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
|
|---|---|---|---|---|---|---|---|---|
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t402.0
|
Now we also want to import Torch and we're going to use two sentences here.
| 402 | 417.2 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t408.8
|
So both of these are from the Wikipedia page on the American Civil War and these are both consecutive sentences.
| 408.8 | 425.2 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t417.2
|
So going back to what we looked at before we would be hoping that BERT would output a 0 label for both of these.
| 417.2 | 430.4 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t425.2
|
Because sentence B is the next sentence after sentence A.
| 425.2 | 436 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t430.4
|
This one being sentence B this one being sentence A.
| 430.4 | 442 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t436.0
|
So execute that and we now have three different steps that we need to take.
| 436 | 452 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t442.0
|
And that is tokenization, create a classification label so the 0 or the 1 so that we can train the model.
| 442 | 458 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t452.0
|
And then from that we calculate the loss. So the first step there is tokenization.
| 452 | 468.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t458.0
|
We tokenize, it's pretty easy. All we do is inputs, tokenizer and then we pass text and text2.
| 458 | 476.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t468.8
|
And we are using PyTorch here so I want to return a PyTorch tensor.
| 468.8 | 483.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t476.8
|
Make sure that's PT.
| 476.8 | 499.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t483.8
|
We need to also initialize those so tokenizer equals BERT tokenizer from pre-trained.
| 483.8 | 510.4 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t499.8
|
And we'll just use BERT base encased for now. Obviously you can use another BERT model if you want.
| 499.8 | 517.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t510.4
|
And copy that and initialize our model as well.
| 510.4 | 521.4 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t517.8
|
Okay now rerun that.
| 517.8 | 528.6 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t521.4
|
And we'll get this warning that's because we're using these models that are used for training or for fine tuning.
| 521.4 | 534 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t528.6
|
So it's just telling us that we shouldn't really use this for inference. You need to train it first.
| 528.6 | 537 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t534.0
|
And that's fine because that's our intention.
| 534 | 543 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t537.0
|
Now from these inputs we get a few different tensors.
| 537 | 546.4 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t543.0
|
So we have input ids, token type ids and attention mask.
| 543 | 552.2 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t546.4
|
Now for next sentence prediction we do need all of these.
| 546.4 | 554.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t552.2
|
So this is a little bit different to mass language modeling.
| 552.2 | 558.2 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t554.8
|
With mass language modeling we don't actually need token type ids.
| 554.8 | 563.6 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t558.2
|
But for next sentence prediction we do.
| 558.2 | 568 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t563.6
|
So let's have a look at what we have inside these.
| 563.6 | 571.6 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t568.0
|
So input ids is just our tokenized text.
| 568 | 575.6 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t571.6
|
And you see that we pass these two sentences here.
| 571.6 | 582.2 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t575.6
|
And they're actually both within the same sentence or the same tensor here, input ids.
| 575.6 | 586.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t582.2
|
And they're separated by this 102 in the middle which is a separated token.
| 582.2 | 593.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t586.8
|
So before that all these tokens that is our text variable or sentence A.
| 586.8 | 600 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t593.8
|
And then afterwards we have our text 2 variable which is sentence B.
| 593.8 | 603.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t600.0
|
And we can see this mirrored in the token type ids tensor as well.
| 600 | 609.2 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t603.8
|
So all the way along here up to here that's our sentence A.
| 603.8 | 611.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t609.1999999999999
|
So we have zeros for sentence A.
| 609.2 | 617.6 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t611.8
|
And then following that we have ones representing sentence B.
| 611.8 | 624.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t617.5999999999999
|
And then we have our attention mask which is just ones because the attention mask is a one where it's a real token and a zero where we have padding token.
| 617.6 | 629.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t624.8
|
So I don't need to really worry about that tensor at all.
| 624.8 | 635.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t629.8
|
Now the next step here is that we need to create a labels tensor.
| 629.8 | 645.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t635.8
|
So to do that we just write labels and we just need to make sure that when we do this we use a long tensor.
| 635.8 | 660.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t645.8
|
OK. So we use a long tensor and in here we need to pass a list containing a single value which is either our zero but is the next sentence or one for is not the next sentence.
| 645.8 | 664.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t660.8
|
In our case our two sentences are supposed to be together.
| 660.8 | 669.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t664.8
|
So we would pass a zero in here.
| 664.8 | 671.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t669.8
|
And run that.
| 669.8 | 678.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t671.8
|
And if we have a look at what we get from there you see that we get this integer tensor.
| 671.8 | 682.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t678.8
|
So now we're ready to calculate our loss which is really easy.
| 678.8 | 686.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t682.8
|
So we have our model up here which we have already initialized.
| 682.8 | 688.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t686.8
|
So we just take that.
| 686.8 | 697.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t688.8
|
And all we do is pass our inputs from here into our model as keyword arguments.
| 688.8 | 700.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t697.8
|
So that's what these two symbols are for.
| 697.8 | 705.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t700.8
|
And then we also pass labels to the labels parameter.
| 700.8 | 716.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t705.8
|
OK. And that will output a couple of tensors for us so we can execute that and let's have a look at what we have.
| 705.8 | 722.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t716.8
|
So you see that we get these two tensors we have the logits and we also have the loss tensor.
| 716.8 | 728.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t722.8
|
So let's have a look at the logits and we should be able to recognize this from early runway.
| 722.8 | 738.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t728.8
|
We saw those two nodes and we had the two values on for the index zero for is next and index one for is not next.
| 728.8 | 741.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t738.8
|
So let's have a look.
| 738.8 | 743.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t741.8
|
You see here that we get both of those.
| 741.8 | 747.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t743.8
|
This is our activation for is the next sentence.
| 743.8 | 751.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t747.8
|
This is our activation for is not the next sentence.
| 747.8 | 762.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t751.8
|
And if we were to take the argmax of those outputs logits we get zero which means it is the next sentence.
| 751.8 | 771.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t762.8
|
OK. And we also have the loss and this loss tensor that will only be output if we pass our labels here.
| 762.8 | 773.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t771.8
|
Otherwise we just get a logits tensor.
| 771.8 | 779.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t773.8
|
So when we're training obviously we need labels so that we can calculate the loss.
| 773.8 | 796.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t779.8
|
And if we just have a look at that we see it's just a loss value which is very small because the model is predicting a zero and the label that we've provided is also a zero.
| 779.8 | 798.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t796.8
|
So the loss is pretty good there.
| 796.8 | 802.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t798.8
|
So that is how NSP works.
| 798.8 | 812.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t802.8
|
Obviously it's slightly different if you're actually training your model and I am going to cover that in the next video.
| 802.8 | 815.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t812.8
|
So I'll leave a link to that in the description.
| 812.8 | 817.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t815.8
|
But for now that's it for this.
| 815.8 | 836.8 |
Training BERT #3 - Next Sentence Prediction (NSP)
|
2021-05-25 14:56:47 UTC
|
https://youtu.be/1gN1snKBLP0
|
1gN1snKBLP0
|
UCv83tO5cePwHMt1952IVVHw
|
1gN1snKBLP0-t817.8
| 817.8 | 836.8 |
|
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t0.0
|
Today we're going to look at how we can actually publish the component that we've been building for Streamlit.
| 0 | 17 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t7.0
|
So what that means is we can actually pip install the component and then use it in any Streamlit app.
| 7 | 23 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t17.0
|
Just as we would a normal Python framework.
| 17 | 29 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t23.0
|
So we pip install it and then we just import and use it.
| 23 | 35 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t29.0
|
So this is an article I've been putting together that kind of covers this.
| 29 | 39 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t35.0
|
So we're basically just going to be going through this.
| 35 | 43 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t39.0
|
And what you're going to see at the end is we can actually do this.
| 39 | 49 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t43.0
|
So you see here we have this from st card component import card component.
| 43 | 53 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t49.0
|
And then we just do card component title subtitle body link.
| 49 | 58 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t53.0
|
And that will create our card.
| 53 | 64 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t58.0
|
So there's not that much to go through. It's pretty straightforward I think.
| 58 | 69 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t64.0
|
So let's jump into it.
| 64 | 76 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t69.0
|
So a little bit of background on how pip is actually working here.
| 69 | 81 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t76.0
|
When you pip install something you're actually installing it from this here.
| 76 | 84 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t81.0
|
The Python package index.
| 81 | 88 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t84.0
|
So put all pi b here. I think that's how you pronounce it.
| 84 | 94 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t88.0
|
And I can go in here and I can search for pandas.
| 88 | 100 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t94.0
|
And it's going to show us pandas or it's going to show us a lot of different pandas.
| 94 | 102 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t100.0
|
I think it's this one.
| 100 | 106 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t102.0
|
OK. Yeah. Pip install pandas at the top there.
| 102 | 114 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t106.0
|
Now we can also find the st card component I've already built beforehand.
| 106 | 119 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t114.0
|
So if we open this. I think it's this one. Yeah.
| 114 | 124 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t119.0
|
It's like version 0.10 at the moment.
| 119 | 129 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t124.0
|
So this is the current component. Look there's me.
| 124 | 132 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t129.0
|
And you can go ahead and install that.
| 129 | 137 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t132.0
|
It's slightly different to the component that we have been putting together.
| 132 | 141 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t137.0
|
Not as generic.
| 137 | 147 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t141.0
|
I built it for a particular use case which you'll probably see pretty soon.
| 141 | 152 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t147.0
|
So what we're going to do is create another st card component 2 or something.
| 147 | 158 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t152.0
|
I don't know. You have to give it a unique name so we can't use the same name again.
| 152 | 162 |
Streamlit for ML #5.3 - Publishing Components to Pip
|
2022-02-28 17:00:29 UTC
|
https://youtu.be/lZ2EaPUnV7k
|
lZ2EaPUnV7k
|
UCv83tO5cePwHMt1952IVVHw
|
lZ2EaPUnV7k-t158.0
|
So we'll go ahead and start with that.
| 158 | 166 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.