Add VAE model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: llamagen
|
| 3 |
+
tags:
|
| 4 |
+
- image-generation
|
| 5 |
+
- image-tokenizer
|
| 6 |
+
- imagenet
|
| 7 |
+
- vq-vae
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# VAE checkpoints for ImageNet LlamaGen GPT-XL
|
| 11 |
+
|
| 12 |
+
This repository contains the two VQ-16 class-conditional ImageNet tokenizers used by
|
| 13 |
+
the paired LlamaGen GPT-XL experiment:
|
| 14 |
+
|
| 15 |
+
- `base_vae_vq_ds16_c2i.pt`: official FoundationVision/LlamaGen VQ-16 c2i checkpoint.
|
| 16 |
+
- `reward_vae_vq_ds16_c2i_seed3101_step1000.pt`: reward-tuned VQ-16 c2i checkpoint,
|
| 17 |
+
exported from the seed 3101, step 1000 tokenizer run.
|
| 18 |
+
|
| 19 |
+
Both checkpoints use a 16,384-entry codebook and produce 24 x 24 = 576 tokens for a
|
| 20 |
+
384px ImageNet input. They are intended for the ImageNet c2i training pipeline in
|
| 21 |
+
https://github.com/hzeroyuke/vae-posttraining.
|
| 22 |
+
|
| 23 |
+
## SHA256
|
| 24 |
+
|
| 25 |
+
| File | SHA256 |
|
| 26 |
+
| --- | --- |
|
| 27 |
+
| `base_vae_vq_ds16_c2i.pt` | `109aa8afb2cf3761eec23cdc8644154cb498f5ab7eef2a35264d25e5e0499f7d` |
|
| 28 |
+
| `reward_vae_vq_ds16_c2i_seed3101_step1000.pt` | `55c06b0ab914f9f40b24eba07033f7c452eecb6233562baf27e69121282434ea` |
|
| 29 |
+
|
| 30 |
+
The checkpoints are distributed for research use. Follow the licenses and terms of
|
| 31 |
+
LlamaGen, ImageNet, PyTorch, and any downstream evaluation dependencies.
|