-
Notifications
You must be signed in to change notification settings - Fork 28
/
Copy pathindex.html
467 lines (426 loc) · 19 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
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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>🪄 Lumos: Language Agents with Unified Data Formats, Modular Design, and Open-Source LLMs</title>
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-PYVRSFMDRL"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag('js', new Date());
gtag('config', 'G-PYVRSFMDRL');
</script>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
rel="stylesheet">
<link rel="stylesheet" href="./static/css/bulma.min.css">
<link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="./static/css/bulma-slider.min.css">
<link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
<link rel="stylesheet"
href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="./static/css/index.css">
<link rel="icon" href="./static/images/dynosaur.png">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="./static/js/fontawesome.all.min.js"></script>
<script src="./static/js/bulma-carousel.min.js"></script>
<script src="./static/js/bulma-slider.min.js"></script>
<script src="./static/js/index.js"></script>
</head>
<body>
<nav class="navbar" role="navigation" aria-label="main navigation">
<div class="navbar-brand">
<a role="button" class="navbar-burger" aria-label="menu" aria-expanded="false">
<span aria-hidden="true"></span>
<span aria-hidden="true"></span>
<span aria-hidden="true"></span>
</a>
</div>
<!-- <div class="navbar-menu">
<div class="navbar-start" style="flex-grow: 1; justify-content: center;">
<a class="navbar-item" href="https://keunhong.com">
<span class="icon">
<i class="fas fa-home"></i>
</span>
</a>
<div class="navbar-item has-dropdown is-hoverable">
<a class="navbar-link">
More Research
</a>
<div class="navbar-dropdown">
<a class="navbar-item" href="https://hypernerf.github.io">
HyperNeRF
</a>
<a class="navbar-item" href="https://nerfies.github.io">
Nerfies
</a>
<a class="navbar-item" href="https://latentfusion.github.io">
LatentFusion
</a>
<a class="navbar-item" href="https://photoshape.github.io">
PhotoShape
</a>
</div>
</div>
</div>
</div> -->
</nav>
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-2 publication-title">🪄 Lumos: Learning Agents with Unified Data, Modular Design, and Open-Source LLMs</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://wadeyin9712.github.io/">Da Yin</a><sup>2</sup>,</span>
<span class="author-block">
<a href="https://fabrahman.github.io/">Faeze Brahman</a><sup>1</sup>,</span>
<span class="author-block">
<a href="https://www.cs.cmu.edu/~aravicha/">Abhilasha Ravichander</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://www.cs.cmu.edu/~kchandu/">Khyathi Chandu</a><sup>1</sup>,
</span>
<br>
<span class="author-block">
<a href="http://web.cs.ucla.edu/~kwchang/">Kai-Wei Chang</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://homes.cs.washington.edu/~yejin/">Yejin Choi</a><sup>1, 3</sup>,
</span>
<span class="author-block">
<a href="https://yuchenlin.xyz/">Bill Yuchen Lin</a><sup>1</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>Allen Institute for AI</span>
<span class="author-block"><sup>2</sup>University of California, Los Angeles </span>
<span class="author-block"><sup>3</sup>University of Washington</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- PDF Link. -->
<span class="link-block">
<a href="https://arxiv.org/abs/2311.05657"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Paper</span>
</a>
</span>
<!-- Model Link. -->
<span class="link-block">
<a href="https://huggingface.co/models?sort=trending&search=ai2lumos"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>Model</span>
</a>
</span>
<!-- Code Link. -->
<span class="link-block">
<a href="https://github.com/allenai/lumos"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
<!-- Dataset Link. -->
<span class="link-block">
<a href="https://huggingface.co/datasets?sort=trending&search=ai2lumos"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="far fa-images"></i>
</span>
<span>Data</span>
</a>
</span>
<span class="link-block">
<a href="https://twitter.com/billyuchenlin/status/1724519911614300390?s=61&t=FyWExVfnr8YanWV5YzaPjg"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-twitter"></i>
</span>
<span>Twitter</span></a>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="content has-text-justified">
We introduce 🪄 <strong>Lumos</strong>, Language Agents with Unified Data Formats, Modular Design, and Open-Source LLMs. <strong>Lumos</strong> unifies a suite of complex interactive tasks and achieves competitive performance with GPT-4/3.5-based and larger open-source agents.
<br><br><strong>Lumos</strong> has following features:
<ul>
<li><strong>🧩 A General Agent Modular Framework</strong>
<ul>
<li>🧩 <strong>Lumos</strong> consists of planning, grounding, and execution modules built based on LLAMA-2-7B and off-the-shelf APIs.</li>
<li>🤗 <strong>Lumos</strong> utilizes a unified data format that encompasses multiple task types, thereby enabling the developed agent framework to conveniently support a range of interactive tasks.</li>
</ul>
</li>
<li><strong>🌍 Diverse Training Data</strong>
<ul>
<li>🌍 <strong>Lumos</strong> is trained with ~40K diverse high-quality subgoal/action annotations from ground-truth reasoning steps in existing benchmarks with GPT-4.</li>
<li>⚒️ <strong>Lumos</strong> data can be instrumental for future research in developing open-source agents for complex interactive tasks.</li>
</ul>
</li>
<li><strong>🚀 Competitive Performance</strong>
<ul>
<li>🚀 <strong>Lumos</strong> is comparable or even beats GPT-series agents on web/complex QA tasks Mind2Web and HotpotQA, and larger open agents on math tasks.</li>
<li>🚀 <strong>Lumos</strong> exceeds contemporaneous agents that have been fine-tuned with in-domain HotpotQA and Mind2Web annotations, such as FiReAct and AgentLM.</li>
<li>🚀 <strong>Lumos</strong> outperforms open agent baseline formulations like chain-of-thoughts and integrated training.</li>
<li>🚀 <strong>Lumos</strong> surpasses larger open LLM agents and domain-specific agents by a large margin on an unseen task, WebShop.</li>
</ul>
</li>
</ul>
</div>
</div>
<!--/ Abstract. -->
<!-- Paper video. -->
<!-- <div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Video</h2>
<div class="publication-video">
<iframe src="https://www.youtube.com/embed/MrKrnHhk8IA?rel=0&showinfo=0"
frameborder="0" allow="autoplay; encrypted-media" allowfullscreen></iframe>
</div>
</div>
</div> -->
<!--/ Paper video. -->
</div>
</section>
<section class="section" id="BibTeX">
<div class="container is-max-desktop content is-size-5">
<h2 class="title">BibTeX</h2>
<pre><code>
@article{yin2023lumos,
title={{Lumos: Learning Agents with Unified Data, Modular Design, and Open-Source LLMs}},
author={Yin, Da and Brahman, Faeze and Ravichander, Abhilasha and Chandu, Khyathi and Chang, Kai-Wei and Choi, Yejin and Lin, Bill Yuchen},
journal={arXiv preprint arXiv:2311.05657},
year={2023}
}
</code></pre>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Results. -->
<div class="columns is-centered">
<div class="column has-text-centered">
<h2 class="title is-3">🪄 Lumos Architecture</h2>
<p align="center">
<img src="static/images/lumos_architecture.png" width="600" align="middle" class="center"/>
</p>
<div class="content has-text-justified">
<strong>Lumos</strong> consists of following modules:
<ul>
<li><strong>Planning Module:</strong>
<ul>
<li>Decompose a complex task into a series of high-level subgoals, which are written in natural language.</li>
</ul>
<li><strong>Grounding Module:</strong>
<ul>
<li>Convert the high-level subgoals produced by the planning module to low-level executable actions.</li>
</ul>
<li><strong>Execution Module:</strong>
<ul>
<li>Parse actions to a series of external tools including APIs, small neural models, and virtual simulators that interact with relevant tools and external environment.</li>
</ul>
</ul>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Results. -->
<div class="columns is-centered">
<div class="column has-text-centered">
<h2 class="title is-3">🪄 Lumos Formulation</h2>
<p align="center">
<img src="static/images/lumos_formulation.png" class="center">
</p>
<div class="content has-text-justified">
We attempt the following two <strong>Lumos</strong> formulations:
<ul>
<li><strong>Lumos-Iterative (Lumos-I):</strong>
<ul>
<li>Generates one subgoal and its corresponding executable actions in each iteration according to the external environment and prior memory.</li>
<li>When generating the current t-th subgoal, the planning module requires the previous planned subgoals and the execution results of their grounded actions.</li>
</ul>
<li><strong>Lumos-Onetime (Lumos-O):</strong>
<ul>
<li>An efficient formulation that generates all the subgoals and grounded actions at once.</li>
</ul>
</ul>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Results. -->
<div class="columns is-centered">
<div class="column has-text-centered">
<h2 class="title is-3">🪄 Lumos Training Annotations</h2>
<p align="center">
<img src="static/images/lumos_annotation_example.png" class="center">
</p>
<div class="content has-text-justified">
<p>
Instead of using Self-Instruct method, we use LLMs to convert ground-truth intermediate reasoning steps into the expected high-quality annotations aligning with our proposed formulations.
</p>
<p>
Finally, we are able to generate ~40K annotations to train Lumos planning and grounding modules (one of the largest resources for language agent fine-tuning). The annotation sources cover web, complex QA and math task types. See our final annotation data in <a href="https://huggingface.co/datasets?search=ai2lumos">Huggingface Dataset</a> and prompt details in <a href="https://github.com/allenai/lumos/tree/main/data">Github</a>.
</p>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Results. -->
<div class="columns is-centered">
<div class="column has-text-centered">
<h2 class="title is-3">Results</h2>
<div class="content has-text-justified">
<p align="center">
<img src="static/images/lumos_results_1.png" width="80%" class="center">
</p>
<br>
<p>
We first evaluate <strong>Lumos</strong> on complex QA, web and maths tasks.
</p>
<p>
We find that <strong>Lumos</strong> outperforms GPT-4/3.5-based agents on complex QA and web tasks.
In particular, <strong>Lumos</strong> outperforms GPT-4 5.1 step success rate on Mind2Web and GPT-3.5-turbo-based ReAct 5.1 LLM accuracy. <strong>Lumos</strong> also achieves better performance than 2-4x bigger language agents on maths tasks.
</p>
</div>
</div>
</div>
</div>
<!--/ Results. -->
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Results. -->
<div class="columns is-centered">
<div class="column has-text-centered">
<h2 class="title is-3">Comparison with Baseline Formulations</h2>
<div class="content has-text-justified">
<p align="center">
<img src="static/images/lumos_results_2.png" class="center">
</p>
<p>
We compare <strong>Lumos</strong> formulation with other baseline formulations to train open-source agents. The baseline formulations are Chain-of-Thought Training
and Integrated Agent Training.
</p>
<p>
<strong>Lumos</strong> performs the best among the baselines on three different complex interative tasks.
</p>
</div>
</div>
</div>
</div>
<!--/ Results. -->
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Results. -->
<div class="columns is-centered">
<div class="column has-text-centered">
<h2 class="title is-3">Generalizability of Lumos</h2>
<div class="content has-text-justified">
<p align="center">
<img src="static/images/lumos_results_3.png" width="250" class="center">
</p>
<p>
We first evaluate <strong>Lumos</strong> trained with the unified annotations composed by task-specific ones. We then test <strong>Lumos</strong> on an unseen complex interactive task, WebShop.
</p>
<p>
We find that after the unified training, <strong>Lumos</strong> would have slightly higher
performance on web and complex QA tasks. We also observe that <strong>Lumos</strong> can bring an improvement over domain-specific agents 5-10 reward improvement, and also better performance than larger agents with 13B and 30B sizes.
</p>
</div>
</div>
</div>
</div>
<!--/ Results. -->
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Results. -->
<div class="columns is-centered">
<div class="column has-text-centered">
<h2 class="title is-3">Further Analysis on Annotations</h2>
<div class="content has-text-justified">
<p align="center">
<img src="static/images/lumos_results_4.png" width="400" class="center">
</p>
<p>
We also conduct deeper analysis about annotation quality and the choice of annotation formats. We answer the following questions:
<ul>
<li><strong>Q1: How good is our converted training annotations?</strong></li>
<li><strong>Q2: Would it be better if we adopt low-level subgoals instead of our proposed high-level subgoals?</strong></li>
</ul>
</p>
<p>
We find that by controling the training annotation size to be the same, our annotations can still help get better performance than the ones produced by Self-Instruct method and passed by rigorous
execution sanity checking. Also, we find that making planning module generate high-level subgoals would be a superior choice to generating a very long sequence of low-level subgoals.
</p>
</div>
</div>
</div>
</div>
<!--/ Results. -->
</section>
<footer class="footer">
<div class="container">
<div class="content has-text-centered">
<a class="icon-link"
href="">
<i class="fas fa-file-pdf"></i>
</a>
<a class="icon-link" href="https://github.com/allenai/lumos" class="external-link" disabled>
<i class="fab fa-github"></i>
</a>
</div>
<div class="columns is-centered">
<div class="column is-8">
<div class="content">
<p>
This website is licensed under a <a rel="license"
href="http://creativecommons.org/licenses/by-sa/4.0/">Creative
Commons Attribution-ShareAlike 4.0 International License</a>.
</p>
<p>
This means you are free to borrow the <a
href="https://github.com/nerfies/nerfies.github.io">source code</a> of this website,
we just ask that you link back to this page in the footer.
Please remember to remove the analytics code included in the header of the website which
you do not want on your website.
</p>
</div>
</div>
</div>
</div>
</footer>
</body>
</html>