-
Notifications
You must be signed in to change notification settings - Fork 232
/
index.html
435 lines (363 loc) · 15.8 KB
/
index.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
<!DOCTYPE html>
<html>
<head>
<title>AlpacaEval Leaderboard</title>
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-VWV023WWP4"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag('js', new Date());
gtag('config', 'G-VWV023WWP4');
</script>
<link rel="stylesheet" href="https://stackpath.bootstrapcdn.com/bootstrap/4.3.1/css/bootstrap.min.css">
<link rel="icon" href="https://raw.githubusercontent.com/tatsu-lab/alpaca_eval/main/docs/AlpacaFarm_small.png">
<link href="https://cdn.jsdelivr.net/css-toggle-switch/latest/toggle-switch.css" rel="stylesheet"/>
<style>
body {
font-family: Arial, sans-serif;
margin: 0;
padding: 50px 20px;
background-color: #FFFFFF;
color: #000000;
}
.container {
max-width: 700px;
margin: auto;
}
#branding {
text-align: center;
margin-bottom: 20px;
}
#branding h1 {
margin: 0;
font-size: 2em;
}
h2 {
margin: 0;
font-size: 1.2em;
color: #777;
}
table {
max-width: 700px;
width: 100%;
table-layout: fixed;
margin: auto;
font-size: 1em;
}
table th,
table td {
padding: 6px;
word-wrap: break-word;
vertical-align: middle;
}
table th {
border-bottom: 2px solid #000;
}
th.rank, td.rank {
width: 9%; /* Adjust as needed */
padding-left: 10px; /* Small margin */
text-align: left;
}
th.name, td.name {
width: 55%;
padding-left: 30px;
text-align: left;
}
th:not(.rank):not(.name),
td:not(.rank):not(.name) {
text-align: right;
padding-right: 10px;
}
th.winRate, td.winRate {
width: 17%;
padding-right: 30px;
}
th {
text-align: right;
padding-bottom: 15px;
}
td {
padding-bottom: 10px;
}
#leaderboard tr th.winRate,
#leaderboard tr td.winRate {
color: #999999;
}
#leaderboard tr th.rank,
#leaderboard tr td.rank {
color: #999999;
}
table tr:nth-child(even) {
background-color: #E8E8E8;
}
table tr:nth-child(odd) {
background-color: #F8F8F8;
}
.switch-toggle {
display: inline-block;
vertical-align: middle;
}
.switch-toggle input + label {
padding: 2px;
padding-left: 7px;
padding-right: 7px;
cursor: pointer;
background-color: lightgrey;
border: 1px solid transparent;
font-size: 16px;
}
.switch-toggle input:checked + label {
border-color: green;
color: green;
}
.switch-toggle input:not(:checked) + label {
color: black;
box-shadow: none !important;
user-select: none;
}
.toggle-line {
display: flex;
justify-content: center;
align-items: center;
margin-bottom: 20px;
font-size: 17px;
}
.toggle-line .switch-toggle {
margin: 0 10px;
}
</style>
<script src="https://cdnjs.cloudflare.com/ajax/libs/PapaParse/5.3.0/papaparse.min.js"></script>
</head>
<body>
<div class="container">
<div id="branding">
<h1>AlpacaEval
<a href="https://github.com/tatsu-lab/alpaca_eval/tree/main">
<img src="https://raw.githubusercontent.com/tatsu-lab/alpaca_eval/main/docs/AlpacaFarm_small.png"
alt="Logo" style="height: 2em; vertical-align: middle;"></a>
Leaderboard
</h1>
<br>
<h2>An Automatic Evaluator for Instruction-following Language Models</h2>
<!-- <small id="alpaca_eval_info" style="color: #777;">-->
<!-- Baseline: GPT-4 Preview | Auto-annotator: GPT-4 Preview-->
<!-- </small>-->
<!-- <br>-->
<small id="caution" style="color: #8C1515;">
<b> Length-controlled</b> (LC) win rates alleviate length biases of GPT-4, but it may favor models finetuned on its outputs.
</small>
<br>
<a href="https://github.com/tatsu-lab/alpaca_eval">
<img src="https://github.githubassets.com/images/modules/logos_page/GitHub-Mark.png" alt="GitHub logo" style="height: 1.5em;/* margin-bottom: 0; */">
</a>
</div>
<div class="toggle-line">
Version:
<div class="switch-toggle switch-evaluator" style="margin-right: 4em">
<input id="alpaca_eval" name="version" type="radio"/>
<label for="alpaca_eval" onclick="">AlpacaEval</label>
<input id="alpaca_eval_2" name="version" type="radio" checked="checked"/>
<label for="alpaca_eval_2" onclick="">AlpacaEval 2.0</label>
</div>
Filter:
<div class="switch-toggle switch-filter">
<input id="community" name="filter" type="radio"/>
<label for="community" onclick="">Community</label>
<input id="verified" name="filter" type="radio" checked="checked"/>
<label for="verified" onclick="">Verified</label>
<!-- <input id="minimal" name="compactness" type="radio"/>-->
<!-- <label for="minimal" onclick="">Minimal</label>-->
</div>
</div>
<div class="container" style="text-align: center; margin-bottom: 10px; margin-top: -10px;">
<small id="alpaca_eval_info" style="color: #777;">
Baseline: GPT-4 Preview (11/06) | Auto-annotator: GPT-4 Preview (11/06)
</small>
</div>
<table id="leaderboard">
<tr>
<th class="rank">Rank</th>
<th class="name">Model Name</th>
<th class="lenWinRate">LC Win Rate</th>
<th class="winRate">Win Rate</th>
</tr>
</table>
<div id="documentation">
<div style="text-align: center;">
<a href="https://github.com/tatsu-lab/alpaca_eval" style="display: inline-block;">
<i class="fab fa-fw fa-github" aria-hidden="true"></i> Github
</a>
</div>
<br>
<h2>About AlpacaEval</h2>
<p>
<a href="https://github.com/tatsu-lab/alpaca_eval" target="_blank">AlpacaEval</a>
an LLM-based automatic evaluation that is fast, cheap, and reliable.
It is based on the
<a href="https://crfm.stanford.edu/2023/05/22/alpaca-farm.html">AlpacaFarm</a>
evaluation set,
which tests the ability of models to follow general user instructions.
These responses are then compared to reference responses (Davinci003 for AlpacaEval, GPT-4 Preview for AlpacaEval 2.0) by
the provided GPT-4 based auto-annotators,
which results in the win rates presented above.
AlpacaEval displays a high agreement rate with ground truth human annotations,
and leaderboard rankings on AlpacaEval are very correlated with leaderboard rankings
based on human annotators.
Please see our
<a href="https://github.com/tatsu-lab/alpaca_eval#analysis" target="_blank">documentation</a>
for more details on our analysis.
</p>
<h2>Adding new models</h2>
<p>
We welcome new model contributions to the leaderboard from the community!
To do so, please follow the steps in the
<a href="https://github.com/tatsu-lab/alpaca_eval#contributing" target="_blank">contributions
section</a>.
Specifically, you'll need to run the model on the evaluation set,
auto-annotate the outputs, and submit a PR with the model config and leaderboard results.
We've also set up a
<a href="https://discord.gg/GJMxJSVZZM" target="_blank">Discord</a>
for community support and discussion.
</p>
<h2>Adding new evaluators or eval sets </h2>
<p>
We also welcome contributions for new evaluators or new eval sets!
For making new evaluators, we release our ground-truth
<a href="https://github.com/tatsu-lab/alpaca_eval#data-release" target="_blank">human annotations</a>
and <a href="https://github.com/tatsu-lab/alpaca_eval#analyzing-an-evaluator" target="_blank">comparison
metrics</a>.
We also release a
<a href="https://github.com/tatsu-lab/alpaca_eval#analyzing-an-eval-set" target="_blank">rough guide</a>
to follow for making new eval sets.
We specifically encourage contributions for harder instructions distributions and for safety testing of
LLMs.
</p>
<h2>AlpacaEval limitations</h2>
<p>
While AlpacaEval provides a useful comparison of model capabilities in following instructions,
it is not a comprehensive or gold-standard evaluation of model abilities.
For one, as detailed in the
<a href="https://arxiv.org/abs/2305.14387" target="_blank">AlpacaFarm paper</a>,
the auto annotator winrates are correlated with length.
Though human annotations also display this bias,
it is unclear if more verbose answers add utility in downstream tasks.
Additionally, the AlpacaFarm eval set, though diverse, consists mainly of simple instructions.
We encourage the community to contribute new, more complex eval sets, such as for tool use.
Finally, AlpacaEval does not evaluate the safety of any of the models.
</p>
</div>
</div>
<script>
const alpacaEvalRadio = document.getElementById('alpaca_eval');
const alpacaEval2Radio = document.getElementById('alpaca_eval_2');
const communityRadio = document.getElementById('community');
const verifiedRadio = document.getElementById('verified');
// const minimalRadio = document.getElementById('minimal');
const table = document.getElementById('leaderboard');
const urls = {
'alpaca_eval': 'https://raw.githubusercontent.com/tatsu-lab/alpaca_eval/main/docs/data_AlpacaEval/alpaca_eval_gpt4_leaderboard.csv',
'alpaca_eval_2': 'https://raw.githubusercontent.com/tatsu-lab/alpaca_eval/main/docs/data_AlpacaEval_2/weighted_alpaca_eval_gpt4_turbo_leaderboard.csv',
// 'claude': 'https://raw.githubusercontent.com/tatsu-lab/alpaca_eval/main/docs/claude_leaderboard.csv',
}
let currentUrl = urls['alpaca_eval_2'];
function updateTable(url) {
while (table.rows.length > 1) {
table.deleteRow(1);
}
Papa.parse(url, {
download: true,
header: true,
complete: function (results) {
console.log(results.data);
let rank = 0; // Initialize rank counter
results.data.forEach(row => {
if (row['name'] || row['win_rate'] || row['length_controlled_winrate']) { //|| row['avg_length']
let filter = row['filter'];
if ((communityRadio.checked && (filter === 'verified' || filter === 'minimal' || filter === 'community')) ||
(verifiedRadio.checked && (filter === 'verified' || filter === 'minimal'))) {
const tr = document.createElement('tr');
const rankTd = document.createElement('td');
const nameTd = document.createElement('td');
const winRateTd = document.createElement('td');
//const lengthTd = document.createElement('td');
const lenWinRateTd = document.createElement('td');
rankTd.classList.add('rank');
nameTd.classList.add('name');
winRateTd.classList.add('winRate');
lenWinRateTd.classList.add('lenWinRate');
// Set the rank value
rank++;
rankTd.textContent = rank;
if (row['link'] && row['link'].trim() !== '') {
const a = document.createElement('a');
a.textContent = row['name'];
a.href = row['link'];
a.target = "_blank";
nameTd.appendChild(a);
} else {
nameTd.textContent = row['name'];
}
if (row['samples'] && row['samples'].trim() !== '') {
const samplesLink = document.createElement('a');
samplesLink.textContent = " 📄"; // adding a space before emoji to separate from name
samplesLink.href = row['samples'];
samplesLink.target = "_blank";
samplesLink.style.textDecoration = "none";
nameTd.appendChild(samplesLink);
}
winRateTd.textContent = Number(row['win_rate']).toFixed(1) + '%';
if (row['length_controlled_winrate'] === '') {
lenWinRateTd.textContent = 'N/A';
} else {
lenWinRateTd.textContent = Number(row['length_controlled_winrate']).toFixed(1) + '%';
}
//lenWinRateTd.textContent = Number(row['length_controlled_winrate']).toFixed(1) + '%';
//lengthTd.textContent = Math.round(Number(row['avg_length'])).toString() ;
tr.appendChild(rankTd);
tr.appendChild(nameTd);
tr.appendChild(lenWinRateTd);
tr.appendChild(winRateTd);
//tr.appendChild(lengthTd);
table.appendChild(tr);
}
}
});
}
});
}
function updateInfoMessage(version) {
let infoText;
if (version === 'alpaca_eval_2') {
infoText = 'Baseline: GPT-4 Preview (11/06) | Auto-annotator: GPT-4 Preview (11/06)';
} else if (version === 'alpaca_eval') {
infoText = 'Baseline: Davinci003 | Auto-annotator: GPT-4';
}
document.getElementById('alpaca_eval_info').innerHTML = infoText;
}
updateTable(urls['alpaca_eval_2']);
alpacaEval2Radio.addEventListener('click', function () {
currentUrl = urls['alpaca_eval_2'];
updateTable(currentUrl);
updateInfoMessage('alpaca_eval_2');
});
alpacaEvalRadio.addEventListener('click', function () {
currentUrl = urls['alpaca_eval'];
updateTable(currentUrl);
updateInfoMessage('alpaca_eval');
});
communityRadio.addEventListener('click', function () {
updateTable(currentUrl);
});
verifiedRadio.addEventListener('click', function () {
updateTable(currentUrl);
});
// minimalRadio.addEventListener('click', function () {
// updateTable(currentUrl);
// });
updateCautionMessage('alpaca_eval_2');
</script>
</body>
</html>