update

transformer/config.json

@@ -4,29 +4,21 @@
   "activation_fn": "gelu-approximate",
   "attention_bias": true,
   "attention_head_dim": 96,
-  "ca_attention_mode": "xformers",
   "caption_channels": 4096,
   "cross_attention_dim": 2304,
-  "double_self_attention": false,
-  "downsampler": null,
   "dropout": 0.0,
   "in_channels": 4,
   "interpolation_scale_h": 2.0,
   "interpolation_scale_t": 2.2,
   "interpolation_scale_w": 2.0,
-  "model_max_length": 300,
   "norm_elementwise_affine": false,
   "norm_eps": 1e-06,
   "norm_type": "ada_norm_single",
   "num_attention_heads": 24,
-  "num_embeds_ada_norm": 1000,
   "num_layers": 32,
-  "only_cross_attention": false,
   "out_channels": 4,
   "patch_size": 2,
   "patch_size_t": 1,
-  "patch_size_temporal": 1,
-  "sa_attention_mode": "flash",
   "sample_frames": 22,
   "sample_height": 90,
   "sample_size": [
@@ -34,9 +26,5 @@
     160
   ],
   "sample_size_t": 22,
-  "sample_width": 160,
-  "upcast_attention": false,
-  "use_additional_conditions": null,
-  "use_linear_projection": false,
-  "use_rope": true
+  "sample_width": 160
 }

vae/config.json

@@ -8,20 +8,6 @@
     512,
     512
   ],
-  "blocks_tempdown_li": [
-    true,
-    true,
-    false,
-    false
-  ],
-  "blocks_tempup_li": [
-    false,
-    true,
-    true,
-    false
-  ],
-  "chunk_len": 24,
-  "down_block_num": 4,
   "down_block_types": [
     "AllegroDownBlock3D",
     "AllegroDownBlock3D",
@@ -32,13 +18,10 @@
   "in_channels": 3,
   "latent_channels": 4,
   "layers_per_block": 2,
-  "load_mode": "full",
   "norm_num_groups": 32,
   "out_channels": 3,
   "sample_size": 320,
-  "scale_factor": 0.13,
-  "scaling_factor": 0.13235,
-  "t_over": 8,
+  "scaling_factor": 0.13,
   "temporal_compression_ratio": 4,
   "temporal_downsample_blocks": [
     true,
@@ -52,11 +35,6 @@
     true,
     false
   ],
-  "tile_overlap": [
-    120,
-    80
-  ],
-  "up_block_num": 4,
   "up_block_types": [
     "AllegroUpBlock3D",
     "AllegroUpBlock3D",