ENV: BACKEND: nccl SOLVER: NAME: LatentDiffusionSolver RESUME_FROM: LOAD_MODEL_ONLY: True USE_FSDP: False SHARDING_STRATEGY: USE_AMP: True DTYPE: float16 CHANNELS_LAST: True MAX_STEPS: 2000 MAX_EPOCHS: -1 NUM_FOLDS: 1 ACCU_STEP: 1 EVAL_INTERVAL: 100 RESCALE_LR: False # WORK_DIR: ./cache/save_data/sdxl_1024_full LOG_FILE: std_log.txt # FILE_SYSTEM: NAME: "ModelscopeFs" TEMP_DIR: "./cache/cache_data" # MODEL: NAME: LatentDiffusionXL PARAMETERIZATION: eps TIMESTEPS: 1000 MIN_SNR_GAMMA: ZERO_TERMINAL_SNR: False PRETRAINED_MODEL: ms://AI-ModelScope/stable-diffusion-xl-base-1.0@sd_xl_base_1.0.safetensors IGNORE_KEYS: [ ] SCALE_FACTOR: 0.13025 SIZE_FACTOR: 8 # DEFAULT_N_PROMPT: 'lowres, error, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, out of frame, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck, username, watermark, signature' DEFAULT_N_PROMPT: SCHEDULE_ARGS: "NAME": "scaled_linear" "BETA_MIN": 0.00085 "BETA_MAX": 0.0120 USE_EMA: False LOAD_REFINER: False # DIFFUSION_MODEL: NAME: DiffusionUNetXL PRETRAINED_MODEL: IN_CHANNELS: 4 OUT_CHANNELS: 4 NUM_RES_BLOCKS: 2 MODEL_CHANNELS: 320 ATTENTION_RESOLUTIONS: [ 4, 2 ] DROPOUT: 0 CHANNEL_MULT: [ 1, 2, 4 ] CONV_RESAMPLE: True DIMS: 2 NUM_CLASSES: sequential USE_CHECKPOINT: False NUM_HEADS: -1 NUM_HEADS_CHANNELS: 64 USE_SCALE_SHIFT_NORM: False RESBLOCK_UPDOWN: False USE_NEW_ATTENTION_ORDER: True USE_SPATIAL_TRANSFORMER: True TRANSFORMER_DEPTH: [ 1, 2, 10 ] CONTEXT_DIM: 2048 DISABLE_MIDDLE_SELF_ATTN: False USE_LINEAR_IN_TRANSFORMER: True ADM_IN_CHANNELS: 2816 USE_SENTENCE_EMB: False USE_WORD_MAPPING: False # FIRST_STAGE_MODEL: NAME: AutoencoderKL EMBED_DIM: 4 PRETRAINED_MODEL: IGNORE_KEYS: [] BATCH_SIZE: 1 # ENCODER: NAME: Encoder CH: 128 OUT_CH: 3 NUM_RES_BLOCKS: 2 IN_CHANNELS: 3 ATTN_RESOLUTIONS: [ ] CH_MULT: [ 1, 2, 4, 4 ] Z_CHANNELS: 4 DOUBLE_Z: True DROPOUT: 0.0 RESAMP_WITH_CONV: True # DECODER: NAME: Decoder CH: 128 OUT_CH: 3 NUM_RES_BLOCKS: 2 IN_CHANNELS: 3 ATTN_RESOLUTIONS: [ ] CH_MULT: [ 1, 2, 4, 4 ] Z_CHANNELS: 4 DROPOUT: 0.0 RESAMP_WITH_CONV: True GIVE_PRE_END: False TANH_OUT: False # COND_STAGE_MODEL: NAME: GeneralConditioner PRETRAINED_MODEL: EMBEDDERS: - NAME: FrozenCLIPEmbedder PRETRAINED_MODEL: ms://AI-ModelScope/clip-vit-large-patch14 TOKENIZER_PATH: ms://AI-ModelScope/clip-vit-large-patch14 MAX_LENGTH: 77 FREEZE: True LAYER: hidden LAYER_IDX: 11 USE_FINAL_LAYER_NORM: False IS_TRAINABLE: False UCG_RATE: 0.0 INPUT_KEYS: [ "prompt" ] LEGACY_UCG_VALUE: - NAME: FrozenOpenCLIPEmbedder2 ARCH: ViT-bigG-14 PRETRAINED_MODEL: MAX_LENGTH: 77 FREEZE: True ALWAYS_RETURN_POOLED: True LEGACY: False LAYER: penultimate IS_TRAINABLE: False UCG_RATE: 0.0 INPUT_KEYS: [ "prompt" ] LEGACY_UCG_VALUE: - NAME: ConcatTimestepEmbedderND OUT_DIM: 256 IS_TRAINABLE: False UCG_RATE: 0.0 INPUT_KEYS: [ "original_size_as_tuple" ] LEGACY_UCG_VALUE: - NAME: ConcatTimestepEmbedderND OUT_DIM: 256 IS_TRAINABLE: False UCG_RATE: 0.0 INPUT_KEYS: [ "crop_coords_top_left" ] LEGACY_UCG_VALUE: - NAME: ConcatTimestepEmbedderND OUT_DIM: 256 IS_TRAINABLE: False UCG_RATE: 0.0 INPUT_KEYS: [ "target_size_as_tuple" ] LEGACY_UCG_VALUE: # REFINER_MODEL: NAME: DiffusionUNetXL PRETRAINED_MODEL: IN_CHANNELS: 4 OUT_CHANNELS: 4 NUM_RES_BLOCKS: 2 MODEL_CHANNELS: 384 ATTENTION_RESOLUTIONS: [ 4, 2 ] DROPOUT: 0 CHANNEL_MULT: [ 1, 2, 4, 4 ] CONV_RESAMPLE: True DIMS: 2 NUM_CLASSES: sequential USE_CHECKPOINT: False NUM_HEADS: -1 NUM_HEADS_CHANNELS: 64 USE_SCALE_SHIFT_NORM: False RESBLOCK_UPDOWN: False USE_NEW_ATTENTION_ORDER: True USE_SPATIAL_TRANSFORMER: True TRANSFORMER_DEPTH: 4 CONTEXT_DIM: [ 1280, 1280, 1280, 1280 ] DISABLE_MIDDLE_SELF_ATTN: False USE_LINEAR_IN_TRANSFORMER: True ADM_IN_CHANNELS: 2560 USE_SENTENCE_EMB: False USE_WORD_MAPPING: False # REFINER_COND_MODEL: NAME: GeneralConditioner PRETRAINED_MODEL: EMBEDDERS: - NAME: FrozenOpenCLIPEmbedder2 ARCH: ViT-bigG-14 PRETRAINED_MODEL: MAX_LENGTH: 77 FREEZE: True ALWAYS_RETURN_POOLED: True LEGACY: False LAYER: penultimate IS_TRAINABLE: False UCG_RATE: 0.0 INPUT_KEYS: [ "prompt" ] LEGACY_UCG_VALUE: - NAME: ConcatTimestepEmbedderND OUT_DIM: 256 IS_TRAINABLE: False UCG_RATE: 0.0 INPUT_KEYS: [ "original_size_as_tuple" ] LEGACY_UCG_VALUE: - NAME: ConcatTimestepEmbedderND OUT_DIM: 256 IS_TRAINABLE: False UCG_RATE: 0.0 INPUT_KEYS: [ "crop_coords_top_left" ] LEGACY_UCG_VALUE: - NAME: ConcatTimestepEmbedderND OUT_DIM: 256 IS_TRAINABLE: False UCG_RATE: 0.0 INPUT_KEYS: [ "aesthetic_score" ] LEGACY_UCG_VALUE: # LOSS: NAME: ReconstructLoss LOSS_TYPE: l2 # SAMPLE_ARGS: SAMPLER: ddim SAMPLE_STEPS: 50 SEED: 2023 GUIDE_SCALE: 5.0 GUIDE_RESCALE: DISCRETIZATION: trailing IMAGE_SIZE: [1024, 1024] RUN_TRAIN_N: False # OPTIMIZER: NAME: AdamW LEARNING_RATE: 0.00001 BETAS: [ 0.9, 0.999 ] EPS: 1e-8 WEIGHT_DECAY: 1e-2 AMSGRAD: False # TRAIN_DATA: NAME: ImageTextPairMSDataset MODE: train MS_DATASET_NAME: style_custom_dataset MS_DATASET_NAMESPACE: damo MS_DATASET_SUBNAME: 3D PROMPT_PREFIX: "" MS_DATASET_SPLIT: train_short MS_REMAP_KEYS: { 'Image:FILE': 'Target:FILE' } REPLACE_STYLE: False PIN_MEMORY: True BATCH_SIZE: 1 NUM_WORKERS: 4 SAMPLER: NAME: LoopSampler TRANSFORMS: - NAME: LoadImageFromFile RGB_ORDER: RGB BACKEND: pillow - NAME: FlexibleResize INTERPOLATION: bicubic SIZE: 1024 INPUT_KEY: [ 'img' ] OUTPUT_KEY: [ 'img' ] BACKEND: pillow - NAME: FlexibleCropXL INPUT_KEY: [ 'img' ] OUTPUT_KEY: [ 'img' ] BACKEND: pillow - NAME: ImageToTensor INPUT_KEY: [ 'img' ] OUTPUT_KEY: [ 'img' ] BACKEND: pillow - NAME: Normalize MEAN: [ 0.5, 0.5, 0.5 ] STD: [ 0.5, 0.5, 0.5 ] INPUT_KEY: [ 'img' ] OUTPUT_KEY: [ 'img' ] BACKEND: torchvision - NAME: Select KEYS: [ 'img', 'prompt', 'img_original_size_as_tuple', 'img_target_size_as_tuple', 'img_crop_coords_top_left' ] META_KEYS: [ 'data_key', 'img_path' ] - NAME: Rename INPUT_KEY: [ 'img', 'img_original_size_as_tuple', 'img_target_size_as_tuple', 'img_crop_coords_top_left' ] OUTPUT_KEY: [ 'image', 'original_size_as_tuple', 'target_size_as_tuple', 'crop_coords_top_left' ] # EVAL_DATA: NAME: ImageTextPairMSDataset MODE: eval MS_DATASET_NAME: style_custom_dataset MS_DATASET_NAMESPACE: damo MS_DATASET_SUBNAME: 3D PROMPT_PREFIX: "" MS_REMAP_KEYS: { 'Image': 'Target:FILE' } MS_DATASET_SPLIT: test_short OUTPUT_SIZE: [ 1024, 1024 ] REPLACE_STYLE: False PIN_MEMORY: True BATCH_SIZE: 2 NUM_WORKERS: 4 FILE_SYSTEM: NAME: "ModelscopeFs" TEMP_DIR: "./cache/cache_data" # TRANSFORMS: - NAME: Select KEYS: [ 'prompt' ] META_KEYS: [ 'image_size' ] # TRAIN_HOOKS: - NAME: BackwardHook PRIORITY: 0 - NAME: LogHook LOG_INTERVAL: 50 - NAME: CheckpointHook INTERVAL: 1000 - NAME: ProbeDataHook PROB_INTERVAL: 100 # EVAL_HOOKS: - NAME: ProbeDataHook PROB_INTERVAL: 100