Views
No views yet
1model:
2 transformer_config:
3 remaining_frames_method: "own_embeddings"
4 transformer_type: "gpt2-large"
5 first_stage_config:
6 vqvae_config:
7 beta: 0.25
8 num_embeddings: 50257
9 embedding_dim: 128
10 autoencoder_config:
11 z_channels: 512
12 channels: 32
13 channels_multiplier:
14 - 2
15 - 4
16 - 8
17 - 8
18 num_res_blocks: 1
19 attention_resolution:
20 - 16
21 resolution: 128
22 dropout: 0.0
23 discriminator_config:
24 num_layers: 3
25 filters: 64
26
27 loss_config:
28 discriminator:
29 loss: "hinge"
30 factor: 1.0
31 iter_start: 16200
32 weight: 0.3
33 vqvae:
34 codebook_weight: 1.0
35 perceptual_weight: 4.0
36 perceptual_loss: "vgg19"
37
38train:
39 batch_size: 64
40 accumulation_size: 1
41 n_epochs: 10000
42 len_x_train: 28213
43 warmup_epoch_percentage: 0.15
44 lr_start: 1e-5
45 lr_max: 2.5e-4
46 perceptual_loss_weight: 1.0
47 n_frames_before: 1
48 stop_ground_truth_after_epoch: 1000