fluxdev/stable-diffusion-webui-forge
1
1# File modified by authors of InstructPix2Pix from original (https://github.com/CompVis/stable-diffusion).2# See more details in LICENSE.3 4model:5 base_learning_rate: 1.0e-046 target: modules.models.diffusion.ddpm_edit.LatentDiffusion7 params:8 linear_start: 0.000859 linear_end: 0.012010 num_timesteps_cond: 111 log_every_t: 20012 timesteps: 100013 first_stage_key: edited14 cond_stage_key: edit15 # image_size: 6416 # image_size: 3217 image_size: 1618 channels: 419 cond_stage_trainable: false # Note: different from the one we trained before20 conditioning_key: hybrid21 monitor: val/loss_simple_ema22 scale_factor: 0.1821523 use_ema: false24 25 scheduler_config: # 10000 warmup steps26 target: ldm.lr_scheduler.LambdaLinearScheduler27 params:28 warm_up_steps: [ 0 ]29 cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases30 f_start: [ 1.e-6 ]31 f_max: [ 1. ]32 f_min: [ 1. ]33 34 unet_config:35 target: ldm.modules.diffusionmodules.openaimodel.UNetModel36 params:37 image_size: 32 # unused38 in_channels: 839 out_channels: 440 model_channels: 32041 attention_resolutions: [ 4, 2, 1 ]42 num_res_blocks: 243 channel_mult: [ 1, 2, 4, 4 ]44 num_heads: 845 use_spatial_transformer: True46 transformer_depth: 147 context_dim: 76848 use_checkpoint: True49 legacy: False50 51 first_stage_config:52 target: ldm.models.autoencoder.AutoencoderKL53 params:54 embed_dim: 455 monitor: val/rec_loss56 ddconfig:57 double_z: true58 z_channels: 459 resolution: 25660 in_channels: 361 out_ch: 362 ch: 12863 ch_mult:64 - 165 - 266 - 467 - 468 num_res_blocks: 269 attn_resolutions: []70 dropout: 0.071 lossconfig:72 target: torch.nn.Identity73 74 cond_stage_config:75 target: ldm.modules.encoders.modules.FrozenCLIPEmbedder76 77data:78 target: main.DataModuleFromConfig79 params:80 batch_size: 12881 num_workers: 182 wrap: false83 validation:84 target: edit_dataset.EditDataset85 params:86 path: data/clip-filtered-dataset87 cache_dir: data/88 cache_name: data_10k89 split: val90 min_text_sim: 0.291 min_image_sim: 0.7592 min_direction_sim: 0.293 max_samples_per_prompt: 194 min_resize_res: 51295 max_resize_res: 51296 crop_res: 51297 output_as_edit: False98 real_input: True99 