| defaults: | |
| - base_pytorch_algo | |
| model: | |
| hidden_size: 128 | |
| temporal_length: 17 | |
| z_channels: 8 | |
| hidden_size_mult: [1, 2, 4, 4] | |
| attn_resolutions: [] | |
| dropout: 0.0 | |
| resolution: ${dataset.resolution} | |
| embed_dim: ${algorithm.model.z_channels} | |
| num_res_blocks: 2 | |
| is_causal: True | |
| first_padding_mode: same | |
| loss: | |
| disc_start: 60001 | |
| disc_weight: 0.5 | |
| kl_weight: 5e-6 | |
| perceptual_weight: 1.0 | |
| loss_type: l1 | |
| training: | |
| warmup_steps: 20000 | |
| gradient_clip_val: 1.0 | |
| optimizer_beta: [0.9, 0.99] | |
| video_lengths: ["${algorithm.model.temporal_length}"] | |
| video_length_probs: [1.0] | |
| validation: | |
| video_lengths: ["${algorithm.model.temporal_length}"] | |
| logging: | |
| metrics: [lpips, fid, fvd, is, mse, ssim, psnr] | |
| loss_freq: 100 | |
| max_num_videos: 64 | |