WEBVTT

1
00:00:01.855 --> 00:00:02.605
<00:00:01.855>Thank <00:00:02.080>you <00:00:02.305>very <00:00:02.530>much.

2
00:00:02.785 --> 00:00:04.845
<00:00:02.785>Um, <00:00:03.251>yeah, <00:00:03.718>I'm <00:00:04.185>Stefan.

3
00:00:04.985 --> 00:00:06.965
<00:00:04.985>Um, <00:00:05.160>and <00:00:05.335>I'm <00:00:05.510>very <00:00:05.685>proud <00:00:05.860>to <00:00:06.035>be <00:00:06.210>here <00:00:06.385>today

4
00:00:06.965 --> 00:00:10.245
<00:00:06.965>because, <00:00:07.140>uh, <00:00:07.314>we <00:00:07.490>present <00:00:07.665>something <00:00:07.840>which <00:00:08.015>is <00:00:08.189>a <00:00:08.365>new

5
00:00:10.245 --> 00:00:11.645
<00:00:10.245>product <00:00:10.875>at <00:00:11.504>Iono.

6
00:00:11.825 --> 00:00:13.485
<00:00:11.825>So <00:00:12.058>it's <00:00:12.291>one <00:00:12.524>of <00:00:12.758>our <00:00:12.991>first <00:00:13.225>steps

7
00:00:13.665 --> 00:00:16.605
<00:00:13.665>as <00:00:13.944>Iono <00:00:14.225>towards <00:00:14.504>an <00:00:14.784>AI <00:00:15.065>journey.

8
00:00:16.745 --> 00:00:18.725
<00:00:16.745>And <00:00:16.920>I'm <00:00:17.095>especially <00:00:17.270>happy <00:00:17.445>to <00:00:17.620>have <00:00:17.795>Kevin <00:00:17.970>with <00:00:18.145>me.

9
00:00:19.835 --> 00:00:22.765
<00:00:19.835>He's <00:00:20.035>from <00:00:20.235>Nvidia, <00:00:20.435>one <00:00:20.635>of <00:00:20.835>our <00:00:21.035>GPU <00:00:21.235>partners

10
00:00:22.865 --> 00:00:24.205
<00:00:22.865>or <00:00:23.106>one <00:00:23.347>of <00:00:23.588>our <00:00:23.829>is <00:00:24.070>Good.

11
00:00:24.315 --> 00:00:27.045
<00:00:24.315>He's <00:00:24.548>Nvidia <00:00:24.781>is <00:00:25.015>just <00:00:25.248>our <00:00:25.481>GPU <00:00:25.715>partner.

12
00:00:27.555 --> 00:00:29.045
<00:00:27.555>Well, <00:00:27.722>and <00:00:27.890>I <00:00:28.057>want <00:00:28.225>to <00:00:28.393>talk <00:00:28.560>a <00:00:28.728>little <00:00:28.896>bit

13
00:00:29.045 --> 00:00:30.205
<00:00:29.045>about <00:00:29.393>our <00:00:29.741>product <00:00:30.089>today.

14
00:00:30.705 --> 00:00:33.845
<00:00:30.705>And <00:00:30.832>the <00:00:30.959>basic <00:00:31.086>idea <00:00:31.214>is <00:00:31.341>that <00:00:31.468>we <00:00:31.595>want <00:00:31.723>to <00:00:31.850>simplify <00:00:31.977>the <00:00:32.104>usage

15
00:00:34.145 --> 00:00:37.205
<00:00:34.145>of <00:00:34.425>foundation <00:00:34.705>models <00:00:34.985>within <00:00:35.265>our <00:00:35.545>infrastructure

16
00:00:37.905 --> 00:00:42.525
<00:00:37.905>and <00:00:38.060>what <00:00:38.216>we'll <00:00:38.371>offer <00:00:38.527>as <00:00:38.682>a <00:00:38.838>product, <00:00:38.993>we <00:00:39.149>will <00:00:39.305>be

17
00:00:42.635 --> 00:00:45.765
<00:00:42.635>what <00:00:42.868>we <00:00:43.101>call <00:00:43.335>the <00:00:43.568>AI <00:00:43.801>model <00:00:44.035>hub.

18
00:00:46.065 --> 00:00:49.365
<00:00:46.065>And <00:00:46.192>we <00:00:46.319>will <00:00:46.446>launch <00:00:46.574>that <00:00:46.701>in <00:00:46.828>two <00:00:46.955>to <00:00:47.083>three <00:00:47.210>weeks <00:00:47.337>from <00:00:47.464>now.

19
00:00:49.465 --> 00:00:52.765
<00:00:49.465>So <00:00:49.640>expect <00:00:49.815>it <00:00:49.990>to <00:00:50.165>be <00:00:50.340>available <00:00:50.515>starting <00:00:50.690>in <00:00:50.865>the

20
00:00:52.765 --> 00:00:53.805
<00:00:52.765>first <00:00:53.077>week <00:00:53.389>of <00:00:53.701>August.

21
00:00:54.225 --> 00:00:58.845
<00:00:54.225>And <00:00:54.400>it's <00:00:54.575>basically <00:00:54.750>a <00:00:54.925>solution <00:00:55.100>that <00:00:55.275>helps <00:00:55.450>you <00:00:55.625>access,

22
00:00:59.585 --> 00:01:02.045
<00:00:59.585>um, <00:00:59.865>a <00:01:00.145>machine <00:01:00.425>learning <00:01:00.705>inference <00:01:00.985>service.

23
00:01:02.195 --> 00:01:06.045
<00:01:02.195>That <00:01:02.395>means, <00:01:02.595>means <00:01:02.795>we <00:01:02.995>offer <00:01:03.195>foundation <00:01:03.395>models, <00:01:03.595>um,

24
00:01:06.155 --> 00:01:09.565
<00:01:06.155>have <00:01:06.330>them <00:01:06.505>easily <00:01:06.680>accessible <00:01:06.855>for <00:01:07.030>you <00:01:07.205>via <00:01:07.380>an <00:01:07.555>API

25
00:01:10.265 --> 00:01:11.365
<00:01:10.265>hosted <00:01:10.759>in <00:01:11.255>Germany.

26
00:01:11.785 --> 00:01:15.965
<00:01:11.785>And, <00:01:12.018>um, <00:01:12.251>yeah, <00:01:12.485>support <00:01:12.718>retrieve <00:01:12.951>augmented <00:01:13.185>generation.

27
00:01:16.545 --> 00:01:18.605
<00:01:16.545>As <00:01:16.778>simple <00:01:17.011>as <00:01:17.245>we <00:01:17.478>can <00:01:17.711>think <00:01:17.945>of,

28
00:01:18.615 --> 00:01:20.365
<00:01:18.615>maybe <00:01:18.964>it's <00:01:19.315>even <00:01:19.664>simpler <00:01:20.015>somehow,

29
00:01:20.465 --> 00:01:23.645
<00:01:20.465>but <00:01:20.592>for <00:01:20.719>the <00:01:20.846>moment <00:01:20.974>it's <00:01:21.101>the <00:01:21.228>simplest <00:01:21.355>way <00:01:21.483>that <00:01:21.610>we <00:01:21.737>could <00:01:21.865>find.

30
00:01:24.065 --> 00:01:27.965
<00:01:24.065>And <00:01:24.205>now, <00:01:24.345>let's <00:01:24.485>put <00:01:24.625>it <00:01:24.765>this <00:01:24.905>way, <00:01:25.045>what <00:01:25.185>do <00:01:25.325>we <00:01:25.465>offer?

31
00:01:28.345 --> 00:01:31.645
<00:01:28.345>We <00:01:28.485>offer <00:01:28.625>the <00:01:28.765>whole <00:01:28.905>thing <00:01:29.045>in <00:01:29.185>a <00:01:29.325>very, <00:01:29.465>very <00:01:29.605>intuitive <00:01:29.745>API.

32
00:01:31.745 --> 00:01:33.485
<00:01:31.745>So <00:01:32.095>we <00:01:32.445>have <00:01:32.795>an <00:01:33.145>API

33
00:01:33.565 --> 00:01:36.645
<00:01:33.565>and <00:01:33.705>restful, <00:01:33.845>API <00:01:33.985>that <00:01:34.125>you <00:01:34.265>can <00:01:34.405>use <00:01:34.545>to <00:01:34.685>access <00:01:34.825>those <00:01:34.965>services.

34
00:01:37.505 --> 00:01:41.565
<00:01:37.505>We <00:01:37.660>have <00:01:37.816>a <00:01:37.971>selection <00:01:38.127>of <00:01:38.282>foundation <00:01:38.438>models <00:01:38.593>from <00:01:38.749>open <00:01:38.905>source,

35
00:01:41.615 --> 00:01:43.245
<00:01:41.615>which <00:01:41.965>are, <00:01:42.315>from <00:01:42.665>our <00:01:43.015>perspective,

36
00:01:43.345 --> 00:01:44.845
<00:01:43.345>the <00:01:43.537>best <00:01:43.730>at <00:01:43.923>the <00:01:44.116>market <00:01:44.309>at <00:01:44.502>the <00:01:44.695>moment.

37
00:01:45.275 --> 00:01:48.885
<00:01:45.275>They <00:01:45.430>are <00:01:45.586>all <00:01:45.741>open <00:01:45.897>source, <00:01:46.052>so <00:01:46.208>they <00:01:46.363>are <00:01:46.519>all, <00:01:46.675>um,

38
00:01:49.405 --> 00:01:54.205
<00:01:49.405>publicly <00:01:49.580>available <00:01:49.755>and <00:01:49.930>we <00:01:50.105>offer <00:01:50.280>a <00:01:50.455>vector <00:01:50.630>database <00:01:50.805>via

39
00:01:54.225 --> 00:01:57.805
<00:01:54.225>an <00:01:54.352>API, <00:01:54.479>so <00:01:54.606>that <00:01:54.734>you <00:01:54.861>can <00:01:54.988>use <00:01:55.115>it <00:01:55.243>without <00:01:55.370>having <00:01:55.497>to <00:01:55.625>do

40
00:01:57.825 --> 00:01:59.365
<00:01:57.825>or <00:01:58.102>implement <00:01:58.379>anything <00:01:58.656>on <00:01:58.933>your <00:01:59.211>own.

41
00:01:59.465 --> 00:02:01.485
<00:01:59.465>You <00:01:59.815>just <00:02:00.165>address <00:02:00.515>your <00:02:00.865>API

42
00:02:01.905 --> 00:02:04.445
<00:02:01.905>and <00:02:02.105>we <00:02:02.305>do <00:02:02.505>everything <00:02:02.705>in <00:02:02.905>the <00:02:03.105>backend <00:02:03.305>ourselves.

43
00:02:05.385 --> 00:02:08.605
<00:02:05.385>In <00:02:05.560>this <00:02:05.735>way, <00:02:05.910>we <00:02:06.085>allow <00:02:06.260>you <00:02:06.435>to <00:02:06.610>use <00:02:06.785>retrieval,

44
00:02:08.605 --> 00:02:11.685
<00:02:08.604>augmented <00:02:08.884>generation <00:02:09.164>or <00:02:09.445>implement <00:02:09.725>retrieval <00:02:10.005>augmented

45
00:02:11.685 --> 00:02:15.925
<00:02:11.685>generation <00:02:11.840>without <00:02:11.996>having <00:02:12.151>to <00:02:12.307>program <00:02:12.462>that <00:02:12.618>much <00:02:12.773>on <00:02:12.929>your <00:02:13.085>own.

46
00:02:16.155 --> 00:02:18.485
<00:02:16.155>It's <00:02:16.330>all <00:02:16.505>done <00:02:16.680>by <00:02:16.855>our <00:02:17.030>API <00:02:17.205>in <00:02:17.380>the <00:02:17.555>end.

47
00:02:18.945 --> 00:02:20.285
<00:02:18.945>And <00:02:19.346>one <00:02:19.749>additional <00:02:20.151>feature

48
00:02:20.285 --> 00:02:23.725
<00:02:20.285>that <00:02:20.460>we'll <00:02:20.635>offer <00:02:20.810>is <00:02:20.984>we <00:02:21.160>follow <00:02:21.335>the <00:02:21.510>OpenAI <00:02:21.685>specification.

49
00:02:24.145 --> 00:02:28.725
<00:02:24.145>So <00:02:24.300>you <00:02:24.456>can <00:02:24.611>basically <00:02:24.767>use <00:02:24.922>your <00:02:25.078>existing <00:02:25.233>tools <00:02:25.389>that <00:02:25.545>use

50
00:02:29.465 --> 00:02:31.925
<00:02:29.465>OpenAI, <00:02:29.931>they <00:02:30.398>OpenAI <00:02:30.865>API

51
00:02:32.545 --> 00:02:36.885
<00:02:32.545>and <00:02:32.700>use <00:02:32.856>our <00:02:33.011>services <00:02:33.167>backend <00:02:33.322>by <00:02:33.478>just <00:02:33.633>switching <00:02:33.789>the <00:02:33.945>URL

52
00:02:36.885 --> 00:02:38.725
<00:02:36.885>that <00:02:37.165>you're <00:02:37.445>accessing <00:02:37.725>the <00:02:38.005>data <00:02:38.285>from.

53
00:02:39.785 --> 00:02:43.845
<00:02:39.785>So <00:02:39.924>what <00:02:40.065>is <00:02:40.204>it <00:02:40.345>or <00:02:40.484>what, <00:02:40.625>uh, <00:02:40.765>models <00:02:40.905>do <00:02:41.045>we <00:02:41.185>offer?

54
00:02:43.845 --> 00:02:48.725
<00:02:43.845>That's <00:02:43.984>basically <00:02:44.125>the <00:02:44.265>list <00:02:44.405>we <00:02:44.545>offer <00:02:44.685>from <00:02:44.825>Meta <00:02:44.965>LAMA <00:02:45.105>three, <00:02:45.245>uh,

55
00:02:48.745 --> 00:02:50.725
<00:02:48.745>the <00:02:48.945>8 <00:02:49.145>billion <00:02:49.345>and <00:02:49.545>the <00:02:49.745>70 <00:02:49.945>billion <00:02:50.145>version.

56
00:02:51.625 --> 00:02:55.165
<00:02:51.625>And <00:02:51.858>we <00:02:52.091>offer <00:02:52.325>a <00:02:52.558>LAMA <00:02:52.791>instruct <00:02:53.025>model.

57
00:02:55.665 --> 00:02:57.525
<00:02:55.665>So <00:02:55.945>basically <00:02:56.225>these <00:02:56.505>are <00:02:56.785>the <00:02:57.065>biggest

58
00:02:57.585 --> 00:03:02.205
<00:02:57.585>and <00:02:57.740>most <00:02:57.896>important <00:02:58.051>models <00:02:58.207>from <00:02:58.362>lama, <00:02:58.518>the <00:02:58.673>US <00:02:58.829>competitor, <00:02:58.985>um,

59
00:03:02.985 --> 00:03:06.685
<00:03:02.985>for <00:03:03.125>open <00:03:03.265>source <00:03:03.405>models <00:03:03.545>and <00:03:03.685>to <00:03:03.825>our <00:03:03.965>best <00:03:04.105>of <00:03:04.245>our <00:03:04.385>knowledge

60
00:03:06.685 --> 00:03:09.885
<00:03:06.685>and <00:03:06.840>to <00:03:06.996>our, <00:03:07.151>based <00:03:07.307>on <00:03:07.462>our <00:03:07.618>all <00:03:07.773>our <00:03:07.929>analysis, <00:03:08.085>the,

61
00:03:09.885 --> 00:03:12.005
<00:03:09.885>the <00:03:10.060>best <00:03:10.234>models <00:03:10.410>that <00:03:10.584>are <00:03:10.760>currently <00:03:10.935>in <00:03:11.109>the <00:03:11.285>market,

62
00:03:12.235 --> 00:03:15.805
<00:03:12.235>when <00:03:12.390>you <00:03:12.546>look <00:03:12.701>at <00:03:12.857>open <00:03:13.012>source <00:03:13.168>models, <00:03:13.323>aside <00:03:13.479>from <00:03:13.635>that,

63
00:03:15.805 --> 00:03:19.325
<00:03:15.805>we <00:03:16.038>also <00:03:16.271>offer <00:03:16.505>mis <00:03:16.738>mistrial, <00:03:16.971>um, <00:03:17.205>mistrial,

64
00:03:19.515 --> 00:03:21.525
<00:03:19.515>basically <00:03:19.795>the <00:03:20.075>7 <00:03:20.355>billion, <00:03:20.635>uh, <00:03:20.915>model

65
00:03:21.585 --> 00:03:24.765
<00:03:21.585>and <00:03:21.785>the <00:03:21.985>mixed <00:03:22.185>trial <00:03:22.385>version <00:03:22.585>as <00:03:22.785>an <00:03:22.985>alternative

66
00:03:24.905 --> 00:03:29.205
<00:03:24.905>of <00:03:25.045>a <00:03:25.185>European <00:03:25.325>model, <00:03:25.465>meaning <00:03:25.605>that <00:03:25.745>we <00:03:25.885>offer <00:03:26.025>a <00:03:26.165>model <00:03:26.305>that

67
00:03:29.995 --> 00:03:32.925
<00:03:29.995>yeah, <00:03:30.275>is <00:03:30.555>better <00:03:30.835>concerning <00:03:31.115>European <00:03:31.395>languages.

68
00:03:32.925 --> 00:03:37.085
<00:03:32.925>Yeah, <00:03:33.080>while <00:03:33.236>llama <00:03:33.391>is <00:03:33.547>very, <00:03:33.702>very <00:03:33.858>good <00:03:34.013>on <00:03:34.169>English <00:03:34.325>texts,

69
00:03:37.415 --> 00:03:39.885
<00:03:37.415>mixed <00:03:37.695>trial <00:03:37.975>is <00:03:38.255>from <00:03:38.535>our <00:03:38.815>perspective,

70
00:03:40.025 --> 00:03:41.565
<00:03:40.025>the <00:03:40.256>current <00:03:40.487>market <00:03:40.718>leader <00:03:40.949>when <00:03:41.180>it <00:03:41.411>comes

71
00:03:41.585 --> 00:03:45.045
<00:03:41.585>to <00:03:41.818>European <00:03:42.051>languages <00:03:42.285>like <00:03:42.518>German, <00:03:42.751>France, <00:03:42.985>French,

72
00:03:45.105 --> 00:03:46.125
<00:03:45.105>and <00:03:45.288>so <00:03:45.472>on <00:03:45.655>and <00:03:45.839>so <00:03:46.023>forth.

73
00:03:46.945 --> 00:03:49.645
<00:03:46.945>And <00:03:47.295>the <00:03:47.644>last <00:03:47.995>foundation <00:03:48.345>model

74
00:03:49.645 --> 00:03:52.525
<00:03:49.645>that <00:03:49.820>we <00:03:49.995>will <00:03:50.170>offer <00:03:50.345>is <00:03:50.520>stab <00:03:50.695>from <00:03:50.870>stability, <00:03:51.045>ai,

75
00:03:52.945 --> 00:03:54.405
<00:03:52.945>stable <00:03:53.602>fusion <00:03:54.259>xl.

76
00:03:54.705 --> 00:03:56.685
<00:03:54.704>So <00:03:54.904>that's <00:03:55.105>basically <00:03:55.304>a <00:03:55.505>text <00:03:55.704>to <00:03:55.904>image <00:03:56.105>model.

77
00:03:56.785 --> 00:03:58.405
<00:03:56.785>And <00:03:57.251>this <00:03:57.718>all, <00:03:58.185>uh,

78
00:03:58.545 --> 00:04:03.485
<00:03:58.545>it <00:03:58.685>will <00:03:58.825>be <00:03:58.965>available <00:03:59.105>via <00:03:59.245>our <00:03:59.385>API <00:03:59.525>on <00:03:59.665>our, <00:03:59.805>uh, <00:03:59.945>infrastructure.

79
00:04:04.905 --> 00:04:08.405
<00:04:04.905>Now, <00:04:05.080>why <00:04:05.255>do <00:04:05.430>I <00:04:05.605>believe, <00:04:05.780>uh, <00:04:05.955>this <00:04:06.130>is <00:04:06.305>cool

80
00:04:08.405 --> 00:04:11.285
<00:04:08.405>or <00:04:08.605>what <00:04:08.805>is <00:04:09.005>from <00:04:09.205>my <00:04:09.405>perspective, <00:04:09.605>the <00:04:09.805>unique

81
00:04:11.305 --> 00:04:12.485
<00:04:11.305>and <00:04:11.570>the <00:04:11.836>cool <00:04:12.101>part <00:04:12.367>here?

82
00:04:12.795 --> 00:04:14.725
<00:04:12.795>Well, <00:04:13.261>what <00:04:13.728>we <00:04:14.195>do

83
00:04:14.725 --> 00:04:18.245
<00:04:14.725>or <00:04:14.852>what <00:04:14.979>we <00:04:15.106>support <00:04:15.234>is <00:04:15.361>we <00:04:15.488>have <00:04:15.615>this <00:04:15.743>easy <00:04:15.870>to <00:04:15.997>use <00:04:16.125>API.

84
00:04:18.865 --> 00:04:22.445
<00:04:18.865>And <00:04:18.981>the <00:04:19.098>thing <00:04:19.215>here <00:04:19.331>is <00:04:19.448>you <00:04:19.565>basically <00:04:19.681>have <00:04:19.798>to <00:04:19.915>set <00:04:20.031>up <00:04:20.148>a <00:04:20.265>header.

85
00:04:22.705 --> 00:04:25.405
<00:04:22.705>And <00:04:22.905>in <00:04:23.104>this <00:04:23.305>header <00:04:23.505>you <00:04:23.705>put <00:04:23.905>your <00:04:24.104>credentials

86
00:04:25.905 --> 00:04:29.205
<00:04:25.905>and <00:04:26.138>using <00:04:26.371>this <00:04:26.604>credentials, <00:04:26.838>you <00:04:27.071>authenticate <00:04:27.304>yourself

87
00:04:29.345 --> 00:04:32.805
<00:04:29.345>to <00:04:29.811>access <00:04:30.278>the <00:04:30.745>service.

88
00:04:33.385 --> 00:04:38.365
<00:04:33.385>And, <00:04:33.525>uh, <00:04:33.664>on <00:04:33.805>top <00:04:33.945>of <00:04:34.085>that <00:04:34.224>you <00:04:34.365>have <00:04:34.505>some, <00:04:34.645>uh, <00:04:34.784>endpoints.

89
00:04:38.585 --> 00:04:40.805
<00:04:38.585>You <00:04:38.818>send <00:04:39.051>your <00:04:39.284>query <00:04:39.518>to <00:04:39.751>these <00:04:39.984>endpoints,

90
00:04:41.185 --> 00:04:44.165
<00:04:41.185>and <00:04:41.360>these <00:04:41.535>endpoints <00:04:41.710>just, <00:04:41.885>uh, <00:04:42.060>give <00:04:42.235>you <00:04:42.410>the <00:04:42.585>result.

91
00:04:44.555 --> 00:04:48.045
<00:04:44.555>What <00:04:44.682>you <00:04:44.809>do <00:04:44.936>need <00:04:45.064>to <00:04:45.191>add <00:04:45.318>is <00:04:45.445>some <00:04:45.573>body <00:04:45.700>with <00:04:45.827>the <00:04:45.955>payload.

92
00:04:48.185 --> 00:04:50.645
<00:04:48.185>So <00:04:48.360>for <00:04:48.535>example, <00:04:48.710>you <00:04:48.885>would <00:04:49.060>add <00:04:49.235>your <00:04:49.410>prompt <00:04:49.585>here

93
00:04:51.105 --> 00:04:53.085
<00:04:51.105>and <00:04:51.338>send <00:04:51.571>the <00:04:51.805>prompt <00:04:52.038>to <00:04:52.271>this <00:04:52.505>endpoint.

94
00:04:53.105 --> 00:04:57.925
<00:04:53.105>And <00:04:53.245>in <00:04:53.385>this <00:04:53.525>way <00:04:53.665>you <00:04:53.805>would <00:04:53.945>get, <00:04:54.085>um, <00:04:54.225>yeah, <00:04:54.365>the <00:04:54.505>output

95
00:04:57.925 --> 00:04:59.605
<00:04:57.925>of <00:04:58.158>a <00:04:58.391>large <00:04:58.625>language <00:04:58.858>model, <00:04:59.091>for <00:04:59.325>example.

96
00:05:00.945 --> 00:05:03.205
<00:05:00.945>And <00:05:01.145>why <00:05:01.344>do <00:05:01.545>we <00:05:01.745>do <00:05:01.945>use <00:05:02.145>this <00:05:02.344>approach?

97
00:05:03.235 --> 00:05:04.525
<00:05:03.235>Well, <00:05:03.525>the <00:05:03.815>big <00:05:04.105>part <00:05:04.395>here

98
00:05:04.525 --> 00:05:06.485
<00:05:04.525>or <00:05:04.724>the <00:05:04.924>big <00:05:05.125>advantages <00:05:05.325>here, <00:05:05.525>we <00:05:05.724>are <00:05:05.924>independent

99
00:05:06.485 --> 00:05:07.525
<00:05:06.485>of <00:05:06.797>the <00:05:07.109>programming <00:05:07.421>language.

100
00:05:07.825 --> 00:05:11.125
<00:05:07.825>You <00:05:08.000>can <00:05:08.175>use <00:05:08.349>this <00:05:08.525>API <00:05:08.700>with <00:05:08.875>every <00:05:09.050>programming <00:05:09.224>language

101
00:05:11.125 --> 00:05:14.205
<00:05:11.125>that <00:05:11.280>you <00:05:11.436>want <00:05:11.591>as <00:05:11.747>long <00:05:11.902>as <00:05:12.058>it <00:05:12.213>supports <00:05:12.369>requests, <00:05:12.525>right?

102
00:05:15.945 --> 00:05:19.845
<00:05:15.945>And <00:05:16.224>now <00:05:16.505>that's <00:05:16.784>just <00:05:17.065>what <00:05:17.344>we

103
00:05:19.905 --> 00:05:21.245
<00:05:19.905>and <00:05:20.206>all <00:05:20.508>the <00:05:20.809>competitors <00:05:21.111>do.

104
00:05:21.245 --> 00:05:22.405
<00:05:21.245>That's <00:05:21.593>not <00:05:21.941>too <00:05:22.289>special.

105
00:05:22.865 --> 00:05:26.125
<00:05:22.865>Now, <00:05:23.020>what <00:05:23.176>we <00:05:23.331>think <00:05:23.487>is <00:05:23.642>special <00:05:23.798>in <00:05:23.953>our <00:05:24.109>case <00:05:24.265>is

106
00:05:26.675 --> 00:05:28.965
<00:05:26.675>that <00:05:26.955>we <00:05:27.235>support <00:05:27.515>retrieve <00:05:27.795>augmented <00:05:28.075>generation.

107
00:05:29.145 --> 00:05:31.645
<00:05:29.145>And <00:05:29.284>I <00:05:29.424>want <00:05:29.565>to <00:05:29.705>emphasize <00:05:29.844>a <00:05:29.984>little <00:05:30.125>bit <00:05:30.265>more <00:05:30.405>on <00:05:30.544>this.

108
00:05:31.785 --> 00:05:35.445
<00:05:31.785>So <00:05:31.960>the <00:05:32.135>basic <00:05:32.310>idea <00:05:32.485>of <00:05:32.660>retrieve <00:05:32.835>augmented <00:05:33.010>generation <00:05:33.185>is

109
00:05:35.835 --> 00:05:39.365
<00:05:35.835>that <00:05:36.010>you <00:05:36.185>typically <00:05:36.359>do <00:05:36.534>not <00:05:36.710>fine <00:05:36.885>tune <00:05:37.060>large <00:05:37.234>language

110
00:05:39.365 --> 00:05:40.485
<00:05:39.365>models <00:05:39.617>on <00:05:39.869>your <00:05:40.121>own, <00:05:40.373>right?

111
00:05:40.755 --> 00:05:45.205
<00:05:40.755>It's, <00:05:40.910>and <00:05:41.066>Kevin, <00:05:41.221>you <00:05:41.377>can, <00:05:41.532>uh, <00:05:41.688>say <00:05:41.843>that <00:05:41.999>it's, <00:05:42.155>uh,

112
00:05:45.265 --> 00:05:47.565
<00:05:45.265>you <00:05:45.440>see <00:05:45.615>it <00:05:45.789>differently <00:05:45.965>because <00:05:46.140>you <00:05:46.315>sell <00:05:46.490>the <00:05:46.664>hardware,

113
00:05:47.565 --> 00:05:50.005
<00:05:47.565>but <00:05:47.765>from <00:05:47.965>our <00:05:48.165>perspective, <00:05:48.365>it's <00:05:48.565>much <00:05:48.765>too <00:05:48.965>expensive

114
00:05:50.385 --> 00:05:55.205
<00:05:50.385>to <00:05:50.525>fine <00:05:50.664>tune <00:05:50.805>large <00:05:50.945>language <00:05:51.085>models <00:05:51.224>on <00:05:51.365>an <00:05:51.505>Nvidia <00:05:51.645>GPU <00:05:51.784>just

115
00:05:55.225 --> 00:05:56.485
<00:05:55.225>for <00:05:55.508>your <00:05:55.792>small <00:05:56.075>business <00:05:56.359>case.

116
00:05:56.485 --> 00:05:59.565
<00:05:56.485>Yeah, <00:05:56.685>you <00:05:56.885>would <00:05:57.085>need <00:05:57.285>millions <00:05:57.485>of <00:05:57.685>training <00:05:57.885>data.

117
00:05:59.745 --> 00:06:01.045
<00:05:59.745>You <00:05:59.979>typically <00:06:00.213>do <00:06:00.447>not <00:06:00.681>have <00:06:00.915>that.

118
00:06:01.425 --> 00:06:03.205
<00:06:01.425>You <00:06:01.891>would <00:06:02.358>need <00:06:02.825>one

119
00:06:03.205 --> 00:06:04.765
<00:06:03.205>of <00:06:03.555>the <00:06:03.905>large <00:06:04.255>expensive <00:06:04.604>Nvidia

120
00:06:04.965 --> 00:06:06.325
<00:06:04.965>hardwares <00:06:05.209>just <00:06:05.454>to <00:06:05.699>do <00:06:05.944>the <00:06:06.188>training.

121
00:06:06.745 --> 00:06:10.685
<00:06:06.745>And <00:06:06.900>typically <00:06:07.056>for <00:06:07.211>most <00:06:07.367>of <00:06:07.522>our <00:06:07.678>customers, <00:06:07.833>that <00:06:07.989>does <00:06:08.145>not

122
00:06:11.225 --> 00:06:12.485
<00:06:11.225>really <00:06:11.792>pay, <00:06:12.359>right?

123
00:06:14.315 --> 00:06:15.325
<00:06:14.315>Glad <00:06:14.618>he <00:06:14.921>he <00:06:15.224>not.

124
00:06:15.665 --> 00:06:17.925
<00:06:15.665>So <00:06:15.898>even <00:06:16.131>Nvidia <00:06:16.365>sees <00:06:16.598>it, <00:06:16.831>oh, <00:06:17.065>sorry,

125
00:06:18.265 --> 00:06:20.045
<00:06:18.265>the <00:06:18.465>over <00:06:18.664>there <00:06:18.865>is <00:06:19.065>one <00:06:19.265>of <00:06:19.465>his <00:06:19.664>bosses.

126
00:06:20.225 --> 00:06:22.485
<00:06:20.225>So, <00:06:20.425>uh, <00:06:20.625>he <00:06:20.825>cannot, <00:06:21.025>uh, <00:06:21.225>comment <00:06:21.425>on <00:06:21.625>this.

127
00:06:22.905 --> 00:06:24.005
<00:06:22.905>We <00:06:23.102>do <00:06:23.301>not <00:06:23.498>see <00:06:23.697>it, <00:06:23.895>right?

128
00:06:24.065 --> 00:06:25.845
<00:06:24.065>Uh, <00:06:24.531>from <00:06:24.998>Nvidia <00:06:25.465>perspective,

129
00:06:25.865 --> 00:06:28.445
<00:06:25.865>you <00:06:26.065>should <00:06:26.265>always <00:06:26.465>fine <00:06:26.665>tune <00:06:26.865>from <00:06:27.065>our <00:06:27.265>perspective,

130
00:06:28.625 --> 00:06:29.725
<00:06:28.625>you <00:06:28.955>should <00:06:29.285>not <00:06:29.615>should

131
00:06:29.785 --> 00:06:31.565
<00:06:29.785>Run <00:06:29.960>on <00:06:30.135>as <00:06:30.310>many <00:06:30.485>GPUs <00:06:30.660>as <00:06:30.835>you <00:06:31.010>ever <00:06:31.185>can.

132
00:06:32.375 --> 00:06:35.965
<00:06:32.375>Sorry, <00:06:32.515>should <00:06:32.655>Run <00:06:32.795>on <00:06:32.935>as <00:06:33.075>many <00:06:33.215>GPUs <00:06:33.355>as <00:06:33.495>you <00:06:33.635>can. <00:06:33.775>Yeah.

133
00:06:37.505 --> 00:06:40.285
<00:06:37.505>And <00:06:37.705>more <00:06:37.905>parameters <00:06:38.105>is <00:06:38.305>always <00:06:38.505>better, <00:06:38.705>right? <00:06:38.905>Always.

134
00:06:41.825 --> 00:06:44.045
<00:06:41.825>So <00:06:42.058>we <00:06:42.291>suggest <00:06:42.525>not <00:06:42.758>to <00:06:42.991>do <00:06:43.224>it,

135
00:06:44.065 --> 00:06:46.125
<00:06:44.065>and <00:06:44.220>I <00:06:44.376>think <00:06:44.531>it's <00:06:44.687>the <00:06:44.842>standard <00:06:44.998>in <00:06:45.153>the <00:06:45.309>market, <00:06:45.465>right?

136
00:06:46.185 --> 00:06:47.805
<00:06:46.185>And <00:06:46.465>then <00:06:46.745>what's <00:06:47.025>the <00:06:47.305>approach <00:06:47.585>now

137
00:06:47.805 --> 00:06:51.005
<00:06:47.805>to <00:06:47.980>get <00:06:48.155>your <00:06:48.330>large <00:06:48.505>language <00:06:48.680>model, <00:06:48.855>answer <00:06:49.030>your <00:06:49.205>queries

138
00:06:51.355 --> 00:06:53.365
<00:06:51.355>with <00:06:51.635>the <00:06:51.915>data <00:06:52.195>from <00:06:52.475>your <00:06:52.755>company?

139
00:06:53.585 --> 00:06:57.005
<00:06:53.585>So <00:06:53.724>think <00:06:53.864>for <00:06:54.005>example, <00:06:54.145>of <00:06:54.284>a <00:06:54.424>knowledge <00:06:54.565>database <00:06:54.705>that <00:06:54.844>you <00:06:54.984>have

140
00:06:57.265 --> 00:06:59.605
<00:06:57.265>and <00:06:57.405>that <00:06:57.544>you <00:06:57.685>want <00:06:57.825>to <00:06:57.965>use <00:06:58.104>in <00:06:58.245>your <00:06:58.385>large <00:06:58.525>language <00:06:58.664>model.

141
00:06:59.605 --> 00:07:01.965
<00:06:59.605>That's <00:06:59.760>a <00:06:59.916>use <00:07:00.071>case <00:07:00.227>that <00:07:00.382>several <00:07:00.538>of <00:07:00.693>our <00:07:00.849>customers <00:07:01.005>have.

142
00:07:02.305 --> 00:07:05.085
<00:07:02.305>So <00:07:02.460>the <00:07:02.616>idea <00:07:02.771>is <00:07:02.927>you <00:07:03.082>have <00:07:03.238>all <00:07:03.393>your <00:07:03.549>company <00:07:03.705>data

143
00:07:05.465 --> 00:07:07.085
<00:07:05.465>and <00:07:05.698>you <00:07:05.931>typically <00:07:06.164>do <00:07:06.398>not <00:07:06.631>find <00:07:06.864>it.

144
00:07:07.105 --> 00:07:09.125
<00:07:07.105>So <00:07:07.305>the <00:07:07.505>idea <00:07:07.705>is <00:07:07.905>putting <00:07:08.105>it <00:07:08.305>into <00:07:08.505>a

145
00:07:09.655 --> 00:07:11.245
<00:07:09.655>large <00:07:10.121>language <00:07:10.588>model <00:07:11.054>somehow

146
00:07:11.745 --> 00:07:14.445
<00:07:11.745>and <00:07:11.900>ask <00:07:12.056>the <00:07:12.211>large <00:07:12.367>language <00:07:12.522>model <00:07:12.678>to <00:07:12.833>summarize <00:07:12.989>your <00:07:13.145>data.

147
00:07:14.445 --> 00:07:15.685
<00:07:14.445>And <00:07:14.724>that <00:07:15.003>works <00:07:15.282>very <00:07:15.561>well

148
00:07:16.075 --> 00:07:18.765
<00:07:16.075>with <00:07:16.354>this <00:07:16.635>retrieve <00:07:16.914>augmented <00:07:17.195>generation <00:07:17.474>approach.

149
00:07:18.945 --> 00:07:22.845
<00:07:18.945>So <00:07:19.061>the <00:07:19.178>idea <00:07:19.295>here <00:07:19.411>is <00:07:19.528>you <00:07:19.645>save <00:07:19.761>all <00:07:19.878>your <00:07:19.995>documents <00:07:20.111>in <00:07:20.228>a <00:07:20.344>vector

150
00:07:23.165 --> 00:07:26.485
<00:07:23.165>database, <00:07:23.305>and <00:07:23.445>the <00:07:23.585>idea <00:07:23.725>of <00:07:23.865>a <00:07:24.005>vector <00:07:24.145>database <00:07:24.285>is <00:07:24.425>just, <00:07:24.565>it's

151
00:07:26.995 --> 00:07:28.605
<00:07:26.995>also <00:07:27.228>a <00:07:27.461>deep <00:07:27.695>learning <00:07:27.928>model <00:07:28.161>in <00:07:28.395>front.

152
00:07:28.705 --> 00:07:30.365
<00:07:28.705>You <00:07:28.984>put <00:07:29.265>your <00:07:29.544>text <00:07:29.825>in <00:07:30.104>there,

153
00:07:30.865 --> 00:07:34.405
<00:07:30.865>it <00:07:31.065>encodes <00:07:31.265>the <00:07:31.465>text <00:07:31.665>into <00:07:31.865>a <00:07:32.065>multidimensional <00:07:32.265>space,

154
00:07:34.985 --> 00:07:37.645
<00:07:34.985>and <00:07:35.218>all <00:07:35.451>texts <00:07:35.685>which <00:07:35.918>are <00:07:36.151>semantically <00:07:36.385>similar

155
00:07:38.345 --> 00:07:40.405
<00:07:38.345>end <00:07:38.545>up <00:07:38.745>in <00:07:38.945>a <00:07:39.145>similar <00:07:39.345>position <00:07:39.545>in <00:07:39.745>this

156
00:07:40.945 --> 00:07:42.085
<00:07:40.945>multi-dimensional <00:07:41.971>space.

157
00:07:42.505 --> 00:07:46.165
<00:07:42.505>And <00:07:42.632>in <00:07:42.759>this <00:07:42.886>way, <00:07:43.014>you <00:07:43.141>can <00:07:43.268>use <00:07:43.395>this <00:07:43.523>vector <00:07:43.650>database <00:07:43.777>to <00:07:43.905>do

158
00:07:46.885 --> 00:07:51.525
<00:07:46.885>searches <00:07:47.164>multi <00:07:47.445>churches <00:07:47.724>based <00:07:48.005>on <00:07:48.284>queries

159
00:07:51.785 --> 00:07:53.125
<00:07:51.784>and <00:07:52.388>find <00:07:52.991>very,

160
00:07:53.155 --> 00:07:56.285
<00:07:53.155>very <00:07:53.388>similar <00:07:53.621>documents <00:07:53.854>from <00:07:54.088>a <00:07:54.321>semantical <00:07:54.554>perspective.

161
00:07:56.385 --> 00:07:58.925
<00:07:56.385>So <00:07:56.560>this <00:07:56.735>is <00:07:56.909>a <00:07:57.085>change <00:07:57.260>compared <00:07:57.435>to, <00:07:57.610>to <00:07:57.784>before,

162
00:07:59.505 --> 00:08:02.085
<00:07:59.505>before <00:07:59.705>you <00:07:59.905>always <00:08:00.105>could <00:08:00.305>look <00:08:00.505>for <00:08:00.705>similar <00:08:00.905>words.

163
00:08:02.105 --> 00:08:05.405
<00:08:02.105>So <00:08:02.260>something <00:08:02.416>like <00:08:02.571>that, <00:08:02.727>this <00:08:02.882>approach <00:08:03.038>helps <00:08:03.193>you <00:08:03.349>to <00:08:03.505>find

164
00:08:05.965 --> 00:08:07.805
<00:08:05.965>semantically <00:08:06.431>similar <00:08:06.898>documents, <00:08:07.364>right?

165
00:08:08.945 --> 00:08:10.845
<00:08:08.945>So <00:08:09.145>that's <00:08:09.344>the <00:08:09.545>idea <00:08:09.745>of <00:08:09.945>the <00:08:10.145>vector <00:08:10.344>database.

166
00:08:11.025 --> 00:08:14.525
<00:08:11.025>Now, <00:08:11.200>when <00:08:11.375>a <00:08:11.549>customer <00:08:11.724>comes <00:08:11.900>and <00:08:12.075>has <00:08:12.250>a <00:08:12.424>question

167
00:08:14.705 --> 00:08:16.445
<00:08:14.705>and <00:08:14.938>you, <00:08:15.171>for <00:08:15.405>example, <00:08:15.638>have <00:08:15.871>an <00:08:16.104>application

168
00:08:16.445 --> 00:08:18.605
<00:08:16.445>or <00:08:16.620>a <00:08:16.795>chatbot <00:08:16.969>who <00:08:17.145>wants <00:08:17.320>to <00:08:17.495>answer <00:08:17.670>this <00:08:17.844>question,

169
00:08:19.385 --> 00:08:22.525
<00:08:19.385>he <00:08:19.618>would <00:08:19.851>then <00:08:20.085>ask <00:08:20.318>this <00:08:20.551>vector <00:08:20.784>database

170
00:08:22.665 --> 00:08:26.445
<00:08:22.665>for <00:08:22.865>the <00:08:23.065>semantically <00:08:23.265>similar <00:08:23.465>articles <00:08:23.665>to <00:08:23.865>this <00:08:24.065>question.

171
00:08:27.105 --> 00:08:30.085
<00:08:27.105>So <00:08:27.245>if <00:08:27.385>you <00:08:27.525>would <00:08:27.665>ask, <00:08:27.805>for <00:08:27.945>example, <00:08:28.085>what <00:08:28.225>does <00:08:28.365>anos <00:08:28.505>do?

172
00:08:30.505 --> 00:08:34.525
<00:08:30.505>It <00:08:30.660>would <00:08:30.816>show <00:08:30.971>you <00:08:31.127>all <00:08:31.282>documents <00:08:31.438>that <00:08:31.593>cope <00:08:31.749>with <00:08:31.905>this,

173
00:08:34.945 --> 00:08:36.245
<00:08:34.945>uh, <00:08:35.530>issue, <00:08:36.115>right?

174
00:08:36.705 --> 00:08:38.925
<00:08:36.705>And <00:08:36.845>then <00:08:36.985>in <00:08:37.125>the <00:08:37.265>next <00:08:37.405>step <00:08:37.545>you <00:08:37.685>would <00:08:37.825>get <00:08:37.965>the <00:08:38.105>results,

175
00:08:38.985 --> 00:08:41.085
<00:08:38.985>the <00:08:39.265>documents <00:08:39.544>from <00:08:39.825>this <00:08:40.105>vector <00:08:40.385>database

176
00:08:41.425 --> 00:08:44.325
<00:08:41.424>and <00:08:41.580>would <00:08:41.736>add <00:08:41.891>this <00:08:42.047>documents <00:08:42.202>to <00:08:42.358>a <00:08:42.513>large <00:08:42.669>language <00:08:42.824>model,

177
00:08:44.385 --> 00:08:46.005
<00:08:44.385>to <00:08:44.585>a <00:08:44.785>prompt <00:08:44.985>of <00:08:45.185>a <00:08:45.385>large <00:08:45.585>language <00:08:45.785>model,

178
00:08:46.385 --> 00:08:48.685
<00:08:46.385>and <00:08:46.618>you <00:08:46.851>would <00:08:47.085>then <00:08:47.318>get <00:08:47.551>an <00:08:47.785>answer,

179
00:08:48.895 --> 00:08:51.045
<00:08:48.895>which <00:08:49.128>is <00:08:49.361>highly <00:08:49.595>customized <00:08:49.828>on <00:08:50.061>your <00:08:50.295>data.

180
00:08:51.505 --> 00:08:54.605
<00:08:51.505>So <00:08:51.645>in <00:08:51.785>this <00:08:51.925>way, <00:08:52.064>the <00:08:52.205>cool <00:08:52.345>thing <00:08:52.485>is <00:08:52.625>you <00:08:52.765>do <00:08:52.905>not

181
00:08:54.605 --> 00:08:55.685
<00:08:54.605>have <00:08:54.929>to <00:08:55.252>train <00:08:55.577>anything.

182
00:08:55.825 --> 00:08:57.685
<00:08:55.825>You <00:08:56.025>do <00:08:56.225>not <00:08:56.425>need <00:08:56.625>to <00:08:56.825>have <00:08:57.025>a <00:08:57.225>very

183
00:08:58.315 --> 00:09:01.085
<00:08:58.315>expensive <00:08:58.515>GPU <00:08:58.715>doing <00:08:58.915>all <00:08:59.115>the <00:08:59.315>training <00:08:59.515>for <00:08:59.715>you,

184
00:09:01.505 --> 00:09:05.285
<00:09:01.505>and <00:09:01.632>you <00:09:01.759>can <00:09:01.886>at <00:09:02.014>the <00:09:02.141>same <00:09:02.268>time <00:09:02.395>have <00:09:02.523>the <00:09:02.650>most <00:09:02.777>recent <00:09:02.905>documents

185
00:09:05.465 --> 00:09:07.365
<00:09:05.465>in <00:09:05.745>your <00:09:06.025>collection <00:09:06.305>and <00:09:06.585>have <00:09:06.865>them

186
00:09:07.385 --> 00:09:08.845
<00:09:07.385>and <00:09:07.604>use <00:09:07.823>them <00:09:08.042>to <00:09:08.261>answer <00:09:08.480>your <00:09:08.699>query.

187
00:09:10.065 --> 00:09:13.405
<00:09:10.065>And <00:09:10.265>that's <00:09:10.465>exactly <00:09:10.665>what <00:09:10.865>retrieve <00:09:11.065>augmented <00:09:11.265>generation <00:09:11.465>is

188
00:09:13.825 --> 00:09:16.405
<00:09:13.825>and <00:09:14.058>what <00:09:14.291>our <00:09:14.525>product <00:09:14.758>now <00:09:14.991>offers <00:09:15.225>is.

189
00:09:16.865 --> 00:09:20.005
<00:09:16.865>You <00:09:17.020>can <00:09:17.176>do <00:09:17.331>that <00:09:17.487>in <00:09:17.642>just <00:09:17.798>one <00:09:17.953>API <00:09:18.109>call. <00:09:18.265>Yeah.

190
00:09:20.005 --> 00:09:22.765
<00:09:20.005>And <00:09:20.145>this <00:09:20.285>is <00:09:20.425>basically <00:09:20.564>the <00:09:20.705>source <00:09:20.845>code <00:09:20.985>for <00:09:21.125>the <00:09:21.265>whole <00:09:21.405>thing.

191
00:09:23.265 --> 00:09:25.805
<00:09:23.265>So <00:09:23.498>we <00:09:23.731>have <00:09:23.965>here <00:09:24.198>a <00:09:24.431>typical <00:09:24.665>prompt.

192
00:09:26.265 --> 00:09:28.445
<00:09:26.265>The <00:09:26.465>structure <00:09:26.665>is <00:09:26.865>a <00:09:27.064>little <00:09:27.265>bit, <00:09:27.465>uh, <00:09:27.665>inconvenient,

193
00:09:28.765 --> 00:09:31.365
<00:09:28.765>probably <00:09:29.045>it's <00:09:29.324>the <00:09:29.605>llama <00:09:29.885>three <00:09:30.165>specification.

194
00:09:32.185 --> 00:09:34.765
<00:09:32.185>So <00:09:32.359>what <00:09:32.535>you <00:09:32.709>do <00:09:32.885>in <00:09:33.060>your <00:09:33.234>prompt, <00:09:33.410>you <00:09:33.584>first

195
00:09:34.765 --> 00:09:38.805
<00:09:34.765>of <00:09:34.905>all <00:09:35.045>have <00:09:35.185>a <00:09:35.324>system <00:09:35.465>prompt <00:09:35.605>where <00:09:35.745>you <00:09:35.885>just <00:09:36.025>specify, <00:09:36.165>well,

196
00:09:38.805 --> 00:09:41.165
<00:09:38.805>you're <00:09:39.005>an <00:09:39.204>ai, <00:09:39.405>answer <00:09:39.604>this <00:09:39.805>question, <00:09:40.005>be <00:09:40.204>nice

197
00:09:41.165 --> 00:09:42.565
<00:09:41.165>and <00:09:41.417>polite, <00:09:41.669>and <00:09:41.921>something <00:09:42.173>like <00:09:42.425>that.

198
00:09:43.195 --> 00:09:44.445
<00:09:43.195>Then <00:09:43.476>you <00:09:43.757>have <00:09:44.038>the <00:09:44.320>context

199
00:09:44.945 --> 00:09:47.165
<00:09:44.945>and <00:09:45.178>specify <00:09:45.411>for <00:09:45.645>the <00:09:45.878>question, <00:09:46.111>the <00:09:46.345>context

200
00:09:47.355 --> 00:09:49.165
<00:09:47.355>that <00:09:47.555>should <00:09:47.755>be <00:09:47.955>used <00:09:48.155>to <00:09:48.355>answer <00:09:48.555>the <00:09:48.755>question.

201
00:09:50.105 --> 00:09:52.685
<00:09:50.105>And <00:09:50.385>here <00:09:50.665>you <00:09:50.945>see <00:09:51.225>something, <00:09:51.505>um,

202
00:09:52.975 --> 00:09:54.605
<00:09:52.975>which <00:09:53.255>is <00:09:53.535>specific <00:09:53.815>to <00:09:54.095>our <00:09:54.375>thing.

203
00:09:54.785 --> 00:09:59.205
<00:09:54.785>So <00:09:54.912>in <00:09:55.039>our <00:09:55.166>approach, <00:09:55.294>you <00:09:55.421>would <00:09:55.548>not <00:09:55.675>put <00:09:55.803>the <00:09:55.930>real <00:09:56.057>context <00:09:56.185>here,

204
00:09:59.505 --> 00:10:04.085
<00:09:59.505>but <00:09:59.660>you <00:09:59.816>can <00:09:59.971>use <00:10:00.127>a <00:10:00.282>variable <00:10:00.438>here, <00:10:00.593>uh, <00:10:00.749>yeah, <00:10:00.905>drawn

205
00:10:04.105 --> 00:10:07.285
<00:10:04.105>or <00:10:04.338>visualized <00:10:04.571>by <00:10:04.805>this, <00:10:05.038>uh, <00:10:05.271>open, <00:10:05.505>um,

206
00:10:08.895 --> 00:10:10.805
<00:10:08.895>curly <00:10:09.245>brackets <00:10:09.595>and <00:10:09.944>dot <00:10:10.295>context.

207
00:10:11.185 --> 00:10:14.405
<00:10:11.185>And <00:10:11.651>the, <00:10:12.118>the <00:10:12.584>solution

208
00:10:14.405 --> 00:10:18.405
<00:10:14.405>of <00:10:14.545>ours <00:10:14.685>will <00:10:14.824>put <00:10:14.964>here <00:10:15.105>the <00:10:15.245>content <00:10:15.385>from <00:10:15.525>the <00:10:15.665>vector <00:10:15.805>database

209
00:10:18.425 --> 00:10:20.685
<00:10:18.425>of <00:10:18.704>the <00:10:18.984>most <00:10:19.265>relevant, <00:10:19.545>uh, <00:10:19.824>documents.

210
00:10:21.105 --> 00:10:22.765
<00:10:21.105>And <00:10:21.305>then <00:10:21.505>you <00:10:21.705>have <00:10:21.905>the <00:10:22.105>user <00:10:22.305>part <00:10:22.505>here

211
00:10:22.935 --> 00:10:24.565
<00:10:22.935>where <00:10:23.214>you <00:10:23.494>put <00:10:23.775>the <00:10:24.055>collection <00:10:24.334>query.

212
00:10:24.665 --> 00:10:26.845
<00:10:24.665>So <00:10:24.839>that's <00:10:25.015>basically <00:10:25.189>the <00:10:25.365>query <00:10:25.540>you <00:10:25.714>want <00:10:25.890>to <00:10:26.064>answer

213
00:10:26.845 --> 00:10:28.165
<00:10:26.845>with <00:10:27.241>this <00:10:27.637>prompt, <00:10:28.033>right?

214
00:10:28.705 --> 00:10:29.925
<00:10:28.705>So <00:10:28.979>this <00:10:29.254>is <00:10:29.528>your <00:10:29.803>prompt,

215
00:10:30.665 --> 00:10:34.005
<00:10:30.665>and <00:10:30.839>then <00:10:31.015>you <00:10:31.189>have <00:10:31.365>to <00:10:31.540>specify <00:10:31.714>two <00:10:31.890>parameters <00:10:32.064>here.

216
00:10:34.385 --> 00:10:36.845
<00:10:34.385>The <00:10:34.585>first <00:10:34.785>of <00:10:34.985>one <00:10:35.185>them <00:10:35.385>is <00:10:35.585>collection <00:10:35.785>id.

217
00:10:36.855 --> 00:10:40.925
<00:10:36.855>Collection <00:10:37.010>ID <00:10:37.166>is <00:10:37.321>just <00:10:37.477>the <00:10:37.632>collection <00:10:37.788>in <00:10:37.943>the <00:10:38.099>vector <00:10:38.255>database

218
00:10:40.925 --> 00:10:44.645
<00:10:40.925>that <00:10:41.064>captures <00:10:41.204>your <00:10:41.344>data <00:10:41.484>and <00:10:41.625>you <00:10:41.765>just <00:10:41.905>specify <00:10:42.045>this <00:10:42.185>ID <00:10:42.324>here.

219
00:10:45.185 --> 00:10:49.285
<00:10:45.185>And <00:10:45.324>the <00:10:45.464>collection <00:10:45.604>query <00:10:45.744>is <00:10:45.885>the <00:10:46.025>user <00:10:46.165>query <00:10:46.305>of <00:10:46.444>the <00:10:46.584>query.

220
00:10:49.425 --> 00:10:51.165
<00:10:49.425>So <00:10:49.599>that's <00:10:49.775>the <00:10:49.949>query <00:10:50.125>that <00:10:50.300>you <00:10:50.474>want <00:10:50.650>to <00:10:50.824>answer.

221
00:10:52.625 --> 00:10:56.685
<00:10:52.625>And <00:10:52.752>in <00:10:52.879>this <00:10:53.006>way, <00:10:53.134>you <00:10:53.261>have <00:10:53.388>just <00:10:53.515>this <00:10:53.643>body <00:10:53.770>for <00:10:53.897>your <00:10:54.025>query

222
00:10:57.235 --> 00:10:59.805
<00:10:57.235>capturing <00:10:57.410>the <00:10:57.585>prompt <00:10:57.760>that <00:10:57.935>you <00:10:58.110>would <00:10:58.285>have, <00:10:58.460>uh, <00:10:58.635>had

223
00:10:59.805 --> 00:11:01.005
<00:10:59.805>to <00:11:00.165>write <00:11:00.525>beforehand <00:11:00.885>too,

224
00:11:01.225 --> 00:11:03.725
<00:11:01.225>but <00:11:01.458>you <00:11:01.691>only <00:11:01.925>specify <00:11:02.158>the <00:11:02.391>collection <00:11:02.625>id.

225
00:11:03.985 --> 00:11:07.245
<00:11:03.985>And <00:11:04.125>you <00:11:04.265>do <00:11:04.405>not <00:11:04.545>have <00:11:04.685>to <00:11:04.825>do <00:11:04.965>the <00:11:05.105>vector <00:11:05.245>database <00:11:05.385>retrieval,

226
00:11:07.945 --> 00:11:11.325
<00:11:07.945>get <00:11:08.072>the <00:11:08.199>content <00:11:08.326>from <00:11:08.454>there, <00:11:08.581>put <00:11:08.708>it <00:11:08.835>together <00:11:08.963>and <00:11:09.090>do <00:11:09.217>it <00:11:09.345>here.

227
00:11:11.475 --> 00:11:14.085
<00:11:11.475>That <00:11:11.675>will <00:11:11.875>all <00:11:12.075>be <00:11:12.275>done <00:11:12.475>by <00:11:12.675>our <00:11:12.875>solution.

228
00:11:14.185 --> 00:11:17.685
<00:11:14.185>So <00:11:14.312>this <00:11:14.439>is <00:11:14.566>your <00:11:14.694>body, <00:11:14.821>you <00:11:14.948>send <00:11:15.075>it <00:11:15.203>to <00:11:15.330>your <00:11:15.457>end <00:11:15.584>point,

229
00:11:18.065 --> 00:11:20.765
<00:11:18.065>and <00:11:18.240>as <00:11:18.415>a <00:11:18.590>result, <00:11:18.765>you <00:11:18.940>will <00:11:19.115>get <00:11:19.290>the <00:11:19.465>result

230
00:11:20.765 --> 00:11:22.325
<00:11:20.765>of <00:11:21.231>retrieval <00:11:21.698>augmented <00:11:22.165>generation.

231
00:11:22.385 --> 00:11:26.605
<00:11:22.385>So <00:11:22.540>basically <00:11:22.696>you <00:11:22.851>can <00:11:23.007>now <00:11:23.162>do <00:11:23.318>retrieve <00:11:23.473>augmented <00:11:23.629>generation <00:11:23.785>in

232
00:11:26.605 --> 00:11:28.005
<00:11:26.605>just <00:11:27.025>one <00:11:27.445>API <00:11:27.865>request.

233
00:11:28.145 --> 00:11:31.205
<00:11:28.145>And <00:11:28.319>we <00:11:28.495>think <00:11:28.670>that <00:11:28.845>simplifies <00:11:29.020>the <00:11:29.194>whole <00:11:29.370>approach, <00:11:29.545>uh,

234
00:11:31.235 --> 00:11:32.285
<00:11:31.235>once <00:11:31.471>again <00:11:31.707>a <00:11:31.943>little <00:11:32.180>bit.

235
00:11:33.305 --> 00:11:37.285
<00:11:33.305>So <00:11:33.538>that <00:11:33.771>said, <00:11:34.005>we <00:11:34.238>have <00:11:34.471>one <00:11:34.704>additional,

236
00:11:37.425 --> 00:11:38.805
<00:11:37.425>uh, <00:11:37.839>cool <00:11:38.252>point <00:11:38.666>here.

237
00:11:39.585 --> 00:11:43.445
<00:11:39.585>We <00:11:39.760>as <00:11:39.935>anos <00:11:40.110>are <00:11:40.285>one <00:11:40.460>of <00:11:40.635>the <00:11:40.810>European <00:11:40.985>hosts.

238
00:11:43.745 --> 00:11:45.965
<00:11:43.745>Um, <00:11:43.978>we <00:11:44.211>have <00:11:44.445>several <00:11:44.678>disease <00:11:44.911>in <00:11:45.145>Europe,

239
00:11:46.015 --> 00:11:47.045
<00:11:46.015>especially <00:11:46.478>in <00:11:46.942>Germany.

240
00:11:47.915 --> 00:11:51.085
<00:11:47.915>This <00:11:48.055>AI <00:11:48.194>model <00:11:48.334>hub <00:11:48.474>solution <00:11:48.615>that <00:11:48.755>I <00:11:48.895>talked <00:11:49.035>about <00:11:49.175>just <00:11:49.314>uh,

241
00:11:51.085 --> 00:11:55.085
<00:11:51.085>minutes <00:11:51.318>ago <00:11:51.551>is <00:11:51.785>placed <00:11:52.018>in <00:11:52.251>our, <00:11:52.485>uh,

242
00:11:55.465 --> 00:11:56.525
<00:11:55.465>DC <00:11:55.783>in, <00:11:56.101>uh, <00:11:56.419>Berlin.

243
00:11:56.705 --> 00:11:58.605
<00:11:56.705>So <00:11:56.938>it's <00:11:57.171>in <00:11:57.405>a <00:11:57.638>German <00:11:57.871>data <00:11:58.105>center.

244
00:11:59.185 --> 00:12:00.885
<00:11:59.185>And <00:11:59.651>we <00:12:00.118>in <00:12:00.584>addition,

245
00:12:00.895 --> 00:12:04.165
<00:12:00.895>offer <00:12:01.128>in <00:12:01.361>our <00:12:01.595>DC's <00:12:01.828>managed <00:12:02.061>Kubernetes <00:12:02.295>services.

246
00:12:04.545 --> 00:12:08.205
<00:12:04.545>So <00:12:04.672>what <00:12:04.799>you <00:12:04.926>can <00:12:05.054>now <00:12:05.181>do <00:12:05.308>is <00:12:05.435>you <00:12:05.563>can <00:12:05.690>implement <00:12:05.817>your <00:12:05.944>solution

247
00:12:10.275 --> 00:12:14.245
<00:12:10.275>with <00:12:10.475>Kubernetes, <00:12:10.675>orchestrating <00:12:10.875>this, <00:12:11.074>uh, <00:12:11.275>logic <00:12:11.475>from <00:12:11.675>here,

248
00:12:14.245 --> 00:12:16.565
<00:12:14.245>integrating <00:12:14.445>it <00:12:14.645>in <00:12:14.845>a <00:12:15.045>web <00:12:15.245>application <00:12:15.445>or <00:12:15.645>whatever,

249
00:12:17.185 --> 00:12:21.605
<00:12:17.185>and <00:12:17.340>have <00:12:17.496>this <00:12:17.651>all <00:12:17.807>run <00:12:17.962>in <00:12:18.118>our <00:12:18.273>infrastructure <00:12:18.429>at <00:12:18.584>iono.

250
00:12:21.825 --> 00:12:24.325
<00:12:21.825>And <00:12:22.025>it'll <00:12:22.225>never <00:12:22.425>leave <00:12:22.625>our <00:12:22.825>German <00:12:23.025>data <00:12:23.225>centers

251
00:12:24.385 --> 00:12:25.685
<00:12:24.385>or <00:12:24.677>never <00:12:24.970>leave, <00:12:25.262>uh, <00:12:25.555>Europe.

252
00:12:25.905 --> 00:12:30.085
<00:12:25.905>So <00:12:26.060>we <00:12:26.216>have <00:12:26.371>a <00:12:26.527>solution <00:12:26.682>here, <00:12:26.838>which <00:12:26.993>allows <00:12:27.149>you <00:12:27.305>to

253
00:12:30.785 --> 00:12:34.765
<00:12:30.785>do <00:12:30.985>generative <00:12:31.185>ai, <00:12:31.385>the <00:12:31.584>whole <00:12:31.785>use <00:12:31.985>case <00:12:32.185>end

254
00:12:34.765 --> 00:12:37.045
<00:12:34.765>to <00:12:34.998>end <00:12:35.231>in <00:12:35.465>our <00:12:35.698>German <00:12:35.931>data <00:12:36.165>centers.

255
00:12:37.705 --> 00:12:40.525
<00:12:37.705>And <00:12:37.905>now <00:12:38.105>I <00:12:38.305>would <00:12:38.505>hand <00:12:38.705>over <00:12:38.905>to <00:12:39.105>Kevin

256
00:12:40.825 --> 00:12:42.325
<00:12:40.825>and <00:12:41.017>he <00:12:41.210>will <00:12:41.403>talk <00:12:41.596>a <00:12:41.789>little <00:12:41.982>bit <00:12:42.175>about

257
00:12:42.875 --> 00:12:45.365
<00:12:42.875>what <00:12:43.075>the <00:12:43.275>future <00:12:43.475>might <00:12:43.675>be <00:12:43.875>or <00:12:44.075>will <00:12:44.275>be.

258
00:12:47.305 --> 00:12:48.445
<00:12:47.305>Uh, <00:12:47.818>thanks <00:12:48.331>Stefan.

259
00:12:49.025 --> 00:12:53.925
<00:12:49.025>Um, <00:12:49.199>so <00:12:49.375>one <00:12:49.550>of <00:12:49.725>the <00:12:49.900>major <00:12:50.074>limitations <00:12:50.250>with <00:12:50.425>the,

260
00:12:54.145 --> 00:12:57.845
<00:12:54.145>uh, <00:12:54.300>architecture <00:12:54.456>that <00:12:54.611>Stefan <00:12:54.767>just <00:12:54.922>talked <00:12:55.078>about, <00:12:55.233>um, <00:12:55.389>is <00:12:55.545>that,

261
00:12:58.265 --> 00:13:01.005
<00:12:58.265>uh, <00:12:58.465>the <00:12:58.665>Ionis <00:12:58.865>managed <00:12:59.064>Kubernetes <00:12:59.265>doesn't <00:12:59.465>provide <00:12:59.665>access

262
00:13:01.005 --> 00:13:02.285
<00:13:01.005>to <00:13:01.389>GPU <00:13:01.773>resources <00:13:02.156>itself.

263
00:13:02.305 --> 00:13:07.005
<00:13:02.305>So <00:13:02.421>this, <00:13:02.538>um, <00:13:02.655>a <00:13:02.771>I <00:13:02.888>AI <00:13:03.005>model <00:13:03.121>hub <00:13:03.238>that <00:13:03.354>they <00:13:03.471>have <00:13:03.588>sits <00:13:03.704>outside

264
00:13:07.005 --> 00:13:08.445
<00:13:07.005>of <00:13:07.329>your <00:13:07.653>managed <00:13:07.977>Kubernetes <00:13:08.301>cluster.

265
00:13:08.745 --> 00:13:11.045
<00:13:08.745>It <00:13:08.920>obviously <00:13:09.095>has <00:13:09.270>GPUs <00:13:09.445>that <00:13:09.620>are <00:13:09.795>on <00:13:09.970>the <00:13:10.145>hardware

266
00:13:11.045 --> 00:13:13.405
<00:13:11.045>that <00:13:11.219>it <00:13:11.395>runs <00:13:11.569>on <00:13:11.745>in <00:13:11.920>order <00:13:12.094>to <00:13:12.270>accelerate <00:13:12.444>any

267
00:13:13.405 --> 00:13:16.125
<00:13:13.405>of <00:13:13.545>the <00:13:13.685>AI <00:13:13.824>inferencing <00:13:13.964>that <00:13:14.105>you <00:13:14.245>are <00:13:14.385>going <00:13:14.525>to <00:13:14.665>be, <00:13:14.805>uh,

268
00:13:16.125 --> 00:13:18.485
<00:13:16.125>consuming <00:13:16.280>from <00:13:16.436>it <00:13:16.591>when <00:13:16.747>you <00:13:16.902>call <00:13:17.058>out <00:13:17.213>to <00:13:17.369>its <00:13:17.525>APIs.

269
00:13:18.785 --> 00:13:19.925
<00:13:18.785>But <00:13:18.990>you <00:13:19.195>as <00:13:19.400>a <00:13:19.605>user <00:13:19.810>of,

270
00:13:19.925 --> 00:13:22.645
<00:13:19.925>of <00:13:20.125>Kubernetes <00:13:20.324>on <00:13:20.525>their <00:13:20.724>managed <00:13:20.925>Kubernetes <00:13:21.125>platform <00:13:21.324>can't

271
00:13:22.805 --> 00:13:24.565
<00:13:22.805>actually <00:13:23.084>make <00:13:23.364>use <00:13:23.645>of <00:13:23.925>GPUs <00:13:24.204>yourself

272
00:13:24.915 --> 00:13:27.125
<00:13:24.915>that <00:13:25.115>doesn't <00:13:25.314>exist <00:13:25.515>in <00:13:25.714>their <00:13:25.915>managed <00:13:26.115>Kubernetes <00:13:26.314>offering.

273
00:13:27.585 --> 00:13:30.725
<00:13:27.585>And <00:13:27.712>so, <00:13:27.839>um, <00:13:27.966>what <00:13:28.094>we'd <00:13:28.221>like <00:13:28.348>to <00:13:28.475>start <00:13:28.603>doing <00:13:28.730>as, <00:13:28.857>as <00:13:28.985>a,

274
00:13:30.725 --> 00:13:33.805
<00:13:30.725>as <00:13:30.880>a <00:13:31.036>partner <00:13:31.191>of <00:13:31.347>Ionis, <00:13:31.502>is <00:13:31.658>to <00:13:31.813>bring <00:13:31.969>the <00:13:32.125>ability

275
00:13:33.985 --> 00:13:36.565
<00:13:33.985>to <00:13:34.218>consume <00:13:34.451>GPUs <00:13:34.685>inside <00:13:34.918>your <00:13:35.151>own <00:13:35.385>workloads,

276
00:13:36.565 --> 00:13:37.885
<00:13:36.565>in <00:13:36.862>their <00:13:37.159>managed <00:13:37.456>Kubernetes <00:13:37.753>offering.

277
00:13:38.425 --> 00:13:40.525
<00:13:38.425>Um, <00:13:38.658>where, <00:13:38.891>you <00:13:39.125>know, <00:13:39.358>instead <00:13:39.591>of <00:13:39.824>just,

278
00:13:40.525 --> 00:13:42.045
<00:13:40.525>you'd <00:13:40.720>still <00:13:40.915>have <00:13:41.111>the <00:13:41.306>ability <00:13:41.502>to <00:13:41.697>call <00:13:41.892>out

279
00:13:42.045 --> 00:13:43.925
<00:13:42.045>to <00:13:42.200>their <00:13:42.356>AI <00:13:42.511>model <00:13:42.667>hub <00:13:42.822>if <00:13:42.978>you, <00:13:43.133>if <00:13:43.289>you <00:13:43.444>want

280
00:13:43.925 --> 00:13:46.205
<00:13:43.925>to <00:13:44.080>just <00:13:44.236>do <00:13:44.391>it <00:13:44.547>over <00:13:44.702>this, <00:13:44.858>this <00:13:45.013>nice <00:13:45.169>simple <00:13:45.324>API,

281
00:13:46.505 --> 00:13:49.405
<00:13:46.505>but <00:13:46.632>you <00:13:46.759>can <00:13:46.886>also <00:13:47.014>now <00:13:47.141>launch <00:13:47.268>your <00:13:47.395>own <00:13:47.523>set <00:13:47.650>of <00:13:47.777>inferencing <00:13:47.905>jobs

282
00:13:49.505 --> 00:13:53.005
<00:13:49.505>or <00:13:49.660>training <00:13:49.816>jobs <00:13:49.971>on <00:13:50.127>your <00:13:50.282>own <00:13:50.438>infrastructure, <00:13:50.593>uh, <00:13:50.749>with <00:13:50.905>you

283
00:13:53.005 --> 00:13:53.965
<00:13:53.005>with <00:13:53.293>the <00:13:53.581>managed <00:13:53.869>Kubernetes

284
00:13:53.965 --> 00:13:55.085
<00:13:53.965>service <00:13:54.133>that <00:13:54.301>they've <00:13:54.469>brought <00:13:54.637>up <00:13:54.805>for <00:13:54.973>you.

285
00:13:55.865 --> 00:13:58.645
<00:13:55.865>Um, <00:13:56.005>and <00:13:56.145>the <00:13:56.285>way <00:13:56.425>that <00:13:56.565>we <00:13:56.705>enable <00:13:56.845>this <00:13:56.985>typically <00:13:57.125>in <00:13:57.265>Kubernetes

286
00:13:58.645 --> 00:14:00.125
<00:13:58.645>in <00:13:58.911>general, <00:13:59.177>is <00:13:59.444>through <00:13:59.710>a <00:13:59.977>component

287
00:14:00.125 --> 00:14:01.765
<00:14:00.125>that <00:14:00.358>we <00:14:00.591>call <00:14:00.825>the <00:14:01.058>the <00:14:01.291>GPU <00:14:01.525>operator.

288
00:14:02.585 --> 00:14:04.485
<00:14:02.585>Um, <00:14:02.725>and <00:14:02.865>so <00:14:03.005>just <00:14:03.145>really <00:14:03.285>quickly, <00:14:03.425>I <00:14:03.565>want <00:14:03.705>to <00:14:03.845>go <00:14:03.985>through

289
00:14:04.485 --> 00:14:06.245
<00:14:04.485>what <00:14:04.718>the <00:14:04.951>components <00:14:05.185>of <00:14:05.418>this <00:14:05.651>operator <00:14:05.885>are.

290
00:14:06.515 --> 00:14:09.085
<00:14:06.515>It's <00:14:06.715>basically <00:14:06.915>a <00:14:07.115>way <00:14:07.314>to <00:14:07.515>allow <00:14:07.715>you <00:14:07.915>to,

291
00:14:09.105 --> 00:14:12.045
<00:14:09.105>on <00:14:09.338>a <00:14:09.571>Kubernetes <00:14:09.805>cluster, <00:14:10.038>install <00:14:10.271>the <00:14:10.505>operator.

292
00:14:12.425 --> 00:14:14.485
<00:14:12.425>It <00:14:12.625>will <00:14:12.824>find <00:14:13.025>any <00:14:13.224>GPU <00:14:13.425>nodes <00:14:13.625>that <00:14:13.824>happen

293
00:14:14.485 --> 00:14:16.885
<00:14:14.485>to <00:14:14.660>exist <00:14:14.835>in <00:14:15.010>the <00:14:15.185>cluster, <00:14:15.360>and <00:14:15.535>it'll <00:14:15.710>install <00:14:15.885>all

294
00:14:16.885 --> 00:14:20.605
<00:14:16.885>of <00:14:17.040>the <00:14:17.196>software <00:14:17.351>necessary <00:14:17.507>for <00:14:17.662>you <00:14:17.818>to <00:14:17.973>start <00:14:18.129>running <00:14:18.285>workloads

295
00:14:20.605 --> 00:14:22.645
<00:14:20.605>that <00:14:20.885>request <00:14:21.165>access <00:14:21.445>to <00:14:21.725>GPUs <00:14:22.005>and,

296
00:14:22.745 --> 00:14:24.325
<00:14:22.745>and <00:14:22.978>run <00:14:23.211>your <00:14:23.445>inferencing <00:14:23.678>jobs <00:14:23.911>on <00:14:24.145>it.

297
00:14:24.545 --> 00:14:26.325
<00:14:24.545>And <00:14:24.745>the <00:14:24.944>components <00:14:25.145>that <00:14:25.344>make <00:14:25.545>up <00:14:25.745>this, <00:14:25.944>uh,

298
00:14:26.485 --> 00:14:28.045
<00:14:26.485>GPU <00:14:26.718>operator <00:14:26.951>are <00:14:27.185>the <00:14:27.418>ones <00:14:27.651>that <00:14:27.885>that,

299
00:14:28.075 --> 00:14:29.605
<00:14:28.075>that <00:14:28.247>you <00:14:28.419>see <00:14:28.591>here <00:14:28.763>on <00:14:28.935>the, <00:14:29.107>on <00:14:29.279>the <00:14:29.452>slide.

300
00:14:30.105 --> 00:14:33.445
<00:14:30.105>Um, <00:14:30.245>it <00:14:30.385>basically <00:14:30.525>consists <00:14:30.665>of <00:14:30.805>a <00:14:30.945>set <00:14:31.085>of <00:14:31.225>host <00:14:31.365>level <00:14:31.505>components,

301
00:14:33.655 --> 00:14:35.645
<00:14:33.655>which <00:14:33.888>are <00:14:34.121>the <00:14:34.355>NVIDIA <00:14:34.588>GPU <00:14:34.821>driver <00:14:35.055>itself.

302
00:14:35.645 --> 00:14:37.005
<00:14:35.645>So <00:14:35.889>as <00:14:36.134>the <00:14:36.379>operator <00:14:36.624>comes <00:14:36.869>online,

303
00:14:37.265 --> 00:14:38.925
<00:14:37.265>it <00:14:37.498>finds <00:14:37.731>a <00:14:37.965>node <00:14:38.198>that <00:14:38.431>has <00:14:38.665>GPUs.

304
00:14:39.385 --> 00:14:41.125
<00:14:39.385>It <00:14:39.665>will <00:14:39.944>go <00:14:40.225>ahead <00:14:40.505>and <00:14:40.785>kickstart

305
00:14:41.125 --> 00:14:42.165
<00:14:41.125>the <00:14:41.312>installation <00:14:41.499>of <00:14:41.686>all <00:14:41.873>of <00:14:42.060>this.

306
00:14:42.165 --> 00:14:44.205
<00:14:42.165>So <00:14:42.320>it'll <00:14:42.476>get <00:14:42.631>your <00:14:42.787>GPU <00:14:42.942>driver <00:14:43.098>running <00:14:43.253>on <00:14:43.409>the <00:14:43.564>host.

307
00:14:44.695 --> 00:14:46.845
<00:14:44.695>It'll <00:14:44.895>install <00:14:45.095>a <00:14:45.295>component <00:14:45.495>called <00:14:45.695>the <00:14:45.895>NVIDIA <00:14:46.095>Container

308
00:14:46.845 --> 00:14:48.885
<00:14:46.845>Toolkit, <00:14:47.195>which <00:14:47.545>enables <00:14:47.895>GPU <00:14:48.245>support

309
00:14:49.145 --> 00:14:50.485
<00:14:49.145>inside <00:14:49.547>containers <00:14:49.949>in <00:14:50.351>general.

310
00:14:50.865 --> 00:14:54.245
<00:14:50.865>So <00:14:51.020>if <00:14:51.176>you've <00:14:51.331>ever <00:14:51.487>interacted <00:14:51.642>with, <00:14:51.798>uh, <00:14:51.953>containers <00:14:52.109>via <00:14:52.265>Docker

311
00:14:54.245 --> 00:14:56.605
<00:14:54.245>or <00:14:54.385>some <00:14:54.525>other <00:14:54.665>command <00:14:54.805>line <00:14:54.945>on <00:14:55.085>your, <00:14:55.225>on <00:14:55.365>your <00:14:55.505>local <00:14:55.645>machine,

312
00:14:56.865 --> 00:14:58.725
<00:14:56.865>uh, <00:14:57.065>the <00:14:57.265>NVIDIA <00:14:57.465>container <00:14:57.665>toolkit <00:14:57.865>is <00:14:58.065>the <00:14:58.265>component

313
00:14:58.725 --> 00:15:00.965
<00:14:58.725>that <00:14:58.900>allows <00:14:59.075>that <00:14:59.250>to <00:14:59.425>work <00:14:59.600>properly <00:14:59.775>under <00:14:59.950>the <00:15:00.125>hood.

314
00:15:01.305 --> 00:15:02.365
<00:15:01.305>Um, <00:15:01.464>and <00:15:01.622>then <00:15:01.781>there's <00:15:01.941>also <00:15:02.100>a <00:15:02.259>set

315
00:15:02.365 --> 00:15:04.165
<00:15:02.365>of <00:15:02.598>Kubernetes <00:15:02.831>specific <00:15:03.065>components <00:15:03.298>that <00:15:03.531>the <00:15:03.765>pu

316
00:15:04.275 --> 00:15:06.205
<00:15:04.275>that <00:15:04.625>the <00:15:04.975>operator <00:15:05.324>orchestrates, <00:15:05.675>such

317
00:15:06.205 --> 00:15:09.125
<00:15:06.205>as <00:15:06.405>the <00:15:06.605>CS <00:15:06.805>device <00:15:07.005>plugin, <00:15:07.205>GPU <00:15:07.405>feature <00:15:07.605>discovery,

318
00:15:09.745 --> 00:15:12.325
<00:15:09.745>NVIDIA <00:15:10.025>MIG <00:15:10.305>Manager, <00:15:10.585>DCGM <00:15:10.865>exporter, <00:15:11.145>some

319
00:15:12.325 --> 00:15:13.405
<00:15:12.325>of <00:15:12.446>you <00:15:12.568>might <00:15:12.689>be <00:15:12.811>familiar <00:15:12.932>with <00:15:13.054>some <00:15:13.175>of <00:15:13.297>these

320
00:15:13.405 --> 00:15:14.565
<00:15:13.405>components, <00:15:13.753>some <00:15:14.101>might <00:15:14.449>not.

321
00:15:14.825 --> 00:15:16.685
<00:15:14.825>But <00:15:14.965>the <00:15:15.105>point <00:15:15.245>is <00:15:15.385>that <00:15:15.525>there's <00:15:15.665>a <00:15:15.805>whole <00:15:15.945>bunch <00:15:16.085>of <00:15:16.225>software

322
00:15:16.685 --> 00:15:18.925
<00:15:16.685>that <00:15:16.885>needs <00:15:17.084>to <00:15:17.285>get <00:15:17.484>orchestrated <00:15:17.685>by <00:15:17.885>this <00:15:18.084>operator,

323
00:15:19.145 --> 00:15:21.405
<00:15:19.145>and <00:15:19.300>by <00:15:19.456>doing <00:15:19.611>a <00:15:19.767>helmet <00:15:19.922>install <00:15:20.078>on <00:15:20.233>your <00:15:20.389>Kubernetes <00:15:20.545>cluster,

324
00:15:21.405 --> 00:15:23.165
<00:15:21.405>everything's <00:15:21.685>kind <00:15:21.964>of <00:15:22.245>managed <00:15:22.525>for <00:15:22.805>you

325
00:15:23.165 --> 00:15:24.325
<00:15:23.165>behind <00:15:23.513>the <00:15:23.861>scenes <00:15:24.209>automatically.

326
00:15:25.385 --> 00:15:28.445
<00:15:25.385>Um, <00:15:25.525>and <00:15:25.665>so <00:15:25.805>an <00:15:25.944>example <00:15:26.085>of <00:15:26.225>what <00:15:26.365>this <00:15:26.505>might <00:15:26.645>look <00:15:26.785>like,

327
00:15:28.685 --> 00:15:29.685
<00:15:28.685>assuming <00:15:28.984>you've, <00:15:29.285>you <00:15:29.584>know,

328
00:15:29.685 --> 00:15:31.805
<00:15:29.685>taken <00:15:29.964>the <00:15:30.244>Ionis <00:15:30.525>managed <00:15:30.805>Kubernetes <00:15:31.084>platform,

329
00:15:31.805 --> 00:15:33.245
<00:15:31.805>you've <00:15:32.021>got <00:15:32.237>the <00:15:32.453>GPU <00:15:32.669>machines <00:15:32.885>out <00:15:33.101>there,

330
00:15:33.245 --> 00:15:35.245
<00:15:33.245>you <00:15:33.445>deploy <00:15:33.645>the <00:15:33.845>operator, <00:15:34.045>what <00:15:34.245>you'll <00:15:34.445>be <00:15:34.645>able

331
00:15:35.245 --> 00:15:37.405
<00:15:35.245>to <00:15:35.372>do <00:15:35.499>now <00:15:35.626>in <00:15:35.754>your <00:15:35.881>pod <00:15:36.008>spec <00:15:36.135>is <00:15:36.263>run <00:15:36.390>something <00:15:36.517>like <00:15:36.645>this.

332
00:15:37.865 --> 00:15:39.885
<00:15:37.865>So <00:15:38.040>you <00:15:38.215>can <00:15:38.390>say, <00:15:38.565>okay, <00:15:38.740>I've <00:15:38.915>got <00:15:39.090>a <00:15:39.265>pod.

333
00:15:40.205 --> 00:15:44.605
<00:15:40.205>I <00:15:40.332>want <00:15:40.459>to <00:15:40.586>run <00:15:40.714>the <00:15:40.841>gen <00:15:40.968>AI <00:15:41.095>sd, <00:15:41.223>uh, <00:15:41.350>NVIDIA <00:15:41.477>inference <00:15:41.605>server,

334
00:15:45.225 --> 00:15:47.725
<00:15:45.225>uh, <00:15:45.505>in <00:15:45.785>Nvidia, <00:15:46.065>uh, <00:15:46.345>inference <00:15:46.625>microservice,

335
00:15:48.185 --> 00:15:50.325
<00:15:48.185>and <00:15:48.359>I <00:15:48.535>want <00:15:48.709>that <00:15:48.885>to <00:15:49.060>run <00:15:49.234>on <00:15:49.410>two <00:15:49.584>GPUs.

336
00:15:50.705 --> 00:15:52.845
<00:15:50.705>And <00:15:50.845>so <00:15:50.985>a <00:15:51.125>simple <00:15:51.265>YAMA <00:15:51.405>file <00:15:51.545>like <00:15:51.685>this <00:15:51.825>will <00:15:51.965>allow <00:15:52.105>you

337
00:15:52.845 --> 00:15:55.845
<00:15:52.845>to <00:15:53.020>bring <00:15:53.195>in, <00:15:53.370>uh, <00:15:53.545>this <00:15:53.720>Nvidia, <00:15:53.895>uh, <00:15:54.070>inference, <00:15:54.245>microservice

338
00:15:56.345 --> 00:15:57.845
<00:15:56.345>and <00:15:56.615>run <00:15:56.885>whatever <00:15:57.155>workload <00:15:57.425>you <00:15:57.695>happen

339
00:15:57.905 --> 00:15:59.165
<00:15:57.905>to, <00:15:58.093>to <00:15:58.283>have <00:15:58.472>on <00:15:58.661>top <00:15:58.849>of <00:15:59.039>that.

340
00:15:59.465 --> 00:16:01.485
<00:15:59.465>Um, <00:15:59.620>I'm <00:15:59.776>not <00:15:59.931>showing <00:16:00.087>in <00:16:00.242>the <00:16:00.398>slide <00:16:00.553>here <00:16:00.709>the <00:16:00.865>workload

341
00:16:01.485 --> 00:16:03.165
<00:16:01.485>that <00:16:01.660>you <00:16:01.835>would <00:16:02.010>actually <00:16:02.185>invoke, <00:16:02.360>but <00:16:02.535>the, <00:16:02.710>you <00:16:02.885>know,

342
00:16:03.165 --> 00:16:04.245
<00:16:03.165>I'm <00:16:03.327>just <00:16:03.489>trying <00:16:03.651>to <00:16:03.813>show <00:16:03.975>from <00:16:04.137>a,

343
00:16:04.245 --> 00:16:05.485
<00:16:04.245>from <00:16:04.524>a <00:16:04.803>high <00:16:05.082>level <00:16:05.361>perspective

344
00:16:05.945 --> 00:16:07.365
<00:16:05.945>how <00:16:06.200>you <00:16:06.456>would <00:16:06.711>actually <00:16:06.967>invoke <00:16:07.223>this

345
00:16:07.365 --> 00:16:09.765
<00:16:07.365>and <00:16:07.540>how <00:16:07.715>simple <00:16:07.890>the, <00:16:08.065>the <00:16:08.240>API <00:16:08.415>would <00:16:08.590>actually <00:16:08.765>be.

346
00:16:10.635 --> 00:16:14.455
<00:16:10.635>Um, <00:16:10.790>if <00:16:10.946>you're <00:16:11.101>not <00:16:11.257>familiar <00:16:11.412>with, <00:16:11.568>uh, <00:16:11.723>uh, <00:16:11.879>these, <00:16:12.035>uh,

347
00:16:14.595 --> 00:16:18.575
<00:16:14.595>NVIDIA <00:16:14.770>inference <00:16:14.945>microservices, <00:16:15.120>um, <00:16:15.295>if <00:16:15.470>you <00:16:15.645>watched <00:16:15.820>the, <00:16:15.995>the,

348
00:16:18.755 --> 00:16:23.415
<00:16:18.755>uh, <00:16:18.910>our, <00:16:19.066>our <00:16:19.221>CEO's <00:16:19.377>keynote <00:16:19.532>from, <00:16:19.688>uh, <00:16:19.843>GTC, <00:16:19.999>our <00:16:20.155>flagship,

349
00:16:23.555 --> 00:16:25.895
<00:16:23.555>uh, <00:16:23.834>GPU <00:16:24.114>conference <00:16:24.395>last <00:16:24.675>March, <00:16:24.954>uh,

350
00:16:25.895 --> 00:16:27.695
<00:16:25.895>that's <00:16:26.069>when <00:16:26.245>we <00:16:26.420>first <00:16:26.595>introduced <00:16:26.770>the <00:16:26.944>notion <00:16:27.120>of <00:16:27.295>these,

351
00:16:27.695 --> 00:16:29.335
<00:16:27.695>these <00:16:28.161>Nvidia <00:16:28.628>inference <00:16:29.095>microservices,

352
00:16:29.395 --> 00:16:31.535
<00:16:29.395>or <00:16:29.595>nims, <00:16:29.795>as <00:16:29.995>we <00:16:30.194>call <00:16:30.395>them <00:16:30.595>for <00:16:30.795>short.

353
00:16:31.955 --> 00:16:34.855
<00:16:31.955>Um, <00:16:32.130>they <00:16:32.305>basically <00:16:32.480>allow <00:16:32.655>you <00:16:32.830>to, <00:16:33.005>uh, <00:16:33.180>run, <00:16:33.355>uh,

354
00:16:35.075 --> 00:16:36.175
<00:16:35.075>AI <00:16:35.405>inferencing <00:16:35.735>at <00:16:36.064>scale.

355
00:16:36.205 --> 00:16:38.655
<00:16:36.205>It's <00:16:36.332>a <00:16:36.459>way <00:16:36.586>to <00:16:36.714>take <00:16:36.841>a <00:16:36.968>container, <00:16:37.095>deploy <00:16:37.223>it <00:16:37.350>on <00:16:37.477>a <00:16:37.605>cluster,

356
00:16:39.035 --> 00:16:41.815
<00:16:39.035>and <00:16:39.209>now <00:16:39.385>you've <00:16:39.560>got <00:16:39.735>access <00:16:39.910>to <00:16:40.084>the <00:16:40.260>APIs, <00:16:40.435>uh,

357
00:16:41.815 --> 00:16:44.135
<00:16:41.815>that <00:16:42.015>you <00:16:42.215>would <00:16:42.415>expect <00:16:42.615>from <00:16:42.815>the <00:16:43.015>specific <00:16:43.215>microservice

358
00:16:44.135 --> 00:16:45.175
<00:16:44.135>that <00:16:44.369>you <00:16:44.603>want <00:16:44.837>access <00:16:45.070>to.

359
00:16:45.675 --> 00:16:47.975
<00:16:45.675>And <00:16:45.830>with <00:16:45.986>these <00:16:46.141>additions <00:16:46.297>that <00:16:46.452>we <00:16:46.608>can <00:16:46.763>add <00:16:46.919>to <00:16:47.074>the,

360
00:16:47.995 --> 00:16:50.255
<00:16:47.995>to <00:16:48.228>the <00:16:48.461>Ionis, <00:16:48.695>uh, <00:16:48.928>managed <00:16:49.161>Kubernetes <00:16:49.395>platform,

361
00:16:50.255 --> 00:16:52.935
<00:16:50.255>you'll <00:16:50.382>be <00:16:50.509>able <00:16:50.636>to, <00:16:50.764>you <00:16:50.891>know, <00:16:51.018>not <00:16:51.145>only <00:16:51.273>use <00:16:51.400>their <00:16:51.527>AI <00:16:51.655>model

362
00:16:53.035 --> 00:16:55.455
<00:16:53.035>hub, <00:16:53.209>but <00:16:53.385>also <00:16:53.560>bring <00:16:53.735>along <00:16:53.910>your <00:16:54.084>own <00:16:54.260>inference <00:16:54.435>services,

363
00:16:55.835 --> 00:16:56.935
<00:16:55.835>uh, <00:16:56.165>based <00:16:56.495>on <00:16:56.824>nims.

364
00:16:57.395 --> 00:17:00.215
<00:16:57.395>Um, <00:16:57.535>and <00:16:57.675>one <00:16:57.814>of <00:16:57.954>the, <00:16:58.095>the <00:16:58.235>cool <00:16:58.375>things <00:16:58.515>that, <00:16:58.655>uh, <00:16:58.795>that

365
00:17:00.815 --> 00:17:02.735
<00:17:00.815>I <00:17:01.048>know <00:17:01.281>it's <00:17:01.515>been <00:17:01.748>announced <00:17:01.981>very <00:17:02.215>recently,

366
00:17:02.835 --> 00:17:05.135
<00:17:02.835>but <00:17:03.035>I'll, <00:17:03.235>I'll <00:17:03.435>just <00:17:03.635>reiterate <00:17:03.835>it <00:17:04.035>here, <00:17:04.235>is

367
00:17:05.135 --> 00:17:08.655
<00:17:05.135>that <00:17:05.310>we're <00:17:05.484>now <00:17:05.660>building <00:17:05.835>a <00:17:06.010>separate <00:17:06.185>operator <00:17:06.359>to <00:17:06.535>orchestrate,

368
00:17:08.955 --> 00:17:11.095
<00:17:08.954>uh, <00:17:09.234>these <00:17:09.514>NIMS <00:17:09.794>on, <00:17:10.074>on <00:17:10.355>Kubernetes.

369
00:17:11.555 --> 00:17:13.655
<00:17:11.555>Um, <00:17:11.710>which <00:17:11.866>means <00:17:12.021>that, <00:17:12.177>you <00:17:12.332>know, <00:17:12.488>now <00:17:12.643>you <00:17:12.799>won't <00:17:12.955>have

370
00:17:13.655 --> 00:17:15.775
<00:17:13.655>to <00:17:13.795>worry <00:17:13.935>about <00:17:14.075>the <00:17:14.214>details <00:17:14.355>of <00:17:14.494>how <00:17:14.635>do <00:17:14.774>I, <00:17:14.915>you <00:17:15.055>know,

371
00:17:15.775 --> 00:17:16.855
<00:17:15.775>both <00:17:16.099>run <00:17:16.423>my <00:17:16.747>containers

372
00:17:16.995 --> 00:17:18.535
<00:17:16.994>and <00:17:17.225>bring <00:17:17.456>the <00:17:17.687>models <00:17:17.919>into <00:17:18.150>the <00:17:18.381>cluster.

373
00:17:18.565 --> 00:17:21.815
<00:17:18.565>This, <00:17:18.720>this, <00:17:18.876>this <00:17:19.031>NIM <00:17:19.187>operator <00:17:19.342>will <00:17:19.498>orchestrate <00:17:19.653>all <00:17:19.809>of <00:17:19.965>that

374
00:17:21.835 --> 00:17:24.215
<00:17:21.835>for <00:17:22.010>you <00:17:22.185>behind <00:17:22.360>the <00:17:22.535>scenes <00:17:22.710>just <00:17:22.885>as <00:17:23.060>the <00:17:23.235>GPU

375
00:17:24.735 --> 00:17:26.055
<00:17:24.734>operator <00:17:25.130>orchestrates <00:17:25.527>getting <00:17:25.923>all

376
00:17:26.055 --> 00:17:27.055
<00:17:26.055>of <00:17:26.355>your <00:17:26.655>software <00:17:26.955>installed.

377
00:17:27.275 --> 00:17:29.295
<00:17:27.275>The <00:17:27.508>NIM <00:17:27.741>operator <00:17:27.975>will <00:17:28.208>orchestrate <00:17:28.441>getting <00:17:28.675>all

378
00:17:29.295 --> 00:17:32.215
<00:17:29.295>of <00:17:29.450>your <00:17:29.606>Nvidia <00:17:29.761>Info <00:17:29.917>Micro <00:17:30.072>in <00:17:30.228>Infras <00:17:30.383>inference <00:17:30.539>micro <00:17:30.695>servers

379
00:17:32.325 --> 00:17:33.495
<00:17:32.325>installed <00:17:32.851>and, <00:17:33.378>and,

380
00:17:33.835 --> 00:17:36.135
<00:17:33.835>and <00:17:33.962>available <00:17:34.089>to <00:17:34.216>you <00:17:34.344>to <00:17:34.471>call <00:17:34.598>out <00:17:34.725>from <00:17:34.853>your <00:17:34.980>other <00:17:35.107>pods <00:17:35.235>via

381
00:17:36.135 --> 00:17:38.415
<00:17:36.135>the <00:17:36.415>APIs <00:17:36.695>that <00:17:36.975>they, <00:17:37.254>um, <00:17:37.535>provide.

382
00:17:40.445 --> 00:17:43.625
<00:17:40.445>Um, <00:17:40.561>and <00:17:40.678>that's <00:17:40.794>pretty <00:17:40.911>much <00:17:41.028>it <00:17:41.145>for, <00:17:41.261>for <00:17:41.378>my <00:17:41.495>part <00:17:41.611>of <00:17:41.728>this <00:17:41.845>talk.

383
00:17:44.005 --> 00:17:47.385
<00:17:44.005>Um, <00:17:44.180>if <00:17:44.355>you <00:17:44.530>want <00:17:44.705>to <00:17:44.880>stick <00:17:45.055>around, <00:17:45.230>uh, <00:17:45.405>myself

384
00:17:47.405 --> 00:17:48.665
<00:17:47.405>and <00:17:47.631>my <00:17:47.858>colleague <00:17:48.085>Evan <00:17:48.312>over <00:17:48.539>here,

385
00:17:48.665 --> 00:17:50.560
<00:17:48.665>we're <00:17:48.898>gonna <00:17:49.131>be <00:17:49.365>giving <00:17:49.598>a <00:17:49.831>talk <00:17:50.065>immediately

386
00:17:50.560 --> 00:17:52.010
<00:17:50.560>after <00:17:50.821>this <00:17:51.081>on <00:17:51.343>this <00:17:51.604>same <00:17:51.865>stage,

387
00:17:52.215 --> 00:17:54.805
<00:17:52.215>which <00:17:52.415>dives <00:17:52.615>into <00:17:52.814>the <00:17:53.014>lower <00:17:53.215>level <00:17:53.415>details <00:17:53.615>about

388
00:17:54.865 --> 00:17:57.085
<00:17:54.865>how <00:17:55.215>GPU <00:17:55.565>support <00:17:55.915>in <00:17:56.265>containers

389
00:17:57.085 --> 00:17:58.645
<00:17:57.085>and <00:17:57.365>Kubernetes <00:17:57.645>works <00:17:57.925>under <00:17:58.205>the <00:17:58.485>hood.

390
00:17:58.705 --> 00:18:00.765
<00:17:58.705>So <00:17:58.905>if <00:17:59.105>you've <00:17:59.304>ever <00:17:59.504>wondered, <00:17:59.705>like, <00:17:59.905>you <00:18:00.105>know,

391
00:18:00.825 --> 00:18:03.725
<00:18:00.825>how <00:18:00.965>does <00:18:01.105>it <00:18:01.245>actually <00:18:01.385>work, <00:18:01.525>come <00:18:01.665>to <00:18:01.805>the <00:18:01.945>talk <00:18:02.085>and, <00:18:02.225>and,

392
00:18:03.725 --> 00:18:04.885
<00:18:03.725>and <00:18:03.933>we'll <00:18:04.142>explain <00:18:04.351>that <00:18:04.560>in <00:18:04.769>detail.

393
00:18:05.545 --> 00:18:09.485
<00:18:05.545>Um, <00:18:05.700>I <00:18:05.856>also <00:18:06.011>encourage <00:18:06.167>you <00:18:06.322>to, <00:18:06.478>uh, <00:18:06.633>tomorrow <00:18:06.789>at, <00:18:06.945>uh,

394
00:18:09.545 --> 00:18:12.405
<00:18:09.545>2:20 <00:18:09.720>PM <00:18:09.895>on <00:18:10.070>the <00:18:10.245>main <00:18:10.420>stage, <00:18:10.595>um, <00:18:10.770>another <00:18:10.945>one

395
00:18:12.405 --> 00:18:13.885
<00:18:12.405>of <00:18:12.627>our <00:18:12.849>colleagues, <00:18:13.071>an <00:18:13.293>kit <00:18:13.514>over <00:18:13.737>here.

396
00:18:14.315 --> 00:18:16.165
<00:18:14.315>He's <00:18:14.595>gonna <00:18:14.875>be <00:18:15.155>talking <00:18:15.435>about <00:18:15.715>NIMS

397
00:18:16.185 --> 00:18:17.685
<00:18:16.185>and <00:18:16.409>some <00:18:16.635>of <00:18:16.860>the <00:18:17.085>details <00:18:17.310>behind <00:18:17.534>some

398
00:18:17.685 --> 00:18:19.325
<00:18:17.685>of <00:18:17.918>the <00:18:18.151>other <00:18:18.385>offerings <00:18:18.618>that <00:18:18.851>NVIDIA <00:18:19.085>has.

399
00:18:19.665 --> 00:18:21.045
<00:18:19.665>Uh, <00:18:19.913>that's, <00:18:20.161>you <00:18:20.410>know, <00:18:20.658>propelled <00:18:20.907>gen

400
00:18:21.105 --> 00:18:22.125
<00:18:21.105>AI <00:18:21.334>forward <00:18:21.564>into <00:18:21.793>the <00:18:22.023>future.

401
00:18:22.945 --> 00:18:25.085
<00:18:22.945>So <00:18:23.225>thank <00:18:23.504>you <00:18:23.784>for <00:18:24.064>your <00:18:24.345>attention

402
00:18:25.585 --> 00:18:28.125
<00:18:25.585>and <00:18:25.740>we'll, <00:18:25.896>we'll <00:18:26.051>take <00:18:26.207>any <00:18:26.362>questions <00:18:26.518>you <00:18:26.673>guys <00:18:26.829>have <00:18:26.985>now.

403
00:18:28.145 --> 00:18:28.565
Thanks.