| { | |
| "depth": 24, | |
| "mlp_ratio": 4, | |
| "vae_latent_dim": 16, | |
| "layer_norm_eps": 1e-06, | |
| "max_latent_resolution": 96, | |
| "patch_size": 2, | |
| "pooled_text_embed_dim": 2048, | |
| "token_level_text_embed_dim": 4096, | |
| "frequency_embed_dim": 256, | |
| "max_period": 10000 | |
| } |
| { | |
| "depth": 24, | |
| "mlp_ratio": 4, | |
| "vae_latent_dim": 16, | |
| "layer_norm_eps": 1e-06, | |
| "max_latent_resolution": 96, | |
| "patch_size": 2, | |
| "pooled_text_embed_dim": 2048, | |
| "token_level_text_embed_dim": 4096, | |
| "frequency_embed_dim": 256, | |
| "max_period": 10000 | |
| } |