Ref2VA/video_vae/source/config.json
| 1 | { |
| 2 | "_class_name": "AutoencoderKLLegacy", |
| 3 | "_diffusers_version": "0.32.2", |
| 4 | "causal_decoder": false, |
| 5 | "causal_encoder": true, |
| 6 | "ch": 128, |
| 7 | "ch_mult": [ |
| 8 | 1, |
| 9 | 2, |
| 10 | 2, |
| 11 | 4, |
| 12 | 4, |
| 13 | 8 |
| 14 | ], |
| 15 | "embed_dim": 24, |
| 16 | "in_channels": 3, |
| 17 | "num_res_blocks": 2, |
| 18 | "num_res_blocks_decoder": null, |
| 19 | "out_ch": 3, |
| 20 | "padding_mode": "reflect", |
| 21 | "padding_mode_t": null, |
| 22 | "pixel_norm_type": "imagenet", |
| 23 | "scaling_factor": 1.0, |
| 24 | "shift_factor": 0.0, |
| 25 | "space_down": [ |
| 26 | 2, |
| 27 | 2, |
| 28 | 2, |
| 29 | 2, |
| 30 | 1, |
| 31 | 1 |
| 32 | ], |
| 33 | "space_up": [ |
| 34 | 1, |
| 35 | 2, |
| 36 | 2, |
| 37 | 2, |
| 38 | 2, |
| 39 | 1 |
| 40 | ], |
| 41 | "time_down": [ |
| 42 | 1, |
| 43 | 2, |
| 44 | 2, |
| 45 | 1, |
| 46 | 1, |
| 47 | 1 |
| 48 | ], |
| 49 | "time_up": null, |
| 50 | "use_3d_conv": true, |
| 51 | "use_t_isolated_gn": true, |
| 52 | "use_vit_decoder": true, |
| 53 | "vae_ratio": 16, |
| 54 | "vae_ratio_t": 4, |
| 55 | "vit_decoder_kwargs": { |
| 56 | "dim_head": 64, |
| 57 | "ffn_activation_fn": "silu", |
| 58 | "ffn_use_gated": true, |
| 59 | "heads": 32, |
| 60 | "norm_affine": true, |
| 61 | "norm_type": "rms_norm", |
| 62 | "num_layers": 36, |
| 63 | "qk_norm_affine": false, |
| 64 | "qk_norm_type": "rms_norm", |
| 65 | "rope_dim_ratio": 0.75, |
| 66 | "rope_theta": 100.0 |
| 67 | }, |
| 68 | "z_channels": 24, |
| 69 | "zq_ch_decoder": null, |
| 70 | "zq_ch_encoder": null |
| 71 | } |
| 72 | |