add vq-f8
Browse files- vq-f8/config.yaml +48 -0
- vq-f8/model.ckpt +3 -0
vq-f8/config.yaml
ADDED
|
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
model:
|
| 2 |
+
base_learning_rate: 4.5e-06
|
| 3 |
+
target: ldm.models.autoencoder.VQModel
|
| 4 |
+
params:
|
| 5 |
+
embed_dim: 4
|
| 6 |
+
n_embed: 16384
|
| 7 |
+
monitor: val/rec_loss
|
| 8 |
+
ddconfig:
|
| 9 |
+
double_z: false
|
| 10 |
+
z_channels: 4
|
| 11 |
+
resolution: 256
|
| 12 |
+
in_channels: 3
|
| 13 |
+
out_ch: 3
|
| 14 |
+
ch: 128
|
| 15 |
+
ch_mult:
|
| 16 |
+
- 1
|
| 17 |
+
- 2
|
| 18 |
+
- 2
|
| 19 |
+
- 4
|
| 20 |
+
num_res_blocks: 2
|
| 21 |
+
attn_resolutions:
|
| 22 |
+
- 32
|
| 23 |
+
dropout: 0.0
|
| 24 |
+
lossconfig:
|
| 25 |
+
target: taming.modules.losses.vqperceptual.VQLPIPSWithDiscriminator
|
| 26 |
+
params:
|
| 27 |
+
disc_conditional: false
|
| 28 |
+
disc_in_channels: 3
|
| 29 |
+
disc_num_layers: 2
|
| 30 |
+
disc_start: 1
|
| 31 |
+
disc_weight: 0.6
|
| 32 |
+
codebook_weight: 1.0
|
| 33 |
+
data:
|
| 34 |
+
target: main.DataModuleFromConfig
|
| 35 |
+
params:
|
| 36 |
+
batch_size: 10
|
| 37 |
+
num_workers: 20
|
| 38 |
+
wrap: true
|
| 39 |
+
train:
|
| 40 |
+
target: ldm.data.openimages.FullOpenImagesTrain
|
| 41 |
+
params:
|
| 42 |
+
size: 384
|
| 43 |
+
crop_size: 256
|
| 44 |
+
validation:
|
| 45 |
+
target: ldm.data.openimages.FullOpenImagesValidation
|
| 46 |
+
params:
|
| 47 |
+
size: 384
|
| 48 |
+
crop_size: 256
|
vq-f8/model.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7a2961629ea47632ec8f8cb6b5e5e8ad4a9c5baa48a3c6592ef980d4d9225cb7
|
| 3 |
+
size 879771703
|