoreKey] > 0 ? "black" : "pink"
-                    })),
-                    Plot.rectY(data, Plot.pointerX(Plot.stackX({
-                        x: "speakers",
-                        order: scoreKey,
-                        reverse: true,
-                        y2: scoreKey, // y2 to avoid stacking by y
-                        fill: "grey",
-                    }))),
-                    Plot.text(data, Plot.stackX({
-                        x: "speakers",
-                        y2: scoreKey,
-                        order: scoreKey,
-                        reverse: true,
-                        text: "language_name",
-                        frameAnchor: "bottom",
-                        textAnchor: "end",
-                        dy: 10,
-                        rotate: 270,
-                        opacity: (d) => d.speakers > 50_000_000 ? 1 : 0,
-                    }))
-                ]
-            });
-            // Add summary plot to the coverage section
-            summaryChartDiv.appendChild(summaryPlot);
-            // Get unique languages with their speaker counts
-            const languageMap = new Map();
-            data.forEach(r => {
-                if (!languageMap.has(r.language_name)) {
-                    languageMap.set(r.language_name, r.speakers);
-                }
-            });
-            // Sort languages by speaker count (descending)
-            const languages = [...languageMap.entries()]
-                .sort((a, b) => b[1] - a[1])
-                .map(([lang]) => lang);
-            // Section for each language
-            languages.forEach(language => {
-                const headerDiv = document.createElement('div');
-                headerDiv.className = 'language-header';
-                const h2 = document.createElement('h2');
-                h2.textContent = language;
-                h2.style.marginBottom = '5px';
-                const speakerP = document.createElement('p');
-                speakerP.className = 'speaker-count';
-                const speakerCount = (languageMap.get(language) / 1_000_000).toFixed(1);
-                speakerP.textContent = `${speakerCount}M speakers`;
-                headerDiv.appendChild(h2);
-                headerDiv.appendChild(speakerP);
-                languageListDiv.appendChild(headerDiv);
-                const languageData = data.filter(r => r.language_name === language)[0]["scores"];
-                const descriptor = code => {
-                    let [org, model] = code.split("/")
-                    return model.split("-")[0]
-                }
-                // Plot for how well the models perform on this language
-                if (languageData && languageData.length > 1) {
-                    const plot = Plot.plot({
-                        width: 400,
-                        height: 200,
-                        margin: 30,
-                        y: {
-                            domain: [0, 1],
-                            label: scoreName
-                        },
-                        marks: [
-                            Plot.barY(languageData, {
-                                x: d => descriptor(d.model),
-                                y: scoreKey
-                            })
-                        ]
-                    });
-                    languageListDiv.appendChild(plot);
-                }
-            });
-        }
-        init();
-    </script>
-</body>
-</html>

observablehq.config.js ADDED Viewed

	@@ -0,0 +1,34 @@

+// See https://observablehq.com/framework/config for documentation.
+export default {
+  // The app’s title; used in the sidebar and webpage titles.
+  title: "AI Language Monitor",
+  // The pages and sections in the sidebar. If you don’t specify this option,
+  // all pages will be listed in alphabetical order. Listing pages explicitly
+  // lets you organize them into sections and have unlisted pages.
+  pages: [
+    { name: "Compare Languages", path: "/compare-languages" },
+    { name: "Compare AI Models", path: "/compare-ai-models" },
+    { name: "Methodology", path: "/methodology" },
+  ],
+  // Content to add to the head of the page, e.g. for a favicon:
+  head: '<link rel="icon" href="data:image/svg+xml,<svg xmlns=%22http://www.w3.org/2000/svg%22 viewBox=%220 0 100 100%22><text y=%22.9em%22 font-size=%2290%22 fill=%22black%22>🌍</text></svg>">',
+  // The path to the source root.
+  root: "src",
+  // Some additional configuration options and their defaults:
+  // theme: "default", // try "light", "dark", "slate", etc.
+  // header: "", // what to show in the header (HTML)
+  // footer: "Built with Observable.", // what to show in the footer (HTML)
+  // sidebar: true, // whether to show the sidebar
+  // toc: true, // whether to show the table of contents
+  // pager: true, // whether to show previous & next links in the footer
+  // output: "dist", // path to the output root for build
+  // search: true, // activate search
+  // linkify: true, // convert URLs in Markdown to links
+  // typographer: false, // smart quotes and other typographic improvements
+  // preserveExtension: false, // drop .html from URLs
+  // preserveIndex: false, // drop /index from URLs
+};

package-lock.json ADDED Viewed

The diff for this file is too large to render. See raw diff

package.json ADDED Viewed

	@@ -0,0 +1,22 @@

+{
+  "type": "module",
+  "private": true,
+  "scripts": {
+    "clean": "rimraf src/.observablehq/cache",
+    "build": "observable build",
+    "dev": "observable preview",
+    "deploy": "observable deploy",
+    "observable": "observable"
+  },
+  "dependencies": {
+    "@observablehq/framework": "^1.13.2",
+    "d3-dsv": "^3.0.1",
+    "d3-time-format": "^4.1.0"
+  },
+  "devDependencies": {
+    "rimraf": "^5.0.5"
+  },
+  "engines": {
+    "node": ">=18"
+  }
+}

results.json DELETED Viewed

The diff for this file is too large to render. See raw diff

src/.gitignore ADDED Viewed

	@@ -0,0 +1 @@


1	+ /.observablehq/cache/

src/compare-ai-models.md ADDED Viewed

	@@ -0,0 +1,65 @@

+---
+theme: dashboard
+title: Compare AI models
+---
+# Compare AI models
+```js
+const data = FileAttachment("data/languagebench.json").json();
+```
+```js
+const scoreKey = "bleu"
+const scoreName = "BLEU Score"
+// Get unique languages with their speaker counts
+const languageMap = new Map();
+data.forEach(r => {
+    if (!languageMap.has(r.language_name)) {
+        languageMap.set(r.language_name, r.speakers);
+    }
+});
+// Sort languages by speaker count (descending)
+const languages = [...languageMap.entries()]
+    .sort((a, b) => b[1] - a[1])
+    .map(([lang]) => lang);
+// Section for each language
+languages.forEach(language => {
+    display(html`<h2 class="language-header">${language}</h2>`)
+    const speakerCount = (languageMap.get(language) / 1_000_000).toFixed(1);
+    display(html`${speakerCount}M speakers`);
+    const languageData = data.filter(r => r.language_name === language)[0]["scores"];
+    console.log(languageData)
+    const descriptor = code => {
+        let [org, model] = code.split("/")
+        return model.split("-")[0]
+    }
+    // Plot for how well the models perform on this language
+    if (languageData && languageData.length >= 1) {
+        console.log("yes")
+        const chart = Plot.plot({
+            width: 400,
+            height: 200,
+            margin: 30,
+            y: {
+                domain: [0, 1],
+                label: scoreName
+            },
+            marks: [
+                Plot.barY(languageData, {
+                    x: d => descriptor(d.model),
+                    y: scoreKey
+                })
+            ]
+        });
+        display(chart)
+    }
+});
+```

src/compare-languages.md ADDED Viewed

	@@ -0,0 +1,20 @@

+---
+theme: dashboard
+title: Compare languages
+---
+# Compare languages
+```js
+import { languageChart } from "./components/language-chart.js";
+const data = FileAttachment("data/languagebench.json").json();
+```
+```js
+const scoreKey = "bleu"
+const scoreName = "BLEU Score"
+// Create summary plot
+display(languageChart(data, {width: 1000, height: 400, scoreKey: scoreKey, scoreName: scoreName}))
+```

src/components/language-chart.js ADDED Viewed

	@@ -0,0 +1,68 @@

+import * as Plot from "npm:@observablehq/plot";
+export function languageChart(
+  languageData,
+  { width, height, scoreKey, scoreName } = {}
+) {
+  // Format captions
+  const formatScore = (score) =>
+    score > 0 ? score.toFixed(2) : "No benchmark available!";
+  const formatTitle = (d) =>
+    d.language_name +
+    "\n" +
+    parseInt(d.speakers / 1_000_00) / 10 +
+    "M speakers\n" +
+    scoreName +
+    ": " +
+    formatScore(d[scoreKey]);
+  return Plot.plot({
+    width: width,
+    height: height,
+    marginBottom: 100,
+    x: { label: "Number of speakers", axis: null },
+    y: { label: `${scoreName} (average across models)` },
+    // color: { scheme: "BrBG" },
+    marks: [
+      Plot.rectY(
+        languageData,
+        Plot.stackX({
+          x: "speakers",
+          order: scoreKey,
+          reverse: true,
+          y2: scoreKey, // y2 to avoid stacking by y
+          title: formatTitle,
+          tip: true,
+          fill: (d) => (d[scoreKey] > 0 ? "black" : "pink"),
+        })
+      ),
+      Plot.rectY(
+        languageData,
+        Plot.pointerX(
+          Plot.stackX({
+            x: "speakers",
+            order: scoreKey,
+            reverse: true,
+            y2: scoreKey, // y2 to avoid stacking by y
+            fill: "grey",
+          })
+        )
+      ),
+      Plot.text(
+        languageData,
+        Plot.stackX({
+          x: "speakers",
+          y2: scoreKey,
+          order: scoreKey,
+          reverse: true,
+          text: "language_name",
+          frameAnchor: "bottom",
+          textAnchor: "end",
+          dy: 10,
+          rotate: 270,
+          opacity: (d) => (d.speakers > 50_000_000 ? 1 : 0),
+        })
+      ),
+    ],
+  });
+}

data.txt → src/data/data.txt RENAMED Viewed

@@ -1,4 +1,4 @@
 floresp-v2.0-rc.3: https://github.com/openlanguagedata/flores
 languages.csv: generated from https://query.wikidata.org/ using the languages.rq query
 LanguageCodes.tab: https://www.ethnologue.com/
-ScriptCodes.csv: https://www.unicode.org/iso15924/iso15924-codes.html

 floresp-v2.0-rc.3: https://github.com/openlanguagedata/flores
 languages.csv: generated from https://query.wikidata.org/ using the languages.rq query
 LanguageCodes.tab: https://www.ethnologue.com/
+ScriptCodes.csv: https://www.unicode.org/iso15924/iso15924-codes.html

languagebench.py → src/data/languagebench.json.py RENAMED Viewed

@@ -1,7 +1,9 @@
 import asyncio
 import json
 import os
 from os import getenv
 import evaluate
 import pandas as pd
@@ -15,14 +17,14 @@ from transformers import NllbTokenizer
 # config
 models = [
-    "openai/gpt-4o",
-    "anthropic/claude-3.5-sonnet",
-    "meta-llama/llama-3.1-405b-instruct",  # lots of slow repetitions for LRLs
-    "mistralai/mistral-large",
     # "google/gemini-flash-1.5",  # very fast
-    "qwen/qwen-2.5-72b-instruct",  # somewhat slow
 ]
-fast_model = "anthropic/claude-3.5-sonnet"
 n_sentences = 30
 # setup
@@ -43,9 +45,9 @@ def reorder(language_name):
         return language_name.split(",")[1] + " " + language_name.split(",")[0]
     return language_name
 # load benchmark languages and scripts
-benchmark_dir = "floresp-v2.0-rc.3/dev"
 benchmark_languages = pd.DataFrame(
     [f.split(".")[1].split("_", 1) for f in os.listdir(benchmark_dir)],
     columns=["language_code", "script_code"],
@@ -56,7 +58,7 @@ benchmark_languages["in_benchmark"] = True
 # load Ethnologue language names
 language_names = (
-    pd.read_csv("LanguageCodes.tab", sep="\t")
     .rename(columns={"LangID": "language_code", "Name": "language_name"})[
         ["language_code", "language_name"]
     ]
@@ -65,7 +67,7 @@ language_names = (
 # load Wikidata speaker stats
 language_stats = (
-    pd.read_csv("languages.tsv", sep="\t")
     .rename(columns={"iso639_3": "language_code", "maxSpeakers": "speakers"})[
         ["language_code", "speakers"]
     ]
@@ -84,7 +86,7 @@ language_stats = language_stats[
 ]
 # load unicode script names
-script_names = pd.read_csv("ScriptCodes.csv").rename(
     columns={"Code": "script_code", "English Name": "script_name"}
 )[["script_code", "script_name"]]
@@ -160,13 +162,13 @@ def load_sentences(language):
 # evaluation!
 async def main():
     results = []
-    for language in languages.itertuples():
         name = (
             language.language_name
             if not pd.isna(language.language_name)
             else language.language_code
         )
-        print(name)
         scores = []
         if language.in_benchmark:
             target_sentences = load_sentences(language)[:n_sentences]
@@ -185,7 +187,7 @@ async def main():
                     load_sentences(lang)[i]
                     for i, lang in enumerate(_original_languages.itertuples())
                 ]
-                print(model)
                 predictions = [
                     translate(
                         model, language.language_name, language.script_name, sentence
@@ -220,8 +222,7 @@ async def main():
                 # "bert_score": mean([s["bert_score"] for s in scores]),
             }
         )
-        with open("results.json", "w") as f:
-            json.dump(results, f, indent=2, ensure_ascii=False)
 if __name__ == "__main__":

 import asyncio
 import json
 import os
+import sys
 from os import getenv
+from pathlib import Path
 import evaluate
 import pandas as pd
 # config
 models = [
+    "openai/gpt-4o-mini",
+    "anthropic/claude-3.5-haiku",
+    # "meta-llama/llama-3.1-405b-instruct",  # lots of slow repetitions for LRLs
+    # "mistralai/mistral-large",
     # "google/gemini-flash-1.5",  # very fast
+    # "qwen/qwen-2.5-72b-instruct",  # somewhat slow
 ]
+fast_model = "anthropic/claude-3.5-haiku"
 n_sentences = 30
 # setup
         return language_name.split(",")[1] + " " + language_name.split(",")[0]
     return language_name
 # load benchmark languages and scripts
+data = Path("src/data")
+benchmark_dir = data / "floresp-v2.0-rc.3/dev"
 benchmark_languages = pd.DataFrame(
     [f.split(".")[1].split("_", 1) for f in os.listdir(benchmark_dir)],
     columns=["language_code", "script_code"],
 # load Ethnologue language names
 language_names = (
+    pd.read_csv(data / "LanguageCodes.tab", sep="\t")
     .rename(columns={"LangID": "language_code", "Name": "language_name"})[
         ["language_code", "language_name"]
     ]
 # load Wikidata speaker stats
 language_stats = (
+    pd.read_csv(data / "languages.tsv", sep="\t")
     .rename(columns={"iso639_3": "language_code", "maxSpeakers": "speakers"})[
         ["language_code", "speakers"]
     ]
 ]
 # load unicode script names
+script_names = pd.read_csv(data / "ScriptCodes.csv").rename(
     columns={"Code": "script_code", "English Name": "script_name"}
 )[["script_code", "script_name"]]
 # evaluation!
 async def main():
     results = []
+    for language in list(languages.itertuples())[:5]:
         name = (
             language.language_name
             if not pd.isna(language.language_name)
             else language.language_code
         )
+        print(name, file=sys.stderr)
         scores = []
         if language.in_benchmark:
             target_sentences = load_sentences(language)[:n_sentences]
                     load_sentences(lang)[i]
                     for i, lang in enumerate(_original_languages.itertuples())
                 ]
+                print(model, file=sys.stderr)
                 predictions = [
                     translate(
                         model, language.language_name, language.script_name, sentence
                 # "bert_score": mean([s["bert_score"] for s in scores]),
             }
         )
+    print(json.dumps(results, indent=2, ensure_ascii=False))
 if __name__ == "__main__":

languages.rq → src/data/languages.rq RENAMED Viewed

File without changes

languages.tsv → src/data/languages.tsv RENAMED Viewed

File without changes

src/index.md ADDED Viewed

	@@ -0,0 +1,53 @@

+---
+toc: false
+---
+<div class="hero">
+  <h1>AI Language Monitor</h1>
+  <h2>Benchmarking all big AI models on all benchmarkable languages.</h2>
+</div>
+```js
+import { languageChart } from "./components/language-chart.js";
+const data = FileAttachment("data/languagebench.json").json();
+```
+<div class="grid grid-cols-2" style="grid-auto-rows: 504px;">
+  <div class="card">
+    <h2 class="hero">Compare languages</h2>
+    ${resize((width) => languageChart(data, {width: 1000, height: 400, scoreKey: "bleu", scoreName: "BLEU Score"}))}
+  </div>
+  <div class="card">
+    <h2 class="hero">Compare AI models</h2>
+    ...
+  </div>
+</div>
+<style>
+.hero {
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  font-family: var(--sans-serif);
+  margin: 4rem 0 8rem;
+  text-wrap: balance;
+  text-align: center;
+}
+.hero h1 {
+  margin: 1rem 0;
+  padding: 1rem 0;
+  max-width: none;
+  font-size: 90px;
+  font-weight: 900;
+  line-height: 1;
+  background: linear-gradient(30deg, var(--theme-foreground-focus), currentColor);
+  -webkit-background-clip: text;
+  -webkit-text-fill-color: transparent;
+  background-clip: text;
+}
+</style>

src/methodology.md ADDED Viewed

	@@ -0,0 +1,12 @@

+---
+title: Methodology
+---
+# Methodology
+Sources:
+1. For AI models: [OpenRouter](https://openrouter.ai/)
+2. For language benchmarks: [FLORES+](https://github.com/openlanguagedata/flores)
+3. For language statistics: [Wikidata](https://gist.github.com/unhammer/3e8f2e0f79972bf5008a4c970081502d), [Ethnologue](https://www.ethnologue.com/browse/names/)