First model version

scheduler/scheduler_config.json

@@ -9,7 +9,7 @@
   "dynamic_thresholding_ratio": 0.995,
   "lower_order_final": true,
   "num_train_timesteps": 1000,
-  "prediction_type": "v_prediction",
+  "prediction_type": "epsilon",
   "sample_max_value": 1.0,
   "set_alpha_to_one": false,
   "solver_order": 2,

unet/config.json

@@ -40,7 +40,7 @@
   "out_channels": 4,
   "projection_class_embeddings_input_dim": null,
   "resnet_time_scale_shift": "default",
-  "sample_size": 96,
+  "sample_size": 64,
   "time_cond_proj_dim": null,
   "time_embedding_type": "positional",
   "timestep_post_act": null,
@@ -50,6 +50,6 @@
     "CrossAttnUpBlock2D",
     "CrossAttnUpBlock2D"
   ],
-  "upcast_attention": true,
+  "upcast_attention": false,
   "use_linear_projection": true
 }

vae/config.json

@@ -19,7 +19,7 @@
   "layers_per_block": 2,
   "norm_num_groups": 32,
   "out_channels": 3,
-  "sample_size": 768,
+  "sample_size": 512,
   "scaling_factor": 0.18215,
   "up_block_types": [
     "UpDecoderBlock2D",