forked from LLM360/website
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathresearch.html
More file actions
629 lines (611 loc) · 34.1 KB
/
research.html
File metadata and controls
629 lines (611 loc) · 34.1 KB
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
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
<!DOCTYPE HTML>
<html lang="en">
<head>
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-0FFN6N7318"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-0FFN6N7318');
</script>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1, user-scalable=no" />
<!-- SEO -->
<title>LLM360 Research Suite | Comprehensive LLM Artifacts</title>
<meta name="description" content="Dive deep into LLM training dynamics with the LLM360 Research Suite. Access detailed resources including training codes, data, and dynamic evaluations of models like Amber, Crystal, and K2. Perfect for researchers and developers interested in advanced LLM training insights."/>
<link rel="canonical" href="https://www.llm360.ai/research.html" />
<meta name="keywords" content="LLM360, LLM Research, LLM Training Dynamics, Advanced AI Training, Open Source LLMs, AI Artifacts, Amber LLM, CrystalCode, K2 Model, Training Checkpoints, LLM Evaluation"/>
<meta name="author" content="LLM360"/>
<meta name="robots" content="index, follow"/>
<!-- Open Graph Protocol -->
<meta property="og:title" content="LLM360 Research Suite | Comprehensive LLM Artifacts"/>
<meta property="og:description" content="Dive deep into LLM training dynamics with the LLM360 Research Suite. Access detailed resources including training codes, data, and dynamic evaluations of models like Amber, Crystal, and K2. Perfect for researchers and developers interested in advanced LLM training insights."/>
<meta property="og:type" content="website"/>
<meta audio property="og:url" content="https://www.llm360.ai/research.html/"/>
<meta property="og:image" content="https://www.llm360.ai/images/open-graph-image.jpg"/>
<!-- Twitter Card -->
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:site" content="@llm360">
<meta name="twitter:title" content="LLM360 Research Suite | Comprehensive LLM Artifacts">
<meta name="twitter:description" content="Dive deep into LLM training dynamics with the LLM360 Research Suite. Access detailed resources including training codes, data, and dynamic evaluations of models like Amber, Crystal, and K2. Perfect for researchers and developers interested in advanced LLM training insights.">
<meta name="twitter:image" content="https://www.llm360.ai/images/open-graph-image.jpg">
<!-- Schema Markup -->
<script type="application/ld+json">
{
"@context": "http://schema.org",
"@type": "WebSite",
"name": "LLM360 Research Suite",
"url": "https://www.llm360.ai/research.html",
"description": "Dive deep into LLM training dynamics with the LLM360 Research Suite. Access detailed resources including training codes, data, and dynamic evaluations of models like Amber, Crystal, and K2. Perfect for researchers and developers interested in advanced LLM training insights.",
"publisher": {
"@type": "Organization",
"name": "LLM360",
"sameAs": [
"https://x.com/llm360",
"https://github.com/LLM360"
]
},
"image": {
"@type": "ImageObject",
"url": "https://www.llm360.ai/images/open-graph-image.jpg",
"width": 1920,
"height": 1080
}
}
</script>
<!-- /SEO -->
<!-- Favicon -->
<link rel="icon" type="image/x-icon" href="assets/favicon/favicon.ico" />
<link rel="icon" type="image/png" sizes="192x192" href="assets/favicon/android-chrome-192x192.png">
<link rel="icon" type="image/png" sizes="512x512" href="assets/favicon/android-chrome-512x512.png">
<link rel="icon" type="image/png" sizes="32x32" href="assets/favicon/favicon-32x32.png">
<link rel="icon" type="image/png" sizes="16x16" href="assets/favicon/favicon-16x16.png">
<link rel="apple-touch-icon" sizes="180x180" href="assets/favicon/apple-touch-icon.png">
<!-- <link rel="manifest" href="/site.webmanifest"> -->
<meta name="msapplication-TileColor" content="#da532c">
<meta name="theme-color" content="#ffffff">
<!-- Style -->
<link rel="stylesheet" href="assets/css/main.css" />
</head>
<body class="is-preload">
<!-- Sidebar -->
<button class="toggle-btn" id="toggleBtn">☰</button>
<section id="sidebar" class="sidebar">
<div class="inner">
<nav>
<a class="alt" href="index.html">
<figure class="hover-rotate">
<img src="images/logo-highres.png" alt="logo" />
</figure>
</a>
<h2>LLM360</h2>
<ul>
<li><a href="index.html#one">Models</a></li>
<li><a href="evaluation.html">Performance and Evaluation</a></li>
<li><a href="index.html#two" >LLM360 Suites</a></li>
<li><a href="index.html#three">Papers</a></li>
<li><a href="index.html#four">Blogs</a></li>
<li><a href="index.html#six">Get in touch</a></li>
<li><a href="community.html">Open-source Communities</a></li>
<li><a href="about.html">About</a></li>
</ul>
</nav>
</div>
</section>
<!-- Wrapper -->
<div id="wrapper">
<!-- Intro -->
<section id="top" class="wrapper fullscreen fade-up">
<div class="inner">
<h1>LLM360 Research Suite</h1>
<p>LLM360 Research Suite is a comprehensive set of large language model (LLM) artifacts from each of our models, for academic and industry researchers to explore LLM training dynamics.</p>
</div>
</section>
<!-- One -->
<section id="one" class="wrapper style3 fullscreen fade-up">
<div class="inner">
<h2>Training Dynamics Collection</h2>
<div class="features three">
<section>
<h3>Amber</h3>
<ul>
<li><a href="https://huggingface.co/LLM360/Amber/tree/main" target="_blank">360 Checkpoints</a></li>
<li><a href="https://github.com/LLM360/amber-train" target="_blank">Training Code</a></li>
<li><a href="https://huggingface.co/datasets/LLM360/AmberDatasets" target="_blank">Data (dataset, code, sequence)</a></li>
<li><a href="https://wandb.ai/llm360/Amber?nw=nwuseraurickqiao" target="_blank">WandB</a></li>
</ul>
</section>
<section>
<h3>Crystal</h3>
<ul>
<li><a href="https://huggingface.co/LLM360/CrystalCoder/tree/main" target="_blank">120 Checkpoints</a></li>
<li><a href="https://github.com/LLM360/crystalcoder-train" target="_blank">Training Code</a></li>
<li><a href="https://huggingface.co/datasets/LLM360/CrystalCoderDatasets" target="_blank">Data (dataset, code, sequence)</a></li>
<li><a href="https://wandb.ai/llm360/CrystalCoder?nw=hdze3lfpuer" target="_blank">WandB</a></li>
</ul>
</section>
<section>
<h3>K2</h3>
<ul>
<li><a href="https://huggingface.co/LLM360/K2/tree/main" target="_blank">120 Checkpoints</a></li>
<li><a href="https://github.com/LLM360/k2-train" target="_blank">Training Code</a></li>
<li><a href="https://huggingface.co/datasets/LLM360/K2Datasets" target="_blank">Data (dataset, code, sequence)</a></li>
<li><a href="https://wandb.ai/llm360/K2?nw=29mu6l0zzqq" target="_blank">WandB</a></li>
<li><a href="https://huggingface.co/spaces/LLM360/k2-gallery" target="_blank">K2 Prompt Gallery</a></li>
<li><a href="https://huggingface.co/spaces/LLM360/k2-eval-gallery" target="_blank">K2 Evaluation Gallery</a></li>
</ul>
</section>
</div>
</div>
</section>
<!-- Two -->
<section id="two" class="wrapper style4 fade-up">
<div class="inner">
<h2>K2 Spikes</h2>
<p> We encountered two major loss spikes while training K2:
<ul>
<li>The first loss spike occurred after 160 checkpoints and lasted over ~34 checkpoints. We restarted training at checkpoint 160 and training returned to normal.</li>
<li>The second loss spike occurred after restarting training to fix the first loss spike at checkpoint 186 and lasted from ~8 checkpoints. Again, we restarted training at checkpoint 186 and training returned to normal.</li>
</ul>
We are releasing these checkpoints so others can study this interesting phenomena in large model training.</p>
<ul>
<li>Spike 1: 34 checkpoints, <a href="https://wandb.ai/llm360/K2?nw=7bxe4sz0vv" target="_blank">WandB</a></li>
<li>Spike 2: 8 checkpoints, <a href="https://wandb.ai/llm360/K2?nw=inng96ujjmr" target="_blank">WandB</a></li>
</ul>
<span class="image fit">
<img src="images/spike_k2.png"/>
</span>
</div>
</section>
<!-- Three -->
<section id="three" class="wrapper fade-up">
<div class="inner">
<h2>Analysis360: Analyze LLMs in 360 degrees</h2>
<p>Analysis360 serves as the single source of truth for all evaluation metrics and provides in-depth analysis from many different angles.</p>
<section>
<h3>Summary of notable open-source LLMs</h3>
<p>
We note a trend of progressively less disclosure of important
pretraining details over time: (1) availability of pretraining code, (2) disclosure of training configurations and
hyperparameters, (3) intermediate checkpoints of model weights, (4) intermediate checkpoints of optimizer
states, (5) disclosure of data mixture and sources, (6) reproducibility of pretraining data sequence, and (7)
availability (or reconstruction scripts) of the pretraining data.
</p>
<table>
<thead>
<tr>
<th>LLM Name</th>
<th>Release Date</th>
<th>Pretraining</th>
<th></th>
<th>Checkpoints</th>
<th></th>
<th>Pretraining Dataset</th>
<th></th>
<th></th>
<th>Tokens</th>
</tr>
<tr>
<th>Name</th>
<th>Date</th>
<th>Code</th>
<th>Config</th>
<th>Model</th>
<th>Optim</th>
<th>Data Mix</th>
<th>Ordering</th>
<th>Available</th>
<th>(T )</th>
</tr>
</thead>
<tbody>
<tr class="alt">
<td>K2</td>
<td>May’24</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>1.4</td>
</tr>
<tr>
<td>OLMo-7B</td>
<td>May’24</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>2.5</td>
</tr>
<tr>
<td>Arctic</td>
<td>Apr’24</td>
<td></td>
<td></td>
<td>✓</td>
<td>✓</td>
<td></td>
<td></td>
<td></td>
<td>1.5</td>
</tr>
<tr class="alt">
<td>CrystalCoder</td>
<td>Dec’23</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>1.4</td>
</tr>
<tr class="alt">
<td>Amber</td>
<td>Dec’23</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>1.3</td>
</tr>
<tr>
<td>Yi</td>
<td>Nov’23</td>
<td></td>
<td></td>
<td></td>
<td></td>
<td></td>
<td></td>
<td></td>
<td>?</td>
</tr>
<tr>
<td>Mistral</td>
<td>Sep’23</td>
<td></td>
<td></td>
<td></td>
<td></td>
<td></td>
<td></td>
<td></td>
<td>?</td>
</tr>
<tr>
<td>Qwen</td>
<td>Aug’23</td>
<td></td>
<td>✓</td>
<td></td>
<td></td>
<td></td>
<td></td>
<td></td>
<td>2.4</td>
</tr>
<tr>
<td>Llama 2</td>
<td>Jul’23</td>
<td></td>
<td>✓</td>
<td></td>
<td></td>
<td></td>
<td></td>
<td></td>
<td>2.0</td>
</tr>
<tr>
<td>Falcon</td>
<td>May’23</td>
<td></td>
<td>✓</td>
<td></td>
<td></td>
<td>✓</td>
<td></td>
<td></td>
<td>1.5</td>
</tr>
<tr>
<td>MPT</td>
<td>May’23</td>
<td>✓</td>
<td>✓</td>
<td></td>
<td></td>
<td>✓</td>
<td></td>
<td></td>
<td>1.0</td>
</tr>
<tr>
<td>INCITE</td>
<td>May’23</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td></td>
<td>✓</td>
<td></td>
<td>✓</td>
<td>1.0</td>
</tr>
<tr>
<td>OpenLLama</td>
<td>May’23</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td></td>
<td>✓</td>
<td></td>
<td>✓</td>
<td>1.0</td>
</tr>
<tr>
<td>LLAMA</td>
<td>Feb’23</td>
<td></td>
<td>✓</td>
<td></td>
<td></td>
<td>✓</td>
<td></td>
<td></td>
<td>1.0</td>
</tr>
<tr>
<td>Pythia</td>
<td>Feb’23</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>0.30</td>
</tr>
<tr>
<td>BLOOM</td>
<td>Nov’22</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>0.34</td>
</tr>
<tr>
<td>OPT</td>
<td>May’22</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td></td>
<td>✓</td>
<td></td>
<td></td>
<td>0.18</td>
</tr>
<tr>
<td>GPT-NeoX</td>
<td>Apr’22</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>0.40</td>
</tr>
<tr>
<td>GPT-J</td>
<td>May’21</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>✓</td>
<td>0.40</td>
</tr>
</tbody>
</table>
<br>
</section>
<section>
<h3>Our Approach</h3>
<p>We run evaluations on a variety of benchmarks, including the conventional benchmarks like MMLU, Hellaswag, ARC, user-preference aligned benchmarks like MT-bench, long-context evaluations like LongEval, and additional studies on safety benchmarks for truthfulness, toxicity, and bias. Moreover, we report results on the model samples we preselected from a suite of LLMs where they all trained on same data seen in the exact same order to better observe and understand how our models develop and evolve over the training process. We also provide public access to all checkpoints, all code and all wandb dashboards for detailed training and evaluation curves.</p>
<br>
</section>
<section>
<h3>List of Analysis and Metrics</h3>
<p>Here's a full list of analysis/metrics we have collected so far. For each model we release, at this point, Amber, CrystalCoder, and K2, we put down the links to specific wandb reports if the evaluation is done. Amber, CrystalCoder, and K2 currently use their own evaluation scripts, we are working on consolidating these in the future, more details can be found in later sections. Please refer to model cards (<a href="https://huggingface.co/collections/LLM360/amber-65e7333ff73c7bbb014f2f2f">Amber</a>, <a href="https://huggingface.co/collections/LLM360/crystal-65e733d14e6a0786c4f5a606">CrystalCoder</a>, and <a href="https://huggingface.co/collections/LLM360/k2-6622ae6911e3eb6219690039">K2</a>) for any terms or technology you find unfamiliar. We will keep updating and expanding the list as our study proceeds, please stay tuned on the upcoming changes!</p>
<table class="alt">
<thead>
<tr>
<th>Metrics/Analysis</th>
<th>Description</th>
<th>Amber</th>
<th>CrystalCoder</th>
</tr>
</thead>
<tbody>
<tr>
<td><a href="https://arxiv.org/abs/2002.03029">mmlu</a></td>
<td>A test to measure a text model’s multitask accuracy. The test covers 57 tasks including elementary mathematics, US history, computer science, law, and more</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/mmlu-23-12-05-12-00-27---Vmlldzo2MTc1Njkx" target="_blank">5 shot</a></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/MMLU-0-shot-23-12-05-12-26-58---Vmlldzo2MTc1OTIw" target="_blank">0 shot</a><br><a href="https://wandb.ai/llm360/CrystalCoder/reports/MMLU-5-shot-23-12-05-12-31-30---Vmlldzo2MTc1OTgy" target="_blank">5 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/1806.03822">race</a></td>
<td>A test to measure reading comprehension ability</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/race-23-12-05-12-01-31---Vmlldzo2MTc1NzAw" target="_blank">0 shot</a></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/RACE-0-shot-23-12-05-12-27-47---Vmlldzo2MTc1OTI5" target="_blank">0 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/1803.05457">arc_challenge</a></td>
<td>A set of grade-school science questions</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/arc-23-12-05-12-02-08---Vmlldzo2MTc1NzA5" target="_blank">25 shot</a></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/ARC-C-0-shot-23-12-06-11-10-01---Vmlldzo2MTg3NjEz" target="_blank">0 shot</a><br><a href="https://wandb.ai/llm360/CrystalCoder/reports/ARC-C-25-shot-23-12-06-11-08-46---Vmlldzo2MTg3NjA0" target="_blank">25 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/1905.10044">boolq</a></td>
<td>A question answering dataset for yes/no questions containing 15942 examples</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/boolq-23-12-05-12-03-24---Vmlldzo2MTc1NzE3" target="_blank">0 shot</a></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/BoolQ-0-shot-23-12-05-12-28-19---Vmlldzo2MTc1OTM3" target="_blank">0 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/1905.07830">hellaswag</a></td>
<td>A test of commonsense inference</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/hellaswag-23-12-05-12-03-55---Vmlldzo2MTc1NzIw" target="_blank">10 shot</a></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/HellaSwag-0-shot-23-12-05-12-25-18---Vmlldzo2MTc1OTA0" target="_blank">0 shot</a>
<br><a href="https://wandb.ai/llm360/CrystalCoder/reports/HellaSwag-10-shot-23-12-05-12-47-16---Vmlldzo2MTc2MTAz" target="_blank">10 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/1809.02789" target="_blank">openbookqa</a></td>
<td>A question-answering dataset modeled after open book exams for assessing human understanding of a subject</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/openbookqa-23-12-05-12-04-39---Vmlldzo2MTc1NzI1" target="_blank">0 shot</a></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/Openbook-QA-0-shot-23-12-05-12-48-00---Vmlldzo2MTc2MTE0" target="_blank">0 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/1911.11641" target="_blank">piqa</a></td>
<td>A test to measure physical commonsense and reasoning</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/piqa-23-12-05-12-05-40---Vmlldzo2MTc1NzMy" target="_blank">0 shot</a></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/PIQA-0-shot-23-12-05-12-46-47---Vmlldzo2MTc2MTAx" target="_blank">0 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/1904.09728" target="_blank">siqa</a></td>
<td>A test to measure commonsense reasoning about social interactions</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/siqa-23-12-05-12-07-33---Vmlldzo2MTc1NzUw" target="_blank">0 shot</a></td>
<td></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/1907.10641" target="_blank">winogrande</a></td>
<td>An adversarial and difficult Winograd benchmark at scale, for commonsense reasoning</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/winogrande-23-12-05-12-08-04---Vmlldzo2MTc1NzU1" target="_blank">0 shot</a></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/Winogrande-0-shot-23-12-05-12-30-16---Vmlldzo2MTc1OTY5" target="_blank">0 shot</a><br><a href="https://wandb.ai/llm360/CrystalCoder/reports/Winogrande-5-shot-23-12-05-12-28-46---Vmlldzo2MTc1OTQ3" target="_blank"> 5 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/2010.00133" target="_blank">crowspairs</a></td>
<td>A challenge set for evaluating what language models (LMs) on their tendency to generate biased outputs</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/crowspairs-23-12-05-12-08-51---Vmlldzo2MTc1NzYz" target="_blank">0 shot</a></td>
<td></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/2109.07958" target="_blank">truthfulqa</a></td>
<td>A test to measure a model’s propensity to reproduce falsehoods commonly found online</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/truthfulqa-23-12-05-12-12-08---Vmlldzo2MTc1Nzg4" target="_blank">0 shot</a></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/Truthful-QA-0-shot-23-12-05-12-49-09---Vmlldzo2MTc2MTIx" target="_blank">0 shot</a></td>
</tr>
<tr>
<td><a href="https://pile.eleuther.ai/" target="_blank">pile</a></td>
<td>A test to measure model's perplexity, we covered 18/22 sub datasets</td>
<td><a href="https://wandb.ai/llm360/Amber/runs/ut4txpqk" target="_blank">perplexity</a></td>
<td></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/1903.00161" target="_blank">drop</a></td>
<td>A reading comprehension benchmark requiring discrete reasoning over paragraphs</td>
<td></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/DROP-3-shot-23-12-05-12-55-44---Vmlldzo2MTc2MTU1" target="_blank">3 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/2108.07732" target="_blank">mbpp</a></td>
<td>Around 1,000 crowd-sourced Python programming problems, designed to be solvable by entry-level programmers</td>
<td></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/MBPP-pass-1-t-0-1-23-12-05-12-42-33---Vmlldzo2MTc2MDcw" target="_blank">pass 1</a><br><a href="https://wandb.ai/llm360/CrystalCoder/reports/MBPP-pass-10-t-0-1-23-12-05-12-41-45---Vmlldzo2MTc2MDYy" target="_blank">pass 10</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/2107.03374" target="_blank">humaneval</a></td>
<td>A test to measure functional correctness for synthesizing programs from docstrings</td>
<td></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/HumanEval-pass-1-t-0-2-23-12-05-12-45-51---Vmlldzo2MTc2MDk0" target="_blank">pass 1</a><br><a href="https://wandb.ai/llm360/CrystalCoder/reports/HumanEval-pass-10-t-0-2-23-12-05-12-48-34---Vmlldzo2MTc2MTE2" target="_blank">pass 10</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/2110.14168" target="_blank">gsm8k</a></td>
<td>Diverse grade school math word problems to measure a model's ability to solve multi-step mathematical reasoning problems</td>
<td></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/GSM8K-5-shot-23-12-05-12-50-29---Vmlldzo2MTc2MTI4" target="_blank">5 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/2203.08398" target="_blank">copa</a></td>
<td>A test to assess progress in open-domain commonsense causal reasoning</td>
<td></td>
<td><a href="https://wandb.ai/llm360/CrystalCoder/reports/COPA-0-shot-23-12-05-12-52-54---Vmlldzo2MTc2MTQy" target="_blank">0 shot</a></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/2203.09509" target="_blank">toxigen</a></td>
<td>A test to measure model's toxicity on text generation</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/toxigen-and-toxicity-identification-23-12-06-15-24-39---Vmlldzo2MTg5NTcy" target="_blank">toxigen</a></td>
<td></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/2305.13169" target="_blank">toxicity identification</a></td>
<td>A test to measure model's capability on identifying toxic text</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/toxigen-and-toxicity-identification-23-12-06-15-24-39---Vmlldzo2MTg5NTcy" target="_blank">toxicity identification</a></td>
<td></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/2101.11718" target="_blank">bold</a></td>
<td>A test to evaluate fairness in open-ended language generation in English language</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/bold-23-12-06-15-27-23---Vmlldzo2MTg5NTky" target="_blank">bold</a></td>
<td></td>
</tr>
<tr>
<td><a href="https://arxiv.org/abs/2202.07646" target="_blank">memorization and token orders analysis</a></td>
<td>An analysis to understand model's memorization abilities</td>
<td><a href="https://wandb.ai/llm360/Amber/reports/memorization-23-12-06-15-29-48---Vmlldzo2MTg5NjEx" target="_blank">memorization</a></td>
<td></td>
</tr>
</table>
</section>
</div>
</section>
</div>
<!-- Footer -->
<footer id="footer" class="wrapper style1-alt">
<div class="inner">
<ul class="menu">
<p>
LLM360, proudly sponsored by Petuum and MBZUAI, is dedicated to advancing the field of AI by providing comprehensive access to large language models.<br>
LLM360 enables community-owned AGI by creating standards and tools to advance the bleeding edge of LLM capability and empower knowledge transfer, research, and development.
</p>
<ul class="actions">
<li><a href="https://twitter.com/llm360" target="_blank" class="icon brands circle fa-twitter"><span class="label">Twitter</span></a></li>
<li><a href="https://github.com/LLM360" target="_blank" class="icon brands circle fa-github"><span class="label">Github</span></a></li>
<li><a href="mailto:team@llm360.ai" target="_blank" class="icon circle fa-envelope"><span class="label">Email</span></a></li>
</ul>
<p class="copyright">© LLM360 2023-2024. All rights reserved.</p>
</ul>
</div>
</footer>
<!-- Scripts -->
<script src="assets/js/jquery.min.js"></script>
<script src="assets/js/jquery.scrollex.min.js"></script>
<script src="assets/js/jquery.scrolly.min.js"></script>
<script src="assets/js/browser.min.js"></script>
<script src="assets/js/breakpoints.min.js"></script>
<script src="assets/js/util.js"></script>
<script src="assets/js/main.js"></script>
</body>
</html>