Commit
•
1461df2
1
Parent(s):
62785a4
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# Imagenet.int8: Entire Imagenet dataset in 5GB.
|
| 2 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
|
| 4 |
Find 138 GB of imagenet dataset too bulky? Did you know entire imagenet actually just fits inside the ram of apple watch?
|
| 5 |
|
|
@@ -20,13 +30,6 @@ These include:
|
|
| 20 |
|
| 21 |
... but so little material online on the actual preprocessed dataset. I'm here to fix that. One thing I noticed was that latent doesn't have to be full precision! Indeed, they can be as small as int-8, and it won't hurt! Here are some of the examples:
|
| 22 |
|
| 23 |
-
<p align="center">
|
| 24 |
-
<img src="contents/vae.png" alt="small" width="800">
|
| 25 |
-
</p>
|
| 26 |
-
|
| 27 |
-
| original, reconstructed from float16, reconstructed from uint8*
|
| 28 |
-
|
| 29 |
-
|
| 30 |
So clearly, it doesn't make sense to download entire Imagenet and do VAE everytime. Just download this, `to('cuda')` the entire dataset just to flex, and call it a day.😌
|
| 31 |
|
| 32 |
(BTW If you think you'll need higher precision, you can always further fine-tune your model on higher precision. But I doubt that.)
|
|
|
|
| 1 |
+
---
|
| 2 |
+
size_categories:
|
| 3 |
+
- 1M<n<10M
|
| 4 |
+
---
|
| 5 |
# Imagenet.int8: Entire Imagenet dataset in 5GB.
|
| 6 |
|
| 7 |
+
<p align="center">
|
| 8 |
+
<img src="contents/vae.png" alt="small" width="800">
|
| 9 |
+
</p>
|
| 10 |
+
|
| 11 |
+
| original, reconstructed from float16, reconstructed from uint8*
|
| 12 |
+
|
| 13 |
|
| 14 |
Find 138 GB of imagenet dataset too bulky? Did you know entire imagenet actually just fits inside the ram of apple watch?
|
| 15 |
|
|
|
|
| 30 |
|
| 31 |
... but so little material online on the actual preprocessed dataset. I'm here to fix that. One thing I noticed was that latent doesn't have to be full precision! Indeed, they can be as small as int-8, and it won't hurt! Here are some of the examples:
|
| 32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
So clearly, it doesn't make sense to download entire Imagenet and do VAE everytime. Just download this, `to('cuda')` the entire dataset just to flex, and call it a day.😌
|
| 34 |
|
| 35 |
(BTW If you think you'll need higher precision, you can always further fine-tune your model on higher precision. But I doubt that.)
|