| { | |
| "base_learning_rate": 2e-05, | |
| "cfg_conds": [ | |
| "txt" | |
| ], | |
| "cfg_scales": [ | |
| 7.5 | |
| ], | |
| "channels": 4, | |
| "cond_stage_config": { | |
| "ckpt_path": null, | |
| "target": "ldm.modules.encoders.modules.FrozenCLIPEmbedder" | |
| }, | |
| "cond_stage_key": null, | |
| "cond_stage_trainable": false, | |
| "conditioning_key": "hybrid-mv", | |
| "fast_attention": true, | |
| "first_stage_config": { | |
| "params": { | |
| "ddconfig": { | |
| "attn_resolutions": [], | |
| "ch": 128, | |
| "ch_mult": [ | |
| 1, | |
| 2, | |
| 4, | |
| 4 | |
| ], | |
| "double_z": true, | |
| "dropout": 0.0, | |
| "in_channels": 3, | |
| "num_res_blocks": 2, | |
| "out_ch": 3, | |
| "resolution": 256, | |
| "z_channels": 4 | |
| }, | |
| "embed_dim": 4, | |
| "lossconfig": { | |
| "target": "torch.nn.Identity" | |
| }, | |
| "monitor": "val/rec_loss" | |
| }, | |
| "target": "ldm.models.autoencoder.AutoencoderKL" | |
| }, | |
| "first_stage_key": null, | |
| "image_size": 32, | |
| "linear_end": 0.012, | |
| "linear_start": 0.00085, | |
| "log_every_t": 200, | |
| "model_type": "spad", | |
| "monitor": "val/loss_simple_ema", | |
| "num_timesteps_cond": 1, | |
| "resume_path": "data/v1-5-pruned.ckpt", | |
| "scale_factor": 0.18215, | |
| "scheduler_config": { | |
| "params": { | |
| "cycle_lengths": [ | |
| 10000000000000 | |
| ], | |
| "f_max": [ | |
| 1.0 | |
| ], | |
| "f_min": [ | |
| 1.0 | |
| ], | |
| "f_start": [ | |
| 1e-06 | |
| ], | |
| "warm_up_steps": [ | |
| 100 | |
| ] | |
| }, | |
| "target": "ldm.lr_scheduler.LambdaLinearScheduler" | |
| }, | |
| "timesteps": 1000, | |
| "transformers_version": "4.42.4", | |
| "unet_config": { | |
| "params": { | |
| "attention_resolutions": [ | |
| 4, | |
| 2, | |
| 1 | |
| ], | |
| "channel_mult": [ | |
| 1, | |
| 2, | |
| 4, | |
| 4 | |
| ], | |
| "context_dim": 768, | |
| "image_size": 32, | |
| "in_channels": 4, | |
| "legacy": false, | |
| "model_channels": 320, | |
| "num_heads": 8, | |
| "num_res_blocks": 2, | |
| "out_channels": 4, | |
| "transformer_depth": 1, | |
| "use_checkpoint": true, | |
| "use_spatial_transformer": true | |
| }, | |
| "target": "spad.mv_unet.SPADUnetModel" | |
| }, | |
| "use_abs_extrinsics": true, | |
| "use_intrinsic": true | |
| } | |