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
|
|---|---|---|---|---|---|---|---|---|
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t606.08
|
11.
| 606.08 | 613.04 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t610.64
|
Okay, this is actually quite useful.
| 610.64 | 619.68 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t613.0400000000001
|
So I've got this error that's popped up because I am currently in another project,
| 613.04 | 620.88 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t619.6800000000001
|
which no longer exists.
| 619.68 | 622.48 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t620.88
|
So I'm getting this error.
| 620.88 | 627.68 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t622.48
|
So I need to make sure that I'm trying to deploy to the right project.
| 622.48 | 630.16 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t627.6800000000001
|
So I just write gcloud app deploy.
| 627.68 | 633.92 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t632.0
|
I can specify the app.yaml.
| 632 | 637.12 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t633.92
|
And I just write project equals.
| 633.92 | 640.16 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t637.12
|
And then in here, I need the project name from before.
| 637.12 | 642.16 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t640.16
|
So I'm just going to go and copy that again.
| 640.16 | 645.76 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t644.8
|
So come to here.
| 644.8 | 652.96 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t651.1999999999999
|
So it's this abstract ring.
| 651.2 | 659.12 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t652.96
|
Abstract ring.
| 652.96 | 665.36 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t662.32
|
And I just write in here abstract ring.
| 662.32 | 666.64 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t666.4000000000001
|
Okay.
| 666.4 | 668.72 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t667.2800000000001
|
Now run that.
| 667.28 | 671.6 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t668.72
|
That should come up with this here.
| 668.72 | 675.52 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t672.24
|
So we have come down here.
| 672.24 | 678.88 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t675.52
|
We have the script and then the target project.
| 675.52 | 679.6 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t678.88
|
So it's correct.
| 678.88 | 682.96 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t679.6
|
We're now going to the correct place.
| 679.6 | 687.36 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t682.96
|
And this URL here is where we will be deploying the API to.
| 682.96 | 689.84 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t688.16
|
So I'm going to click continue.
| 688.16 | 690.16 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t689.84
|
Yes.
| 689.84 | 696.24 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t691.84
|
And this can take a little bit of time for it to run.
| 691.84 | 699.12 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t697.36
|
So just wait a moment.
| 697.36 | 705.36 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t700.48
|
One thing that's actually pretty useful is when you try,
| 700.48 | 709.92 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t705.36
|
when you first deploy, gcloud will create this gcloud ignore file.
| 705.36 | 711.12 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t709.92
|
It's like a git ignore.
| 709.92 | 718.8 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t712.4
|
And in here, it will just ignore the typical things that you do actually want to ignore.
| 712.4 | 722.16 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t719.36
|
So you have PyCache here, which has been created up here.
| 719.36 | 728.4 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t722.16
|
So we don't need to exclude that from the upload by default, which is useful, I think.
| 722.16 | 732.96 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t729.76
|
So see how that is going.
| 729.76 | 734.08 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t732.96
|
Still uploading.
| 732.96 | 735.6 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t734.08
|
Still uploading.
| 734.08 | 737.52 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t736.72
|
Okay.
| 736.72 | 740.08 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t737.5200000000001
|
So good to go.
| 737.52 | 741.92 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t741.0400000000001
|
I'm going to copy this.
| 741.04 | 745.36 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t744.4000000000001
|
Open in the browser.
| 744.4 | 752.56 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t747.5200000000001
|
And we should see, so what we saw before with this, we should end up seeing the same.
| 747.52 | 756 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t752.5600000000001
|
So we'll go first to that address.
| 752.56 | 758.64 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t758.4000000000001
|
Okay.
| 758.4 | 760.72 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t758.6400000000001
|
We get that not found that we saw before.
| 758.64 | 763.04 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t760.72
|
So expected.
| 760.72 | 764.4 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t763.04
|
And we'll just go to users.
| 763.04 | 767.6 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t767.04
|
There we go.
| 767.04 | 769.44 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t769.04
|
Okay.
| 769.04 | 775.6 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t769.4399999999999
|
So we see that our API is now deployed to Google Cloud.
| 769.44 | 777.44 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t775.5999999999999
|
I can also see that over here.
| 775.6 | 785.2 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t777.4399999999999
|
So if I refresh, we should see the request pop up over here.
| 777.44 | 788.16 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t786.64
|
Or they will do eventually.
| 786.64 | 792.88 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t788.16
|
So this will give you a summary of how many requests are made to your API.
| 788.16 | 794.56 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t792.88
|
That's it for this video.
| 792.88 | 798.24 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t795.4399999999999
|
I hope it's been useful.
| 795.44 | 801.44 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t799.1999999999999
|
And I will see you in the next one.
| 799.2 | 819.44 |
API Series #3 - How to Deploy Flask APIs to the Cloud (GCP)
|
2021-11-02 14:30:00 UTC
|
https://youtu.be/3fsIcMgUOY8
|
3fsIcMgUOY8
|
UCv83tO5cePwHMt1952IVVHw
|
3fsIcMgUOY8-t801.44
| 801.44 | 819.44 |
|
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t0.0
|
Hi, welcome to the video. Here we're going to have a look at using next sentence prediction or NSP
| 0 | 14.4 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t6.8
|
for fine tuning our BERT models. Now a few of the previous videos we covered mass language modeling
| 6.8 | 21.68 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t14.4
|
and how we use mass language modeling to fine tune our models. NSP is like the other half of fine
| 14.4 | 29.92 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t21.68
|
tuning for BERT. So both of those techniques during the actual training of BERT, so when Google train
| 21.68 | 38.96 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t29.92
|
BERT initially, they use both of these methods and whereas MLM is identifying or almost training on
| 29.92 | 46.88 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t38.96
|
the relationships between words, next sentence prediction is training on more long-term relationships
| 38.96 | 56 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t46.88
|
between sentences rather than words. And in the original BERT paper it was found that without
| 46.88 | 62.72 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t56.0
|
NSP because they tried training BERT without NSP as well, BERT performed worse on every single metric.
| 56 | 69.2 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t62.72
|
So it is pretty important and obviously if we take this approach, we take mass language modeling and
| 62.72 | 76.32 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t69.2
|
NSP and apply both those to training our models, fine tuning our models, we're going to get better
| 69.2 | 85.76 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t76.32
|
results and if we just use MLM. So what is NSP? NSP consists of giving BERT two sentences, sentence
| 76.32 | 93.04 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t85.76
|
A and sentence B and saying paper does sentence B come after sentence A and then BERT will say okay
| 85.76 | 101.44 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t94.4
|
sentence B is the next sentence after sentence A or it is not the next sentence after the sentence
| 94.4 | 110.32 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t101.44
|
A. So if we took these three sentences that are on the screen, we have one, two and three, right?
| 101.44 | 118.64 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t110.32
|
One and two, if you ask BERT does sentence two come after sentence one, then we'd kind of want BERT to
| 110.32 | 125.68 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t118.63999999999999
|
say no, right? Because clearly they're talking about completely different topics and the type
| 118.64 | 130.4 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t125.67999999999999
|
of language and everything in there just doesn't really match up. But then if we have a look at
| 125.68 | 139.04 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t130.4
|
sentence three and sentence one, they do match up. So sentence three is quite possibly a
| 130.4 | 148.08 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t139.04
|
the follow-on sentence after sentence one. So in that case we would expect BERT to say this is the
| 139.04 | 158.88 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t148.07999999999998
|
next sentence. So let's have a look at how NSP looks within BERT itself. So here we have the core
| 148.08 | 171.2 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t158.88
|
BERT model and during fine-tuning or pre-training we add this other head on top of BERT. So this is the BERT for pre-training head.
| 158.88 | 177.44 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t172.0
|
And the BERT for pre-training head contains two different heads inside it.
| 172 | 184.64 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t180.48
|
And that is our NSP head and our mass language modeling head.
| 180.48 | 193.68 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t184.64
|
Now we just want to focus on the NSP head for now. And as well we don't need to fine-tune or
| 184.64 | 198.88 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t193.67999999999998
|
train our models with both of these heads. We can actually do it one by one. We could use it mass
| 193.68 | 206.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t198.88
|
language modeling only or we could use NSP only. But the full approach to pre-training BERT is using both.
| 198.88 | 213.2 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t206.8
|
So if we have a look inside our NSP head we'll find that we have a feed forward neural network
| 206.8 | 223.92 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t213.92000000000002
|
and that will output two different values. Now these two values are our is not the next sequence
| 213.92 | 236.56 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t223.92
|
there and our is the next sequence which is there. Okay so value zero is the next sentence value one is not the next sentence.
| 223.92 | 245.92 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t238.56
|
Now we have the final outputs from our final encoder in BERT at the bottom here.
| 238.56 | 256 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t245.92
|
And we don't actually use all of these activations. We only use the CLS token activation which is over at the left here.
| 245.92 | 265.68 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t256.0
|
So this here is our CLS token. Okay and when I say this is our CLS token I mean more that this is not our CLS token.
| 256 | 285.04 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t265.68
|
The CLS token is down here so we input the CLS token and this output is the subsequent output after being processed by 12 or so encoders within BERT itself.
| 265.68 | 299.76 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t285.04
|
So this is the output representation of that CLS token. Now the activations from that get fed into our feed forward neural network.
| 285.04 | 310 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t299.76000000000005
|
And the dimensionality that we have here is 768 for that single token. This is in the
| 299.76 | 321.36 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t310.0
|
BERT base model by the way. And that gets translated into our dimensionality here which is just the two outputs.
| 310 | 332.24 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t321.36
|
So that's essentially how NSP works. Once we have our two outputs here we just take the arg max of both of those.
| 321.36 | 343.6 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t332.24
|
So we take both over here and we just take an arg max function of that and that will output us either 0 or 1.
| 332.24 | 353.6 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t343.6
|
Where 0 is the is next class and 1 is the not next class.
| 343.6 | 368 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t353.6
|
And that's how NSP works. So let's dive into the code and see how all this works in Python.
| 353.6 | 376 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t368.0
|
Okay so we're going to be using HuggingFace's transformers and PyTorch. So we'll import both of those.
| 368 | 394 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t376.0
|
And from transformers we just need the BERT tokenizer class and the BERT for next sentence prediction class.
| 376 | 402 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t394.0
|
And BERT for next sentence prediction.
| 394 | 408.8 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.