title
stringlengths 12
112
| published
stringlengths 19
23
| url
stringlengths 28
28
| video_id
stringlengths 11
11
| channel_id
stringclasses 5
values | id
stringlengths 16
31
| text
stringlengths 0
596
| start
float64 0
37.8k
| end
float64 2.18
37.8k
|
|---|---|---|---|---|---|---|---|---|
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t628.96
|
the sentence transformers people have actually trained sentence transformers that can use all
| 628.96 | 642.64 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t635.92
|
more than 50 languages at once. And the performance is good. It's not just that they, you know,
| 635.92 | 650.16 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t642.64
|
managed to get a few phrases correct. The performance is actually quite good. So I think,
| 642.64 | 655.76 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t650.16
|
you know, it's pretty impressive. And, you know, the training time for these is super quick,
| 650.16 | 664.8 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t656.56
|
as we'll see. And like I said, it's using just translation data, parallel data, which is reasonably
| 656.56 | 673.76 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t664.8
|
easy to get for almost every language. So I think that's pretty useful. Now, well, let's have a look
| 664.8 | 680.72 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t673.76
|
at what that multi-lingual knowledge distillation training process actually looks like. So it's what
| 673.76 | 687.2 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t680.7199999999999
|
we have here. So same example as before. I've got, I like plants this time and mi piaccia
| 680.72 | 694.8 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t687.2
|
no le piante, which is again, the same thing in Italian. Now we have both of those. We have a
| 687.2 | 702.08 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t694.8000000000001
|
teacher model and a student model. Now, when we say knowledge distillation, that means where you
| 694.8 | 708.96 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t702.08
|
basically take one model and you distill the knowledge from that one model into another model
| 702.08 | 714.72 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t708.96
|
here. The model that already knows some of the stuff that we want, that we want to distill
| 708.96 | 722.48 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t714.72
|
knowledge from is called the teacher model. Okay. Now the teacher model in this case is going to be
| 714.72 | 726.96 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t722.48
|
a monolingual model. So it's probably going to be a sentence transformer. That's very good
| 722.48 | 736.4 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t727.52
|
at English tests only. And what we do is we take the student model, which is going to be,
| 727.52 | 742.32 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t736.4
|
it doesn't have to be a sentence transformer. It's just a pre-trained transform model. We'll be using
| 736.4 | 751.52 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t742.32
|
XLM Roberta later on, and it needs to be capable of understanding multiple languages. Okay. So
| 742.32 | 757.6 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t752.24
|
in this case, we feed the English sentence into both our teacher model and student model. And
| 752.24 | 763.44 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t757.6
|
then we optimize the student model to reduce the difference between the two vectors output
| 757.6 | 771.12 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t764.08
|
by those two models. And that makes the student model almost mimic the monolingual aspect of the
| 764.08 | 777.76 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t771.12
|
teacher model. But then we take it a little further and we process the Italian or the target language
| 771.12 | 783.68 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t777.76
|
through the student model. And then we do the same thing. So we try to reduce the difference
| 777.76 | 789.52 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t783.68
|
between the Italian vector and the teacher's English vector. And what we're doing there is
| 783.68 | 796.48 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t789.52
|
making the student model mimic the teacher for a different language. Okay. So through that process,
| 789.52 | 804 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t796.48
|
you can add more and more languages to a student model, which mimics your teacher model, which,
| 796.48 | 813.92 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t805.04
|
yeah, I mean, it seems at least really simple just to think of it like that in my opinion anyway,
| 805.04 | 824.72 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t815.2
|
but it works really well. So it's a very cool technique in my opinion. I do like it. So just
| 815.2 | 831.36 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t824.72
|
a more visual way of going through that. We have these different circles, they represent different
| 824.72 | 837.2 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t831.36
|
language tasks or different languages, but pretty similar or the same task in each one of those.
| 831.36 | 844.56 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t837.9200000000001
|
We have our monolingual teacher model and that can perform on one of these languages, but fails
| 837.92 | 851.6 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t844.5600000000001
|
on the others. We take that monolingual model or our teacher model, and then we also take a pre-trained
| 844.56 | 857.2 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t851.6
|
multilingual model. So the important thing here is that it can handle new languages. Like I said,
| 851.6 | 863.28 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t857.2
|
with Excel learning Roberta, this is our student. We perform multilingual knowledge distillation,
| 857.2 | 868.32 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t863.28
|
meaning the student learns how the teacher performs well on the single task by mimicking
| 863.28 | 874.16 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t868.32
|
its sentence vector outputs. The student then performs this mimicry across multiple languages.
| 868.32 | 881.28 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t874.16
|
And then hopefully the student model can now perform across all of the languages that we are
| 874.16 | 889.04 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t881.28
|
wanting to train on. That's how the multilingual knowledge distillation works. Let's have a look
| 881.28 | 895.92 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t889.04
|
at that in code. Okay. So we're in our code here. And the first thing I'm going to do is actually
| 889.04 | 904 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t895.92
|
get our data. So in the paper that introduced the multilingual knowledge distillation,
| 895.92 | 914.72 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t904.0
|
Reimers and Guravich use the focus partly on this TED subtitles data. So we know TED talks,
| 904 | 919.92 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t914.7199999999999
|
they're just low talks where people present on a particular topic, usually pretty interesting.
| 914.72 | 928.8 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t919.92
|
And those TED talks have subtitles in loads of different languages. So they scraped that subtitle
| 919.92 | 936.24 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t928.8
|
data and use that as sentence pairs for the different languages. Okay. So that's the parallel
| 928.8 | 944.24 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t936.24
|
data. Now, what I'm going to do is use Hug and Face Transformers to download that. So we just
| 936.24 | 950.08 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t944.24
|
import datasets here. Sorry, I said Hug and Face Transformers actually Hug and Face datasets here.
| 944.24 | 957.84 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t951.36
|
So import datasets and I'm going to load that dataset and just have a look at what the structure
| 951.36 | 964.48 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t957.84
|
of that dataset is. So it's the TEDMortar and I'm just getting the training data here. You see in
| 957.84 | 971.6 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t964.48
|
here we have this features, translations and talk name. Now it's not really very clear.
| 964.48 | 978.72 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t971.6
|
But inside the translations data, we have the language tag. So these are language codes,
| 971.6 | 983.28 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t978.72
|
ISO language codes. If you type that into Google, they'll pop up. If you don't know which one,
| 978.72 | 993.52 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t983.9200000000001
|
which are which. And below we also have in here, it's not very clear again. So if I come here,
| 983.92 | 998.32 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t993.52
|
we have translations and each one of those corresponds to the language code up here.
| 993.52 | 1,007.28 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t998.32
|
Okay. So if we came here, we see EN English and we find it here. Okay. And then we also have
| 998.32 | 1,015.04 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1007.2800000000001
|
talk name. It's not really important for us. So we can get the index of our English text. We need
| 1,007.28 | 1,022 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1015.0400000000001
|
to extract that for our source language. So we extract that, we get a number four. So we're
| 1,015.04 | 1,028.24 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1022.0
|
going into those language pairs, finding EN. And then we use that index to get the corresponding
| 1,022 | 1,035.76 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1028.24
|
translation, which is here. And then we'd use that to create all of our pairs. Now here,
| 1,028.24 | 1,041.68 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1035.76
|
I've just created loads of pairs. This first one, so this is English to Arabic. But if we have a
| 1,035.76 | 1,046.56 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1041.68
|
load, there's actually loads of pairs here. So we have 27 in total, which is obviously quite a lot.
| 1,041.68 | 1,054.24 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1046.56
|
We're probably not going to use all of those. I mean, you could do if you wanted, it depends on
| 1,046.56 | 1,059.04 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1054.24
|
what you're trying to build. But I think most of us are probably not going to be trying to build
| 1,054.24 | 1,065.92 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1060.48
|
some model that crosses all these different languages. So what I'm going to do is just
| 1,060.48 | 1,075.44 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1065.9199999999998
|
initialize a list of languages that we would like to train on. So we're going to be feeding all of
| 1,065.92 | 1,083.36 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1075.44
|
this into a sentence transformer class called parallel sentences dataset. And that requires
| 1,075.44 | 1,092.24 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1084.24
|
that we one, separate our pairs using a tab character. And two, keep all those pairs separated
| 1,084.24 | 1,100.48 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1092.24
|
in different GZIP files. So that's why I'm using this particular structure. So data pre-processing
| 1,092.24 | 1,103.92 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1100.48
|
steps here, I'm just running through them quickly because I want to focus more on the actual
| 1,100.48 | 1,110.8 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1103.92
|
sentence transformer training part. So run that, we can, well, it's actually going to take a moment.
| 1,103.92 | 1,118.48 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1110.8000000000002
|
So let me skip forwards. Okay. And then we want to see how many pairs, well, I just want to see,
| 1,110.8 | 1,122.16 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1118.48
|
we don't have to do this. I want to see how many pairs we have for each language.
| 1,118.48 | 1,129.52 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1122.88
|
And you see here we have about 200,000 for each of them. The German one is slightly less.
| 1,122.88 | 1,134.56 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1129.52
|
Okay. And then let's have a look at what those source and translations look like. So here we
| 1,129.52 | 1,144.16 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1134.56
|
have applause and applause. Now I think that's Italian, it seems so. But here we can see, okay,
| 1,134.56 | 1,151.6 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1144.16
|
the end of the talk ends in applause. Obviously the subtitles say applause, or hopefully it ends
| 1,144.16 | 1,157.28 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1151.6
|
in applause. And then we just have the tab character and that separates the source language,
| 1,151.6 | 1,166.72 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1157.28
|
English in this case, from the translated language. Now, okay. Now what we want to do is save that
| 1,157.28 | 1,173.04 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1166.72
|
data. So we sort all that in these dictionaries. Okay. So initialize dictionary here and access
| 1,166.72 | 1,184.08 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1173.04
|
them here. So we have enit, es, ar, fr, and de. And now I'm just going to save them. So run this.
| 1,173.04 | 1,190.48 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1184.08
|
That will save. And what I'll do is, sorry, OSLister. So we can see what is in there.
| 1,184.08 | 1,201.44 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1191.76
|
Where is it? It's data. Just data. Is that right? Okay. And then we have these five files. Okay.
| 1,191.76 | 1,210.32 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1202.24
|
Now let's continue. So now what we want to do is, okay, we have, that's our training data. It's
| 1,202.24 | 1,217.92 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1210.32
|
ready or mostly ready before we feed it into the Sentence Transformer's parallel sentences
| 1,210.32 | 1,224.56 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1217.9199999999998
|
dataset object later on. So, okay, let's leave that for now and move on to the next step, which is
| 1,217.92 | 1,232.08 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1224.56
|
choosing our teacher and student models. So, you know, I already mentioned before, we want our
| 1,224.56 | 1,238.8 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1232.08
|
student model to be capable of multilingual comprehension. So what I mean by that, or
| 1,232.08 | 1,245.52 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1238.8
|
not just what I mean, but one big component of that is, can the Transformer tokenizer deal with
| 1,238.8 | 1,253.36 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1245.52
|
different languages? In some cases they really can't. So let me show you what the Debert tokenizer
| 1,245.52 | 1,259.52 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1253.36
|
does with these four different sentences. So we'll just loop through each one. So four text in
| 1,253.36 | 1,266.64 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1260.08
|
sentences. And what I'm going to do is just print. I'm going to print the output of Debert
| 1,260.08 | 1,277.36 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1266.64
|
tokenizer. And if I tokenize that text, now what does it give me? Okay. So what we have here, okay,
| 1,266.64 | 1,285.68 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1277.3600000000001
|
English, of course, is fine. But the tokenizer or the vocabulary of the tokenizer is, I think,
| 1,277.36 | 1,294.64 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1285.68
|
roughly 30,000 tokens. Okay. And most of those are English based. Okay. Some, like you can see here,
| 1,285.68 | 1,298.72 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1294.64
|
that it has picked up some Chinese characters because it does, you know, other languages do
| 1,294.64 | 1,303.36 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1298.72
|
feed into it a little bit because it's just, you know, all the data is pulled from the internet,
| 1,298.72 | 1,310.08 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1303.92
|
other bits do get in there, but it's mostly English. So that's why we see, okay, we have these
| 1,303.92 | 1,316.88 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1310.08
|
unknown characters. Now, as soon as we have an unknown character in our sentence, the tokenizer,
| 1,310.08 | 1,322.8 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1316.8799999999999
|
or no, sorry, the transformer is read against the wood to understand, you know, what is in that
| 1,316.88 | 1,329.44 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1322.8
|
position? What is that unknown token supposed to represent? In the case of, you know, I think of it
| 1,322.8 | 1,336.64 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1329.4399999999998
|
as it's like, you know, when you're a kid in school and they had those, you know, had like a
| 1,329.44 | 1,341.76 |
All You Need to Know on Multilingual Sentence Vectors (1 Model, 50+ Languages)
|
2021-11-04 13:00:10 UTC
|
https://youtu.be/NNS5pOpjvAQ
|
NNS5pOpjvAQ
|
UCv83tO5cePwHMt1952IVVHw
|
NNS5pOpjvAQ-t1336.64
|
paragraph and you had to fill in the blanks, right? So you had a paragraph and occasionally
| 1,336.64 | 1,346.56 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.