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
|
|---|---|---|---|---|---|---|---|---|
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1310.6
|
Okay, let's save this and
| 1,310.6 | 1,317.96 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1312.6
|
And try again. Okay, so i'm going to copy this over into home face come here
| 1,312.6 | 1,320.68 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1318.6799999999998
|
Uh not here here
| 1,318.68 | 1,322.76 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1320.9199999999998
|
edit
| 1,320.92 | 1,327.8 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1322.76
|
And come here select all paste and I am going to commit those changes
| 1,322.76 | 1,333.32 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1329.08
|
Now let's have a look at what happens if we load the data set so
| 1,329.08 | 1,336.12 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1334.12
|
Come back over here test data set
| 1,334.12 | 1,338.44 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1336.4399999999998
|
Uh, let's run this
| 1,336.44 | 1,340.52 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1338.76
|
Let's see what happens
| 1,338.76 | 1,345 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1340.52
|
Okay, it loaded well it loaded correctly. That's a good sign come down here
| 1,340.52 | 1,351.72 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1345.72
|
And now we can see that these are no longer strings, but they're actually flowing point numbers. Okay, so
| 1,345.72 | 1,354.76 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1353.32
|
That is
| 1,353.32 | 1,359.96 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1354.76
|
That's everything there are maybe a few aesthetic things to change here. So the
| 1,354.76 | 1,362.6 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1360.68
|
Like the citation
| 1,360.68 | 1,367.24 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1362.6
|
We'll change that up here. I can change this as well, but we're not going to go through that in this
| 1,362.6 | 1,373.74 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1367.24
|
Uh in this video, but I think you want to watch me change citations. So yeah, that's everything
| 1,367.24 | 1,376.2 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1374.76
|
For this video in the next video
| 1,374.76 | 1,380.52 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1376.2
|
What we're going to do is take a look at taking this a a little bit further
| 1,376.2 | 1,385.64 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1381.0
|
And adding more advanced data types like images into our data sets
| 1,381 | 1,387.48 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1386.28
|
so
| 1,386.28 | 1,389.96 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1387.48
|
Until then I hope this has been useful
| 1,387.48 | 1,396.86 |
Hugging Face Datasets #2 - Dataset Builder Scripts
|
2022-09-23 14:45:22 UTC
|
https://youtu.be/ODdKC30dT8c
|
ODdKC30dT8c
|
UCv83tO5cePwHMt1952IVVHw
|
ODdKC30dT8c-t1389.96
| 1,389.96 | 1,396.86 |
|
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t0.0
|
In this video we're going to have a quick introduction to OpenAI's clip and how we can use it to almost move between the modalities of both language and images.
| 0 | 19.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t15.0
|
Now before we dive in let's just quickly understand what clip is.
| 15 | 23 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t19.5
|
So it consists of two big models.
| 19.5 | 29.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t23.0
|
In this implementation we're going to be using a vision transformer that will embed images.
| 23 | 35 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t29.5
|
And we're going to use a normal text transformer that will embed text.
| 29.5 | 51 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t35.0
|
During pre-training OpenAI trained the model on pairs of images and text and it trained them to both output embedding vectors that are as close as possible to each other.
| 35 | 68 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t51.0
|
So the text transformer was trained to output a single embedding 512 dimensional embedding that was as close as possible to the vision transformer's image embedding for the image text pair.
| 51 | 79.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t68.0
|
So what that means is that clip is able to take both images and text and embed them both into a similar vector space.
| 68 | 81.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t79.5
|
And with that we can do a lot of things.
| 79.5 | 84.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t81.5
|
You can do image and text classification.
| 81.5 | 88.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t84.5
|
You can do image and text search and a huge number of things.
| 84.5 | 93.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t88.5
|
Anything to do with images and text there's a good chance we can do it with clip.
| 88.5 | 98 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t93.5
|
So let's have a look at how we actually use clip.
| 93.5 | 102 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t98.0
|
OpenAI released a GitHub repository OpenAI clip here.
| 98 | 107.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t102.0
|
This contains clip but we're not going to use this implementation.
| 102 | 111 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t107.5
|
We're actually going to use this implementation of clip.
| 107.5 | 113 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t111.0
|
So this is on Hugging Face.
| 111 | 118 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t113.0
|
So we're going to be using Hugging Face transformers and this is still from OpenAI.
| 113 | 119 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t118.0
|
It's still clip.
| 118 | 134 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t119.0
|
It's just an easy to use implementation of it through the Hugging Face transformers library which is a more standard library for actually doing anything with NLP and also now computer vision and some other things as well.
| 119 | 137.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t134.0
|
So to get started I'd recommend you install these libraries.
| 134 | 145.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t137.5
|
To install Torch you should probably go through the PyTorch.org instructions rather than following this here.
| 137.5 | 157.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t145.5
|
So go to PyTorch.org and just install PyTorch using the specific install command they use for your platform or your iOS from here.
| 145.5 | 160.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t157.5
|
And then pip install transformers and datasets.
| 157.5 | 166 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t160.5
|
You can still just use this command I'd recommend installing PyTorch from here instead.
| 160.5 | 171 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t166.0
|
Now after that we're going to need our dataset.
| 166 | 175.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t171.0
|
So this is just a very simple dataset.
| 171 | 183.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t175.5
|
It contains I think just under 10,000 images and we only care about the images here.
| 175.5 | 191.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t183.5
|
So if we have a look we have ImageNet we'll go the first item and we'll just have a look at image.
| 183.5 | 196.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t191.5
|
And we have this Sony radio and we have other things as well.
| 191.5 | 200 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t196.5
|
So if we go ImageNet.
| 196.5 | 204 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t200.0
|
It's 6494.
| 200 | 206.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t204.0
|
There's another image here of a dog.
| 204 | 214.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t206.5
|
OK just to point out that we have a lot of images in here in the dataset that cover a range of things.
| 206.5 | 220.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t214.5
|
There's not a huge number of different categories here but they have dogs they have radios and a few other things.
| 214.5 | 224.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t220.5
|
Now I'm just going to go ahead and initialize everything.
| 220.5 | 227.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t224.5
|
So there's a few things here.
| 224.5 | 231.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t227.5
|
From transformers we're importing the clip tokenizer.
| 227.5 | 240.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t231.5
|
So the tokenizer is what's going to handle the pre-processing of our text into token ID tensors and other tensors.
| 231.5 | 246.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t240.5
|
We have the clip processor that's like the tokenizer but for images.
| 240.5 | 259.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t246.5
|
So this is actually just going to resize our images into the size that clip expects and also modify the pixel values as well.
| 246.5 | 263.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t259.5
|
And then we have clip model. Clip model is clip itself.
| 259.5 | 274.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t263.5
|
OK so if you have CUDA or MPS if you're on M1 Mac you just set that with this.
| 263.5 | 279.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t274.5
|
OK and then we're ready to actually initialize all of this.
| 274.5 | 285.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t279.5
|
So the model ID is going to be what we saw before.
| 279.5 | 294.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t285.5
|
So you come over here we have the tokenizer clip VIT base patch 32 copy that.
| 285.5 | 297.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t294.5
|
And here we go. OK.
| 294.5 | 301.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t297.5
|
And now we just need to look I'm being told what to do already.
| 297.5 | 305.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t301.5
|
OK so model clip model from pre-trained model ID.
| 301.5 | 308.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t305.5
|
I'm going to I don't normally set device like that.
| 305.5 | 313.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t308.5
|
I don't know if you can. I am going to do it like this.
| 308.5 | 319.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t313.5
|
OK and tokenizer. OK good job. And processor. Cool.
| 313.5 | 325.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t319.5
|
Almost there. It's from pre-trained.
| 319.5 | 333.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t325.5
|
OK and you got a little bit confused. So model ID.
| 325.5 | 337.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t333.5
|
OK that looks good. Let's run that.
| 333.5 | 345.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t337.5
|
OK cool. So now what we're going to do is take a look how we actually create the text embeddings through clip.
| 337.5 | 350.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t345.5
|
So we start with a prompt. I'm going to go with a dog in the snow.
| 345.5 | 356.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t350.5
|
There's not many pictures of dogs in the snow in the dataset but there are some.
| 350.5 | 359.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t356.5
|
And what we need to do is is tokenize the prompt.
| 356.5 | 363.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t359.5
|
Yeah that's true. OK. I'm not going to do it like that.
| 359.5 | 371.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t363.5
|
We're going to go with tokenize prompt and the we need to return tensors using Pytorch.
| 363.5 | 375.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t371.5
|
So we're using we're going to be using Pytorch behind the scenes here.
| 371.5 | 383.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t375.5
|
So make sure we do that. And let's just have a look at what is actually in inputs.
| 375.5 | 391.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t383.5
|
OK so we get the this input ID is tensor so you'll you'll recognize this if you if you use a face transformers before.
| 383.5 | 397.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t391.5
|
These are just the ID token IDs that represent the words from this.
| 391.5 | 402.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t397.5
|
OK. And these this is the attention mask.
| 397.5 | 410.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t402.5
|
Now for us it is going to all be ones but if we had padding in here anything beyond the length of our prompt would become a zero.
| 402.5 | 415.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t410.5
|
Telling the model to not pay attention to to that part of the prompt.
| 410.5 | 429.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t415.5
|
And from there we can process this through clips so we do model get text features I think.
| 415.5 | 437.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t429.5
|
And we pass in those inputs. OK. And let's have a look at the shape of that.
| 429.5 | 442.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t437.5
|
OK so we have a five hundred and twelve dimensional vector.
| 437.5 | 447.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t442.5
|
OK. So that's the text embedding side of things.
| 442.5 | 451.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t447.5
|
Now we need to go ahead and do the image embedding side of things.
| 447.5 | 456.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t451.5
|
OK. So we're going to resize the image first with the processor.
| 451.5 | 462.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t456.5
|
We're not adding any text in here so you can also process text through this processor.
| 456.5 | 466.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t462.5
|
I'm just keeping it separate because it makes more sense to me.
| 462.5 | 472.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t466.5
|
The image should be images actually.
| 466.5 | 476.5 |
Fast intro to multi-modal ML with OpenAI's CLIP
|
2022-08-11 13:03:08 UTC
|
https://youtu.be/989aKUVBfbk
|
989aKUVBfbk
|
UCv83tO5cePwHMt1952IVVHw
|
989aKUVBfbk-t472.5
|
Again we want to return tensors using PyTorch.
| 472.5 | 483.5 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.