reorg configs

configs/inference/inference-v1.yaml

@@ -1,26 +1,23 @@
 unet_additional_kwargs:
   unet_use_cross_frame_attention: false
-  unet_use_temporal_attention: false
-  use_motion_module: true
-  motion_module_resolutions:
-  - 1
-  - 2
-  - 4
-  - 8
-  motion_module_mid_block: false
-  motion_module_decoder_only: false
-  motion_module_type: Vanilla
+  unet_use_temporal_attention:    false
+  use_motion_module:              true
+  motion_module_resolutions:      [1,2,4,8]
+  motion_module_mid_block:        false
+  motion_module_decoder_only:     false
+  motion_module_type:             "Vanilla"
+  
   motion_module_kwargs:
-    num_attention_heads: 8
-    num_transformer_block: 1
-    attention_block_types:
-    - Temporal_Self
-    - Temporal_Self
-    temporal_position_encoding: true
+    num_attention_heads:                8
+    num_transformer_block:              1
+    attention_block_types:              [ "Temporal_Self", "Temporal_Self" ]
+    temporal_position_encoding:         true
     temporal_position_encoding_max_len: 24
-    temporal_attention_dim_div: 1
+    temporal_attention_dim_div:         1
 
 noise_scheduler_kwargs:
-  beta_start: 0.00085
-  beta_end: 0.012
+  beta_start:    0.00085
+  beta_end:      0.012
   beta_schedule: "linear"
+  steps_offset:  1
+  clip_sample:   False

configs/inference/inference-v2.yaml

@@ -1,27 +1,24 @@
 unet_additional_kwargs:
-  use_inflated_groupnorm: true
+  use_inflated_groupnorm:         true
   unet_use_cross_frame_attention: false
-  unet_use_temporal_attention: false
-  use_motion_module: true
-  motion_module_resolutions:
-  - 1
-  - 2
-  - 4
-  - 8
-  motion_module_mid_block: true
-  motion_module_decoder_only: false
-  motion_module_type: Vanilla
+  unet_use_temporal_attention:    false
+  use_motion_module:              true
+  motion_module_resolutions:      [1,2,4,8]
+  motion_module_mid_block:        true
+  motion_module_decoder_only:     false
+  motion_module_type:             "Vanilla"
+
   motion_module_kwargs:
-    num_attention_heads: 8
-    num_transformer_block: 1
-    attention_block_types:
-    - Temporal_Self
-    - Temporal_Self
-    temporal_position_encoding: true
+    num_attention_heads:                8
+    num_transformer_block:              1
+    attention_block_types:              [ "Temporal_Self", "Temporal_Self" ]
+    temporal_position_encoding:         true
     temporal_position_encoding_max_len: 32
-    temporal_attention_dim_div: 1
+    temporal_attention_dim_div:         1
 
 noise_scheduler_kwargs:
-  beta_start: 0.00085
-  beta_end: 0.012
+  beta_start:    0.00085
+  beta_end:      0.012
   beta_schedule: "linear"
+  steps_offset:  1
+  clip_sample:   False

configs/inference/inference-v3.yaml

@@ -0,0 +1,22 @@
+unet_additional_kwargs:
+  use_inflated_groupnorm:     true
+  use_motion_module:          true
+  motion_module_resolutions:  [1,2,4,8]
+  motion_module_mid_block:    false
+  motion_module_type:         Vanilla
+
+  motion_module_kwargs:
+    num_attention_heads:                 8
+    num_transformer_block:               1
+    attention_block_types:               [ "Temporal_Self", "Temporal_Self" ]
+    temporal_position_encoding:          true
+    temporal_position_encoding_max_len:  32
+    temporal_attention_dim_div:          1
+    zero_initialize:                     true
+
+noise_scheduler_kwargs:
+  beta_start:    0.00085
+  beta_end:      0.012
+  beta_schedule: "linear"
+  steps_offset:  1
+  clip_sample:   False

configs/inference/sparsectrl/image_condition.yaml

@@ -0,0 +1,17 @@
+controlnet_additional_kwargs:
+  set_noisy_sample_input_to_zero:     true
+  use_simplified_condition_embedding: false
+  conditioning_channels:              3
+
+  use_motion_module:         true
+  motion_module_resolutions: [1,2,4,8]
+  motion_module_mid_block:   false
+  motion_module_type:        "Vanilla"
+
+  motion_module_kwargs:
+    num_attention_heads:                8
+    num_transformer_block:              1
+    attention_block_types:              [ "Temporal_Self" ]
+    temporal_position_encoding:         true
+    temporal_position_encoding_max_len: 32
+    temporal_attention_dim_div:         1

configs/inference/sparsectrl/latent_condition.yaml

@@ -0,0 +1,17 @@
+controlnet_additional_kwargs:
+  set_noisy_sample_input_to_zero:     true
+  use_simplified_condition_embedding: true
+  conditioning_channels:              4
+
+  use_motion_module:         true
+  motion_module_resolutions: [1,2,4,8]
+  motion_module_mid_block:   false
+  motion_module_type:        "Vanilla"
+
+  motion_module_kwargs:
+    num_attention_heads:                8
+    num_transformer_block:              1
+    attention_block_types:              [ "Temporal_Self" ]
+    temporal_position_encoding:         true
+    temporal_position_encoding_max_len: 32
+    temporal_attention_dim_div:         1