| { | |
| "_class_name": "AutoencoderVQ", | |
| "_quantizer_name": "VQuantizer", | |
| "in_channels": 3, | |
| "latent_channels": 256, | |
| "layers_per_block": 4, | |
| "norm_num_groups": 32, | |
| "out_channels": 3, | |
| "sample_size": 1024, | |
| "num_vq_embeddings": 131072, | |
| "vq_embed_dim": 256, | |
| "temporal_stride": 1, | |
| "spatial_stride": 16, | |
| "attn_down_block": true, | |
| "attn_up_block": true, | |
| "decoder_dtype": "bfloat16", | |
| "block_out_channels": [ | |
| 256, | |
| 256, | |
| 512, | |
| 512, | |
| 1024 | |
| ] | |
| } | |