Add model_index.json and update modeling configs (#2)

- add model_index.json; update configs (43d06687e1b23afacab5980e452638293b3722b0)
- update (7defb981dbb797a900f485c84f483d6d21e9520a)


Co-authored-by: Aryan V S <a-r-r-o-w@users.noreply.huggingface.co>

model_index.json

@@ -0,0 +1,24 @@
+{
+  "_class_name": "AllegroPipeline",
+  "_diffusers_version": "0.31.0.dev0",
+  "scheduler": [
+    "diffusers",
+    "EulerAncestralDiscreteScheduler"
+  ],
+  "text_encoder": [
+    "transformers",
+    "T5EncoderModel"
+  ],
+  "tokenizer": [
+    "transformers",
+    "T5Tokenizer"
+  ],
+  "transformer": [
+    "diffusers",
+    "AllegroTransformer3DModel"
+  ],
+  "vae": [
+    "diffusers",
+    "AutoencoderKLAllegro"
+  ]
+}

transformer/config.json

@@ -1,38 +1,30 @@
 {
   "_class_name": "AllegroTransformer3DModel",
-  "_diffusers_version": "0.28.0",
+  "_diffusers_version": "0.31.0.dev0",
   "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,
-  "sa_attention_mode": "flash",
+  "sample_frames": 22,
+  "sample_height": 90,
   "sample_size": [
     90,
     160
   ],
   "sample_size_t": 22,
-  "upcast_attention": false,
+  "sample_width": 160
-  "use_additional_conditions": null,
-  "use_linear_projection": false,
-  "use_rope": true
 }

vae/config.json

@@ -1,6 +1,6 @@
 {
-  "_class_name": "AllegroAutoencoderKL3D",
+  "_class_name": "AutoencoderKLAllegro",
-  "_diffusers_version": "0.28.0",
+  "_diffusers_version": "0.31.0.dev0",
   "act_fn": "silu",
   "block_out_channels": [
     128,
@@ -8,33 +8,37 @@
     512,
     512
   ],
-  "blocks_tempdown_li": [
+  "down_block_types": [
-    true,
+    "AllegroDownBlock3D",
-    true,
+    "AllegroDownBlock3D",
-    false,
+    "AllegroDownBlock3D",
-    false
+    "AllegroDownBlock3D"
   ],
-  "blocks_tempup_li": [
-    false,
-    true,
-    true,
-    false
-  ],
-  "chunk_len": 24,
-  "down_block_num": 4,
   "force_upcast": true,
   "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": 0.13,
-  "t_over": 8,
+  "temporal_compression_ratio": 4,
-  "tile_overlap": [
+  "temporal_downsample_blocks": [
-    120,
+    true,
-    80
+    true,
+    false,
+    false
   ],
-  "up_block_num": 4
+  "temporal_upsample_blocks": [
+    false,
+    true,
+    true,
+    false
+  ],
+  "up_block_types": [
+    "AllegroUpBlock3D",
+    "AllegroUpBlock3D",
+    "AllegroUpBlock3D",
+    "AllegroUpBlock3D"
+  ]
 }