Update benchmark leaderboard
Browse files- index.html +4 -4
index.html
CHANGED
|
@@ -284,9 +284,9 @@
|
|
| 284 |
|
| 285 |
<script>
|
| 286 |
(function() {
|
| 287 |
-
const ALL_ROWS = [{"model": "CYFRAGOVPL/Llama-PLLuM-70B-chat-250801", "avg_exact": 0.
|
| 288 |
const ALL_LANGS = ["ab", "ar", "az", "be", "bg", "bo", "ca", "cn", "cs", "cy", "da", "de", "el", "en", "es", "et", "eu", "fa", "fi", "fr", "gl", "he", "hi", "hu", "hv", "ir", "is", "it", "ka", "kz", "la", "li", "lv", "mk", "mt", "nl", "no", "pl"];
|
| 289 |
-
const LANG_NAMES = {"af": "Afrikaans", "ab": "Arabic", "az": "Azerbaijani", "ar": "Armenian", "be": "Belarusian", "bo": "Bosnian", "bg": "Bulgarian", "bn": "Brunei", "ca": "Catalan", "cs": "Czech", "cn": "Chinese", "cy": "Welsh", "da": "Danish", "de": "German", "el": "Greek", "en": "English", "eo": "Esperanto", "es": "Spanish", "et": "Estonian", "eu": "Basque", "fa": "Faroese", "fi": "Finnish", "fr": "French", "ga": "Irish", "gl": "Galician", "gu": "Gujarati", "he": "Hebrew", "hi": "Hindi", "hr": "Croatian", "hu": "Hungarian", "hv": "Croatia", "id": "Indonesian", "is": "Icelandic", "it": "Italian", "ir": "Irish", "ja": "Japanese", "ka": "Georgian", "
|
| 290 |
const LANG_COUNTS = {"ab": 600, "ar": 400, "az": 400, "be": 400, "bg": 362, "bo": 400, "ca": 359, "cn": 600, "cs": 400, "cy": 600, "da": 400, "de": 600, "el": 400, "en": 600, "es": 333, "et": 400, "eu": 597, "fa": 600, "fi": 400, "fr": 382, "gl": 600, "he": 600, "hi": 544, "hu": 400, "hv": 400, "ir": 600, "is": 400, "it": 400, "ka": 318, "kz": 600, "la": 513, "li": 400, "lv": 400, "mk": 400, "mt": 400, "nl": 400, "no": 400, "pl": 571, "pt": 400, "ro": 400, "ru": 600, "sk": 400, "sl": 400, "sq": 400, "sr": 400, "sv": 400, "tr": 400, "uk": 574};
|
| 291 |
const LANG_DIST = {"ab": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "ar": {"1": 100, "2": 100, "5": 100, "6": 100}, "az": {"1": 100, "2": 100, "5": 100, "6": 100}, "be": {"1": 100, "2": 100, "5": 100, "6": 100}, "bg": {"1": 100, "2": 100, "5": 100, "6": 62}, "bo": {"1": 100, "2": 100, "5": 100, "6": 100}, "ca": {"1": 73, "2": 100, "5": 100, "6": 86}, "cn": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "cs": {"1": 100, "2": 100, "5": 100, "6": 100}, "cy": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "da": {"1": 100, "2": 100, "5": 100, "6": 100}, "de": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "el": {"1": 100, "2": 100, "5": 100, "6": 100}, "en": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "es": {"1": 100, "2": 100, "5": 100, "6": 33}, "et": {"1": 100, "2": 100, "5": 100, "6": 100}, "eu": {"1": 97, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "fa": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "fi": {"1": 100, "2": 100, "5": 100, "6": 100}, "fr": {"1": 100, "2": 100, "5": 100, "6": 82}, "gl": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "he": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "hi": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 44}, "hu": {"1": 100, "2": 100, "5": 100, "6": 100}, "hv": {"1": 100, "2": 100, "5": 100, "6": 100}, "ir": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "is": {"1": 100, "2": 100, "5": 100, "6": 100}, "it": {"1": 100, "2": 100, "5": 100, "6": 100}, "ka": {"1": 18, "2": 100, "5": 100, "6": 100}, "kz": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "la": {"1": 100, "2": 96, "3": 17, "4": 100, "5": 100, "6": 100}, "li": {"1": 100, "2": 100, "5": 100, "6": 100}, "lv": {"1": 100, "2": 100, "5": 100, "6": 100}, "mk": {"1": 100, "2": 100, "5": 100, "6": 100}, "mt": {"1": 100, "2": 100, "5": 100, "6": 100}, "nl": {"1": 100, "2": 100, "5": 100, "6": 100}, "no": {"1": 100, "2": 100, "5": 100, "6": 100}, "pl": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 71}, "pt": {"1": 100, "2": 100, "5": 100, "6": 100}, "ro": {"1": 100, "2": 100, "5": 100, "6": 100}, "ru": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "sk": {"1": 100, "2": 100, "5": 100, "6": 100}, "sl": {"1": 100, "2": 100, "5": 100, "6": 100}, "sq": {"1": 100, "2": 100, "5": 100, "6": 100}, "sr": {"1": 100, "2": 100, "5": 100, "6": 100}, "sv": {"1": 100, "2": 100, "5": 100, "6": 100}, "tr": {"1": 100, "2": 100, "5": 100, "6": 100}, "uk": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 74}};
|
| 292 |
|
|
@@ -318,7 +318,7 @@
|
|
| 318 |
function dsLangs() { return activeDS === 1 ? ALL_LANGS : ALL_LANGS_2; }
|
| 319 |
function dsLangCounts() { return activeDS === 1 ? LANG_COUNTS : LANG_COUNTS_2; }
|
| 320 |
function dsLangDist() { return activeDS === 1 ? LANG_DIST : LANG_DIST_2; }
|
| 321 |
-
function dsTotalPreds() { return activeDS === 1 ?
|
| 322 |
|
| 323 |
window.setDataset = function(ds) {
|
| 324 |
if (ds === 2 && !HAS_SECOND) return;
|
|
@@ -479,7 +479,7 @@
|
|
| 479 |
}).join('');
|
| 480 |
|
| 481 |
document.getElementById('footer').textContent =
|
| 482 |
-
`
|
| 483 |
}
|
| 484 |
|
| 485 |
// ── dataset distribution ──
|
|
|
|
| 284 |
|
| 285 |
<script>
|
| 286 |
(function() {
|
| 287 |
+
const ALL_ROWS = [{"model": "CYFRAGOVPL/Llama-PLLuM-70B-chat-250801", "avg_exact": 0.196988, "avg_wp": 0.301437, "avg_bias": -1.7929, "avg_parsed": 0.751704, "avg_mae": 2.0205, "total": 17463, "lang_exact": {"ab": 0.045, "ar": 0.2225, "az": 0.26, "be": 0.235, "bg": 0.21547, "bo": 0.3275, "ca": 0.144847, "cn": 0.2, "cs": 0.25, "cy": 0.233333, "da": 0.2, "de": 0.163333, "el": 0.1875, "en": 0.135, "es": 0.114114, "et": 0.25, "eu": 0.132328, "fa": 0.288333, "fi": 0.24, "fr": 0.230366, "gl": 0.135, "he": 0.08, "hi": 0.176471, "hu": 0.1175, "hv": 0.4075, "ir": 0.275, "is": 0.3425, "it": 0.13, "ka": 0.031447, "kz": 0.095, "la": 0.350877, "li": 0.2425, "lv": 0.23, "mk": 0.1925, "mt": 0.2425, "nl": 0.2225, "no": 0.1725, "pl": 0.087912}, "lang_wp": {"ab": 0.0625, "ar": 0.40125, "az": 0.39375, "be": 0.35, "bg": 0.319061, "bo": 0.47125, "ca": 0.232591, "cn": 0.291667, "cs": 0.36125, "cy": 0.344167, "da": 0.3125, "de": 0.2475, "el": 0.2675, "en": 0.230833, "es": 0.15015, "et": 0.38125, "eu": 0.201005, "fa": 0.458333, "fi": 0.3525, "fr": 0.332461, "gl": 0.205, "he": 0.144167, "hi": 0.286765, "hu": 0.13875, "hv": 0.57, "ir": 0.439167, "is": 0.57, "it": 0.18375, "ka": 0.157233, "kz": 0.145833, "la": 0.519493, "li": 0.36125, "lv": 0.355, "mk": 0.2825, "mt": 0.39125, "nl": 0.355, "no": 0.2575, "pl": 0.113187}, "lang_parsed": {"ab": 0.16, "ar": 0.785, "az": 0.94, "be": 0.9275, "bg": 0.801105, "bo": 0.875, "ca": 0.62117, "cn": 0.751667, "cs": 0.7525, "cy": 0.806667, "da": 0.925, "de": 0.958333, "el": 0.69, "en": 0.848333, "es": 0.228228, "et": 1.0, "eu": 0.716918, "fa": 0.878333, "fi": 0.905, "fr": 0.730366, "gl": 0.651667, "he": 0.346667, "hi": 0.851103, "hu": 0.205, "hv": 0.9525, "ir": 0.973333, "is": 0.99, "it": 0.4, "ka": 0.889937, "kz": 0.77, "la": 1.0, "li": 0.9725, "lv": 0.945, "mk": 0.6475, "mt": 0.845, "nl": 0.9225, "no": 0.71, "pl": 0.283516}, "lang_mae": {"ab": 1.9271, "ar": 1.2962, "az": 2.1011, "be": 2.4798, "bg": 2.3138, "bo": 1.6314, "ca": 2.3946, "cn": 1.8271, "cs": 1.8771, "cy": 1.6756, "da": 2.6757, "de": 2.4417, "el": 2.4565, "en": 2.2318, "es": 1.0789, "et": 2.3775, "eu": 2.3762, "fa": 1.1917, "fi": 2.4144, "fr": 2.0036, "gl": 2.2558, "he": 1.5, "hi": 1.8553, "hu": 1.2439, "hv": 1.3307, "ir": 1.4366, "is": 0.9697, "it": 2.0812, "ka": 3.318, "kz": 2.8117, "la": 1.2768, "li": 2.5244, "lv": 2.4392, "mk": 2.1969, "mt": 1.7337, "nl": 2.2222, "no": 2.5634, "pl": 1.6124}, "confusion": {"1": {"1": 0.9206, "3": 0.0453, "4": 0.0172, "5": 0.0115, "2": 0.003, "6": 0.0024}, "2": {"1": 0.8714, "3": 0.0706, "5": 0.0219, "4": 0.0272, "2": 0.0046, "6": 0.0042}, "3": {"3": 0.1209, "5": 0.0644, "1": 0.7559, "4": 0.0588}, "4": {"1": 0.6881, "5": 0.0887, "4": 0.1264, "3": 0.0877, "6": 0.0082, "2": 0.001}, "5": {"1": 0.754, "5": 0.1214, "4": 0.0862, "3": 0.0236, "6": 0.013, "2": 0.0018}, "6": {"6": 0.0505, "1": 0.6997, "5": 0.1745, "4": 0.0616, "3": 0.0133, "2": 0.0004}}}];
|
| 288 |
const ALL_LANGS = ["ab", "ar", "az", "be", "bg", "bo", "ca", "cn", "cs", "cy", "da", "de", "el", "en", "es", "et", "eu", "fa", "fi", "fr", "gl", "he", "hi", "hu", "hv", "ir", "is", "it", "ka", "kz", "la", "li", "lv", "mk", "mt", "nl", "no", "pl"];
|
| 289 |
+
const LANG_NAMES = {"af": "Afrikaans", "ab": "Arabic", "az": "Azerbaijani", "ar": "Armenian", "be": "Belarusian", "bo": "Bosnian", "bg": "Bulgarian", "bn": "Brunei", "ca": "Catalan", "cs": "Czech", "cn": "Chinese", "cy": "Welsh", "da": "Danish", "de": "German", "el": "Greek", "en": "English", "eo": "Esperanto", "es": "Spanish", "et": "Estonian", "eu": "Basque", "fa": "Faroese", "fi": "Finnish", "fr": "French", "ga": "Irish", "gl": "Galician", "gu": "Gujarati", "he": "Hebrew", "hi": "Hindi", "hr": "Croatian", "hu": "Hungarian", "hv": "Croatia", "id": "Indonesian", "is": "Icelandic", "it": "Italian", "ir": "Irish", "ja": "Japanese", "ka": "Georgian", "kz": "Kazakh", "km": "Khmer", "kn": "Kannada", "ko": "Korean", "la": "Latin", "li": "Lithuanian", "lv": "Latvian", "mk": "Macedonian", "ml": "Malayalam", "mn": "Mongolian", "mr": "Marathi", "ms": "Malay", "mt": "Maltese", "my": "Burmese", "ne": "Nepali", "nl": "Dutch", "no": "Norwegian", "pa": "Punjabi", "pe": "Persian", "pl": "Polish", "pt": "Portuguese", "ro": "Romanian", "ru": "Russian", "si": "Sinhala", "sk": "Slovak", "sl": "Slovenian", "sq": "Albanian", "sr": "Serbian", "sv": "Swedish", "sw": "Swahili", "ta": "Tamil", "te": "Telugu", "th": "Thai", "tl": "Filipino", "tr": "Turkish", "uk": "Ukrainian", "ur": "Urdu", "uz": "Uzbek", "vi": "Vietnamese", "zh": "Chinese", "zu": "Zulu"};
|
| 290 |
const LANG_COUNTS = {"ab": 600, "ar": 400, "az": 400, "be": 400, "bg": 362, "bo": 400, "ca": 359, "cn": 600, "cs": 400, "cy": 600, "da": 400, "de": 600, "el": 400, "en": 600, "es": 333, "et": 400, "eu": 597, "fa": 600, "fi": 400, "fr": 382, "gl": 600, "he": 600, "hi": 544, "hu": 400, "hv": 400, "ir": 600, "is": 400, "it": 400, "ka": 318, "kz": 600, "la": 513, "li": 400, "lv": 400, "mk": 400, "mt": 400, "nl": 400, "no": 400, "pl": 571, "pt": 400, "ro": 400, "ru": 600, "sk": 400, "sl": 400, "sq": 400, "sr": 400, "sv": 400, "tr": 400, "uk": 574};
|
| 291 |
const LANG_DIST = {"ab": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "ar": {"1": 100, "2": 100, "5": 100, "6": 100}, "az": {"1": 100, "2": 100, "5": 100, "6": 100}, "be": {"1": 100, "2": 100, "5": 100, "6": 100}, "bg": {"1": 100, "2": 100, "5": 100, "6": 62}, "bo": {"1": 100, "2": 100, "5": 100, "6": 100}, "ca": {"1": 73, "2": 100, "5": 100, "6": 86}, "cn": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "cs": {"1": 100, "2": 100, "5": 100, "6": 100}, "cy": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "da": {"1": 100, "2": 100, "5": 100, "6": 100}, "de": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "el": {"1": 100, "2": 100, "5": 100, "6": 100}, "en": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "es": {"1": 100, "2": 100, "5": 100, "6": 33}, "et": {"1": 100, "2": 100, "5": 100, "6": 100}, "eu": {"1": 97, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "fa": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "fi": {"1": 100, "2": 100, "5": 100, "6": 100}, "fr": {"1": 100, "2": 100, "5": 100, "6": 82}, "gl": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "he": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "hi": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 44}, "hu": {"1": 100, "2": 100, "5": 100, "6": 100}, "hv": {"1": 100, "2": 100, "5": 100, "6": 100}, "ir": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "is": {"1": 100, "2": 100, "5": 100, "6": 100}, "it": {"1": 100, "2": 100, "5": 100, "6": 100}, "ka": {"1": 18, "2": 100, "5": 100, "6": 100}, "kz": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "la": {"1": 100, "2": 96, "3": 17, "4": 100, "5": 100, "6": 100}, "li": {"1": 100, "2": 100, "5": 100, "6": 100}, "lv": {"1": 100, "2": 100, "5": 100, "6": 100}, "mk": {"1": 100, "2": 100, "5": 100, "6": 100}, "mt": {"1": 100, "2": 100, "5": 100, "6": 100}, "nl": {"1": 100, "2": 100, "5": 100, "6": 100}, "no": {"1": 100, "2": 100, "5": 100, "6": 100}, "pl": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 71}, "pt": {"1": 100, "2": 100, "5": 100, "6": 100}, "ro": {"1": 100, "2": 100, "5": 100, "6": 100}, "ru": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 100}, "sk": {"1": 100, "2": 100, "5": 100, "6": 100}, "sl": {"1": 100, "2": 100, "5": 100, "6": 100}, "sq": {"1": 100, "2": 100, "5": 100, "6": 100}, "sr": {"1": 100, "2": 100, "5": 100, "6": 100}, "sv": {"1": 100, "2": 100, "5": 100, "6": 100}, "tr": {"1": 100, "2": 100, "5": 100, "6": 100}, "uk": {"1": 100, "2": 100, "3": 100, "4": 100, "5": 100, "6": 74}};
|
| 292 |
|
|
|
|
| 318 |
function dsLangs() { return activeDS === 1 ? ALL_LANGS : ALL_LANGS_2; }
|
| 319 |
function dsLangCounts() { return activeDS === 1 ? LANG_COUNTS : LANG_COUNTS_2; }
|
| 320 |
function dsLangDist() { return activeDS === 1 ? LANG_DIST : LANG_DIST_2; }
|
| 321 |
+
function dsTotalPreds() { return activeDS === 1 ? 17463 : TOTAL_PREDS_2; }
|
| 322 |
|
| 323 |
window.setDataset = function(ds) {
|
| 324 |
if (ds === 2 && !HAS_SECOND) return;
|
|
|
|
| 479 |
}).join('');
|
| 480 |
|
| 481 |
document.getElementById('footer').textContent =
|
| 482 |
+
`17463 predictions · ${dsLangs().length} languages · ${dsRows().length} models`;
|
| 483 |
}
|
| 484 |
|
| 485 |
// ── dataset distribution ──
|