Update `unet`'s config file

unet/config.json

@@ -5,7 +5,9 @@
   "addition_embed_type": null,
   "addition_embed_type_num_heads": 64,
   "addition_time_embed_dim": null,
+  "attention_bias": true,
   "attention_head_dim": 8,
+  "attention_pre_only": true,
   "attention_type": "default",
   "block_out_channels": [
     256,
@@ -18,7 +20,7 @@
   "conv_in_kernel": 3,
   "conv_out_kernel": 3,
   "cross_attention_dim": 2048,
-  "cross_attention_norm": null,
+  "cross_attention_norm": "layer_norm",
   "down_block_types": [
     "DownBlock2D",
     "CrossAttnDownBlock2D",
@@ -30,6 +32,7 @@
   "encoder_hid_dim": null,
   "encoder_hid_dim_type": null,
   "ff_act_fn": "gelu",
+  "ff_norm_type": "group_norm_matryoshka",
   "flip_sin_to_cos": true,
   "freq_shift": 0,
   "in_channels": 3,
@@ -39,6 +42,7 @@
   "mid_block_type": "UNetMidBlock2DCrossAttn",
   "norm_eps": 1e-05,
   "norm_num_groups": 32,
+  "norm_type": "layer_norm_matryoshka",
   "num_attention_heads": null,
   "num_class_embeds": null,
   "only_cross_attention": false,
@@ -48,7 +52,7 @@
   "resnet_skip_time_act": false,
   "resnet_time_scale_shift": "scale_shift",
   "reverse_transformer_layers_per_block": null,
-  "sample_size": 64,
+  "sample_size": null,
   "time_cond_proj_dim": null,
   "time_embedding_act_fn": null,
   "time_embedding_dim": null,
@@ -65,5 +69,5 @@
     "UpBlock2D"
   ],
   "upcast_attention": false,
-  "use_linear_projection": false
+  "use_linear_projection": "no_projection"
 }