-
Notifications
You must be signed in to change notification settings - Fork 3
/
index.html
392 lines (355 loc) · 18.2 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
<!-- ----->
<!--permalink: /cocon/-->
<!-- ----->
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description"
content="Evaluating Very Long-Term Conversational Memory of LLM Agents">
<meta name="keywords" content="dialog, memory, locomo, llm-agent, conversational agent">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Evaluating Very Long-Term Conversational Memory of LLM Agents</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/favicon.svg">
<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>
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h2 class="title is-2 publication-title">Evaluating Very Long-Term Conversational Memory of LLM Agents</h2>
<!-- <h3 class="title is-3">Accepted to <a href="https://jmlr.org/tmlr/">TMLR</a> 02/2024</h3> -->
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://adymaharana.github.io/">Adyasha Maharana</a><sup>1</sup>,</span>
<span class="author-block">
<a href="https://www.danny-lee.info/">Dong-Ho Lee</a><sup>2</sup>,</span>
<span class="author-block">
<a href="http://www.stulyakov.com/">Sergey Tulyakov</a><sup>3</sup>,
</span>
<span class="author-block">
<a href="https://www.cs.unc.edu/~mbansal/">Mohit Bansal</a><sup>1†</sup>,
</span>
<span class="author-block">
<a href="https://fvancesco.github.io/">Francesco Barbieri</a><sup>†</sup>,
</span>
<span class="author-block">
<a href="https://yuwfan.github.io/">Yuwei Fang</a><sup>3†</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>UNC Chapel Hill</span>
<span class="author-block"><sup>2</sup>University of Southern California</span>
<span class="author-block"><sup>3</sup>Snap Inc.</span>
<span class="author-block"><sup>†</sup>Equal advising</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- PDF Link. -->
<!-- <span class="link-block">
<a href="https://openreview.net/forum?id=ue9igTDLN2"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Open Review</span>
</a>
</span> -->
<span class="link-block">
<a href="http://arxiv.org/abs/2402.17753"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>arXiv</span>
</a>
</span>
<!-- <!– Video Link. –>-->
<!-- <span class="link-block">-->
<!-- <a href="https://www.youtube.com/watch?v=MrKrnHhk8IA"-->
<!-- class="external-link button is-normal is-rounded is-dark">-->
<!-- <span class="icon">-->
<!-- <i class="fab fa-youtube"></i>-->
<!-- </span>-->
<!-- <span>Video</span>-->
<!-- </a>-->
<!-- </span>-->
<!-- Code Link. -->
<span class="link-block">
<a href="https://github.com/snap-research/LoCoMo"
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://github.com/snap-research/locomo/tree/main/data"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-clock"></i>
</span>
<span>Data</span>
</a>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<!-- <iframe src="./static/images/main_figure_arxiv.pdf" style="width: 100%;height: 100%;border: none;"></iframe>-->
<img src="./static/images/evaluation_framework.svg" alt="Logo" style="display:block;border: none;"/>
<!-- <video id="teaser" autoplay muted loop playsinline height="100%">-->
<!-- <source src="./static/images/main_figure.svg"-->
<!-- type="image/svg+xml">-->
<!-- </video>-->
<h2 class="subtitle has-text-centered">
<b>Overview of our evaluation framework.</b> We propose three tasks: question answering, event summarization and multimodal dialog generation
to evaluate models' comprehension in very long-term dialogues.
</h2>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<!-- Abstract. -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Motivation</h2>
<div class="content has-text-justified">
<p>
Existing works on long-term open-domain dialogues focus on evaluating model responses
within contexts spanning no more than five
chat sessions.
</p>
<p>
We introduce a machine-human
pipeline to generate high-quality, very long-term dialogues by leveraging LLM-based agent
architectures and grounding their dialogues on personas and temporal event graphs. Moreover,
we equip each agent with the capability of sharing and reacting to images.
The generated conversations are verified and edited by human annotators for long-range consistency and grounding to the event graphs.
Using this pipeline, we collect LoCoMo, a dataset of very long-term conversations, each encompassing 300
turns and 9K tokens on avg., over up to 35 sessions.
</p>
<p>
Based on LOCOMO, we present a comprehensive evaluation benchmark to measure long-term memory in models, encompassing
question answering, event summarization, and multi-modal dialogue generation tasks.
<!-- Our experimental results indicate that LLMs exhibit -->
<!-- challenges in understanding lengthy conversations and comprehending long-range temporal -->
<!-- and causal dynamics within dialogues. -->
</p>
</div>
</div>
</div>
<!--/ Abstract. -->
<!-- Motivation. -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">How do we generate <i>very</i> long-term conversations?</h2>
<div class="intro-figure">
<img src="./static/images/intro_figure_conv_only_v2.svg" alt="Intro" style="float: right; margin: 8px;"/>
<!-- <iframe src="https://www.youtube.com/embed/MrKrnHhk8IA?rel=0&showinfo=0"-->
<!-- frameborder="0" allow="autoplay; encrypted-media" allowfullscreen></iframe>-->
</div>
<div class="content has-text-justified">
<p>
An example of a conversation in LoCoMo is shown to the right.
<ul>
<li>We create two <b>virtual agents</b>, each initialized with a LLM.</li>
<li>To start, unique <b>persona statements</b> are assigned to each agent, ensuring the integration of distinct personalities into their dialogues.</li>
<li>To mirror real-life experiences, we create a <b>temporal event graph</b> for each agent, which illustrates a realistic sequence of life events.</li>
<li>The LLM <b>agent architecture</b> is utilized for each agent, enabling them to effectively memorize and reflect conversation history into ongoing dialogues.</li>
<li>Further, each agent can share coherent images, thereby enhancing the <b>multi-modal dialogue</b> aspect.</li>
<li>Finally, human annotators are tasked with <b>manually filtering and refining</b> the generated data.</li>
</ul>
</p>
</div>
</div>
</div>
<!--/ Paper video. -->
</div>
</section>
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<!-- <iframe src="./static/images/main_figure_arxiv.pdf" style="width: 100%;height: 100%;border: none;"></iframe>-->
<img src="./static/images/main_v4.svg" alt="Logo" style="margin-top:50px;"/>
<!-- <video id="teaser" autoplay muted loop playsinline height="100%">-->
<!-- <source src="./static/images/main_figure.svg"-->
<!-- type="image/svg+xml">-->
<!-- </video>-->
<h2 class="subtitle has-text-centered">
<b>Overview of the generative pipeline for LoCoMo.</b> Each LLM agent is assigned a distinct persona and
a timeline of causally connected events in their file. The agent is equipped with a memory and reflection module to
retrieve relevant history for dialog generation and is also enabled for image-sharing and image-reaction behaviors
(left). The generated conversations are edited by human annotators to maintain long-range consistency (right).
</h2>
</div>
</div>
</section>
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<!-- <iframe src="./static/images/main_figure_arxiv.pdf" style="width: 100%;height: 100%;border: none;"></iframe>-->
<img src="./static/images/events.svg" alt="Logo"/>
<!-- <video id="teaser" autoplay muted loop playsinline height="100%">-->
<!-- <source src="./static/images/main_figure.svg"-->
<!-- type="image/svg+xml">-->
<!-- </video>-->
<h2 class="subtitle has-text-centered">
<b>Temporal Event Graph Creation.</b> Each event is generated in accordance with the specified persona p
and causal connections l between events are depicted to illustrate the casual relationships among them.
</h2>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Evaluation Framework</h2>
<div class="content has-text-justified">
<p>
In this study, we present a holistic evaluation framework to assess an agent’s proficiency in managing and responding within long-term contexts.
<ul>
<li><b>Question Answering.</b> Agents need to “recall” past
context correctly to integrate relevant information
into future responses. We present a direct examination of their memory via a question answering
task. We classify questions into five distinct
reasoning types to evaluate memory from multiple perspectives: <i>single-hop</i>, <i>multi-hop</i>, <i>temporal</i>,
<i>commonsense or world knowledge</i>, and <i>adversarial</i>.</li>
<li><b>Event Graph Summarization.</b> Agents also need to recognize long-range causal and temporal connections in the dialogues to generate empathetic and relevant responses. We
propose a measurement of their causal and temporal understanding with an event graph summarization task where the event graphs
linked to each LLM speaker serve as the correct
answers, and models are tasked with extracting this
information from the conversation history.</li>
<li><b>Multi-modal Dialog Generation.</b> Conversational agents need to utilize relevant context recalled from past conversations to generate
responses that are consistent with the ongoing narrative. We assess this ability via the multi-modal
dialog generation task.</li>
</ul>
</p>
</div>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Findings</h2>
<div class="content has-text-justified">
<p>
We present extensive experimental results on the LoCoMo benchmark using instruction-based LLMs, long-context LLMs, and RAG techniques. Our findings include:
<ul>
<li> <b>Long-context LLMs and RAG demonstrate effectiveness in QA tasks </b>, improving ‘memory’ capabilities of LLMs (with improvements ranging
from 22-66%), but still significantly lag behind human levels (by 56%), especially in temporal reasoning, (by 73%);
</li>
<li><b>Long-context LLMs demonstrate significant hallucinations</b>, leading to difficulty with adversarial questions in the QA task and in the event graph summarization task.</li>
<li><b>RAG offers a balanced compromise</b>, combining the accuracy of short-context LLMs with the
extensive comprehension of wide-context LLMs, and does particularly well when dialogues are transformed into a database of assertions (observations)
about each speaker’s life and persona.
</li>
</ul>
</p>
</div>
</div>
</div>
</div>
</section>
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<!-- <iframe src="./static/images/main_figure_arxiv.pdf" style="width: 100%;height: 100%;border: none;"></iframe>-->
<img src="./static/images/tables.svg" alt="Logo"/>
<!-- <video id="teaser" autoplay muted loop playsinline height="100%">-->
<!-- <source src="./static/images/main_figure.svg"-->
<!-- type="image/svg+xml">-->
<!-- </video>-->
<h2 class="subtitle has-text-centered">
<b>Question answering performance of Base, Long-context and RAG models.</b> Optimal performance is in bold.
Results are based on F1-score for answer prediction; higher is better.
</h2>
</div>
<div class="hero-body">
<!-- <iframe src="./static/images/main_figure_arxiv.pdf" style="width: 100%;height: 100%;border: none;"></iframe>-->
<img src="./static/images/minigpt5_results.svg" alt="Logo"/>
<!-- <video id="teaser" autoplay muted loop playsinline height="100%">-->
<!-- <source src="./static/images/main_figure.svg"-->
<!-- type="image/svg+xml">-->
<!-- </video>-->
<h2 class="subtitle has-text-centered">
<b>Multimodal dialog generation performance of MiniGPT-5</b>. (A) an example of multimodal dialog
predicted using MiniGPT5 with and without observation as retrieved context, (B) Variation of MM-Relevance score
with length of dialog history, and (C) comparison of RAG-based MiniGPT-5 methods.
</h2>
</div>
</div>
</section>
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>@article{maharana2024lococmo,
author = {Maharana, Adyasha and Lee, Dong-Ho and Tulyakov, Sergey and Bansal, Mohit and Barbieri, Francesco and Fang, Yuwei},
title = {Evaluating Very Long-Term Conversational Memory of LLM Agents.},
journal = {arxiv},
year = {2024},
}</code></pre>
</div>
</section>
<footer class="footer">
<div class="container">
<div class="content has-text-centered">
<a class="icon-link"
href="https://arxiv.org/abs/2303.16133">
<i class="fas fa-file-pdf"></i>
</a>
<a class="icon-link" href="https://github.com/adymaharana" 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>
Website design borrowed from <a
href="https://github.com/nerfies/nerfies.github.io">nerfies</a>.
</p>
</div>
</div>
</div>
</div>
</footer>
</body>
</html>