Upload new ct_imagenet64 consistency model checkpoint (fix mid block attention group norm bug) (#1)

- Upload new ct_imagenet64 consistency model checkpoint (fix mid block attention group norm bug) (59bf6f8db33514ac1761753c1528b337bad166a8)


Co-authored-by: Daniel Gu <dg845@users.noreply.huggingface.co>

model_index.json

@@ -1,6 +1,7 @@
 {
   "_class_name": "ConsistencyModelPipeline",
-  "_diffusers_version": "0.17.0.dev0",
+  "_diffusers_version": "0.21.0.dev0",
+  "_name_or_path": "ct_imagenet64",
   "scheduler": [
     "diffusers",
     "CMStochasticIterativeScheduler"

scheduler/scheduler_config.json

@@ -1,6 +1,6 @@
 {
   "_class_name": "CMStochasticIterativeScheduler",
-  "_diffusers_version": "0.17.0.dev0",
+  "_diffusers_version": "0.21.0.dev0",
   "clip_denoised": true,
   "num_train_timesteps": 201,
   "rho": 7.0,

unet/config.json

@@ -1,9 +1,11 @@
 {
   "_class_name": "UNet2DModel",
-  "_diffusers_version": "0.17.0.dev0",
+  "_diffusers_version": "0.21.0.dev0",
+  "_name_or_path": "ct_imagenet64/unet",
   "act_fn": "silu",
   "add_attention": true,
   "attention_head_dim": 64,
+  "attn_norm_num_groups": 32,
   "block_out_channels": [
     192,
     384,
@@ -19,6 +21,7 @@
     "AttnDownBlock2D"
   ],
   "downsample_padding": 1,
+  "downsample_type": "resnet",
   "flip_sin_to_cos": true,
   "freq_shift": 0,
   "in_channels": 3,
@@ -31,12 +34,11 @@
   "resnet_time_scale_shift": "scale_shift",
   "sample_size": 64,
   "time_embedding_type": "positional",
-  "upsample_type": "resnet",
-  "downsample_type": "resnet",
   "up_block_types": [
     "AttnUpBlock2D",
     "AttnUpBlock2D",
     "AttnUpBlock2D",
     "ResnetUpsampleBlock2D"
-  ]
+  ],
+  "upsample_type": "resnet"
 }

unet/diffusion_pytorch_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf93ba5bd7a176927b78b55eec357bacc22f42940d262c401b4ed051b537c4f1
+size 1183665484