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 #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1214.08
|
And labels.
| 1,214.08 | 1,222.32 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1214.8
|
Okay. So that's quite a lot going into our model. And now what we want to do is extract the loss
| 1,214.8 | 1,230.32 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1222.32
|
from that. Then we calculate loss for every parameter in our model. And then using that,
| 1,222.32 | 1,237.84 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1230.32
|
we can update our gradients using our optimizer. And then what we want to do is print the relevant
| 1,230.32 | 1,246.96 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1237.84
|
info to our progress bar that we set up using TQDM and loop. So loop set description.
| 1,237.84 | 1,256.32 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1250.24
|
And here I was going to put the epoch info. So the epoch we're currently on.
| 1,250.24 | 1,260.64 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1258.08
|
And then I also want to set the post fix.
| 1,258.08 | 1,268.64 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1260.64
|
Which will contain the loss information. So loss.item. Okay. We can run that.
| 1,260.64 | 1,276.4 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1270.24
|
And you see that our model is now training. So we're now training a model using both
| 1,270.24 | 1,283.04 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1276.4
|
assignment modeling and net sentence prediction. And we haven't needed to take any structured data
| 1,276.4 | 1,288.48 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1283.0400000000002
|
to set up the model. So we can just run that. And we can see that our model is now training.
| 1,283.04 | 1,294.64 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1288.48
|
We haven't needed to take any structured data. We've just taken a book and pulled all data and
| 1,288.48 | 1,299.2 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1294.64
|
formatted it in the correct way for us to actually train a better model, which I think is really
| 1,294.64 | 1,318.72 |
Training BERT #5 - Training With BertForPretraining
|
2021-06-15 15:00:19 UTC
|
https://youtu.be/IC9FaVPKlYc
|
IC9FaVPKlYc
|
UCv83tO5cePwHMt1952IVVHw
|
IC9FaVPKlYc-t1299.2
| 1,299.2 | 1,318.72 |
|
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t0.0
|
Okay, so in the previous video what we did was set up our Elasticsearch document store to contain all of our
| 0 | 11.64 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t9.64
|
paragraphs from meditations
| 9.64 | 15.2 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t12.3
|
so we did that in this script here and
| 12.3 | 23.06 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t16.2
|
All together we only have not that much data, 508 paragraphs or documents within our document store
| 16.2 | 25.2 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t23.76
|
so
| 23.76 | 30 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t25.2
|
What we now want to do is set up the next part of our
| 25.2 | 34.44 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t31.04
|
Retriever reader stack, which is the retriever and
| 31.04 | 37.2 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t35.2
|
What the retriever will do is
| 35.2 | 43.48 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t37.8
|
given a query it will communicate with our Elasticsearch document store and
| 37.8 | 45.8 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t44.28
|
return a
| 44.28 | 52.6 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t45.8
|
Certain number of contexts which are the paragraphs in our case that it thinks are most relevant to our query
| 45.8 | 54.6 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t52.6
|
So
| 52.6 | 58.56 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t55.760000000000005
|
That's what we are going to be doing here and
| 55.76 | 67.14 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t59.120000000000005
|
The first thing that we need to do is initialize our document store again, so I'm just going to copy these
| 59.12 | 70.16 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t68.48
|
and
| 68.48 | 72.16 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t70.16
|
paste them here and
| 70.16 | 81.04 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t74.92
|
This would just initialize it from what we've already built so it's using the same index that already exists
| 74.92 | 82.92 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t81.04
|
so
| 81.04 | 89.36 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t82.92
|
Just initialize that and once we have our document store. Okay, cool. We have that now
| 82.92 | 96.88 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t90.58000000000001
|
Now what we want to do is set up our DPR, which is a dense passage retriever, which
| 90.58 | 99.92 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t97.92
|
essentially uses
| 97.92 | 102.42 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t100.42
|
dense vectors and
| 100.42 | 105.6 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t102.56
|
a type of efficient similarity search
| 102.56 | 109.72 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t106.78
|
to embed these indexes as
| 106.78 | 114.24 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t109.72
|
dense vectors and then once it comes to actually searching
| 109.72 | 117.44 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t114.8
|
And finding the most similar or the most relevant
| 114.8 | 124.78 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t118.68
|
Documents later on it will use those dense vectors and find the most similar ones
| 118.68 | 130 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t125.8
|
So I'll explain that a little bit better in a moment
| 125.8 | 135.46 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t132.0
|
So first what we want to do is actually initialize that
| 132 | 143.3 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t135.46
|
So we do from Haystack dense retriever
| 135.46 | 148.06 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t145.14000000000001
|
Import dense passage retriever
| 145.14 | 158.54 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t155.14000000000001
|
Sorry, it's the other way around here so retriever dense
| 155.14 | 163.54 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t162.34
|
And
| 162.34 | 166.18 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t163.54
|
then we'll put into a
| 163.54 | 175.38 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t169.29999999999998
|
Variable called retriever which uses the dense passage retriever from up here
| 169.3 | 184.42 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t176.57999999999998
|
And in here we need to pass a few parameters. So the first thing is the document store. So the document store is
| 176.58 | 187.78 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t185.38
|
just what we've already initialized up so and
| 185.38 | 195.98 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t187.78
|
Then we need to initialize two different models so it's the query embedding model
| 187.78 | 204.06 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t200.86
|
And the passage embedding model
| 200.86 | 212.06 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t207.66
|
Now behind the scenes Haystack is using the Hugging Face Transformers library
| 207.66 | 214.54 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t212.06
|
So what we'll do is we'll head over to the
| 212.06 | 219.66 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t214.54
|
Models over there and see which embedding models we can use for DPR
| 214.54 | 237.42 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t229.34
|
Okay, so here let's just search for DPR and you'll find we have all of these models from Facebook AI
| 229.34 | 240.86 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t238.85999999999999
|
Now with DPR
| 238.86 | 246.22 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t240.86
|
the reason that it's so useful for question answering is that we have
| 240.86 | 251.5 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t247.26000000000002
|
What are two different models that encode
| 247.26 | 254.46 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t252.06
|
the text that we pass into it so we have
| 252.06 | 258.06 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t255.34
|
this sort of setup during training and
| 255.34 | 261.26 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t259.26
|
What we see down here
| 259.26 | 265.34 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t263.34000000000003
|
Are these two models we have this
| 263.34 | 267.98 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t265.98
|
EP, BERT, and EMP
| 265.98 | 271.26 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t267.98
|
models we have this EP, BERT encoder
| 267.98 | 279.98 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t271.98
|
And we also have this EQ, BERT encoder. Now the EP, BERT encoder encodes the passages or the context
| 271.98 | 283.74 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t280.3
|
So essentially the paragraphs that we have
| 280.3 | 286.54 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t284.46000000000004
|
fed into our elastic search model
| 284.46 | 289.26 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t287.26
|
This is what we'll be
| 287.26 | 293.02 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t289.74
|
encoding them into these vectors here
| 289.74 | 299.1 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t293.02
|
Now this is during training this whole graph. So all we will actually see
| 293.02 | 303.5 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t299.74
|
when we're encoding these vectors is we will see the
| 299.74 | 307.98 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t305.97999999999996
|
EP encoder
| 305.98 | 315.98 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t312.78
|
And this will create the EP vectors
| 312.78 | 320.06 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t315.98
|
And all we're going to do is feed in all of the documents
| 315.98 | 324.62 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t321.26
|
from elastic search into this
| 321.26 | 330.3 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t326.86
|
Now once all of these have been encoded
| 326.86 | 335.66 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t331.34000000000003
|
We then have a new set of dense vectors
| 331.34 | 343.26 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t339.98
|
And then we'll have a new set of dense vectors
| 339.98 | 345.26 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t343.26
|
And all of those
| 343.26 | 351.42 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t346.3
|
Will be fed back into our document store so back into elastic
| 346.3 | 358.7 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t354.3
|
Now when it comes to performing similarity search later on
| 354.3 | 365.42 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t359.58
|
We're going to ask a question and that question will be processed by the EQ encoder
| 359.58 | 369.82 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t367.65999999999997
|
So here we have our
| 367.66 | 372.46 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t369.82
|
EQ encoder
| 369.82 | 377.5 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t373.98
|
And we have our question so that will go into here
| 373.98 | 383.58 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t379.9
|
And that will encode our question
| 379.9 | 389.58 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t384.38
|
And then send it over to elastic and say okay what are the most similar vectors to
| 384.38 | 392.86 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t390.3
|
this vector that we created from a question
| 390.3 | 397.18 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t394.14
|
And the reason that we're asking this question is because
| 394.14 | 400.86 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t397.18
|
We're going to be using a new set of vectors to train our question
| 397.18 | 405.18 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t402.3
|
And the reason that DPR is so good is
| 402.3 | 408.14 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t406.14
|
That if you look at the training down here
| 406.14 | 417.26 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t408.86
|
We are creating these EP vectors and these EQ vectors that are matching so where we have a matching question to a matching context
| 408.86 | 420.14 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t418.54
|
We are training them
| 418.54 | 423.1 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t420.14
|
To maximize the dot product
| 420.14 | 428.22 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t423.1
|
And the alignment between those two vectors so what happens is
| 423.1 | 432.46 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t429.1
|
That a relevant passage and a relevant question
| 429.1 | 437.18 |
Q&A Document Retrieval With DPR
|
2021-04-15 15:00:10 UTC
|
https://youtu.be/DBsxUSUhfRg
|
DBsxUSUhfRg
|
UCv83tO5cePwHMt1952IVVHw
|
DBsxUSUhfRg-t433.82000000000005
|
Will come out to have a very similar vector
| 433.82 | 441.82 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.