Upload folder using huggingface_hub

2e-5_kl_naive_globalscale_channelmean_sampling/log.txt

@@ -67,7 +67,7 @@ Disc shape (1, 16, 16, 512)
 Disc shape (1, 8, 8, 512)
 Disc shape (1, 4, 4, 512)
 Total num of Discriminator parameters: 23998017
-Loaded checkpoint from 13763719 seconds ago.
+Loaded checkpoint from 13891605 seconds ago.
 Loaded model with step 511001
 ┌──────────────────────────────────────────────────────────────────────────────┐
 │                                    TPU 0                                     │
@@ -768,7 +768,7 @@ DiT: Input of shape (4, 32, 32, 4) dtype float32
 DiT: After patch embed, shape is (4, 256, 768) dtype bfloat16
 DiT: Patch Embed of shape (4, 256, 768) dtype bfloat16
 DiT: Conditioning of shape (1, 768) dtype float32
-Loaded checkpoint from 21940 seconds ago.
+Loaded checkpoint from 149826 seconds ago.
 
  parameter shapes:
 ('PatchEmbed_0', 'Conv_0', 'kernel'): (2, 2, 4, 768)
@@ -1938,501 +1938,60 @@ Decoder layer (128, 64, 64, 512)
 Decoder layer (128, 128, 128, 512)
 Decoder layer (128, 256, 256, 256)
 Decoder layer (128, 256, 256, 128)
-FID is 36.831787109375
-(512, 256, 256, 3)
-Calc FID for CFG 1.0 and denoise_timesteps 64
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 37.423683166503906
-(512, 256, 256, 3)
-Calc FID for CFG 1.0 and denoise_timesteps 32
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 39.304908752441406
-(512, 256, 256, 3)
-Calc FID for CFG 1.0 and denoise_timesteps 16
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 45.72352600097656
-(512, 256, 256, 3)
-Calc FID for CFG 1.0 and denoise_timesteps 8
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 67.60384368896484
-(512, 256, 256, 3)
-Calc FID for CFG 1.0 and denoise_timesteps 4
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 152.43223571777344
-(512, 256, 256, 3)
-Calc FID for CFG 1.0 and denoise_timesteps 2
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 325.9117431640625
-(512, 256, 256, 3)
-Calc FID for CFG 1.0 and denoise_timesteps 1
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 265.09783935546875
+FID is 36.43708801269531
 (512, 256, 256, 3)
 Calc FID for CFG 1.25 and denoise_timesteps 128
 DiT: Input of shape (512, 32, 32, 4) dtype float32
 DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
 DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
 DiT: Conditioning of shape (512, 768) dtype float32
-FID is 22.617610931396484
-(512, 256, 256, 3)
-Calc FID for CFG 1.25 and denoise_timesteps 64
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 23.02716827392578
-(512, 256, 256, 3)
-Calc FID for CFG 1.25 and denoise_timesteps 32
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 24.476266860961914
-(512, 256, 256, 3)
-Calc FID for CFG 1.25 and denoise_timesteps 16
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 29.616016387939453
-(512, 256, 256, 3)
-Calc FID for CFG 1.25 and denoise_timesteps 8
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 49.262271881103516
-(512, 256, 256, 3)
-Calc FID for CFG 1.25 and denoise_timesteps 4
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 128.205322265625
-(512, 256, 256, 3)
-Calc FID for CFG 1.25 and denoise_timesteps 2
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 313.4750671386719
-(512, 256, 256, 3)
-Calc FID for CFG 1.25 and denoise_timesteps 1
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 256.4454345703125
+FID is 22.433515548706055
 (512, 256, 256, 3)
 Calc FID for CFG 1.5 and denoise_timesteps 128
 DiT: Input of shape (512, 32, 32, 4) dtype float32
 DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
 DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
 DiT: Conditioning of shape (512, 768) dtype float32
-FID is 14.177565574645996
-(512, 256, 256, 3)
-Calc FID for CFG 1.5 and denoise_timesteps 64
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 14.451891899108887
-(512, 256, 256, 3)
-Calc FID for CFG 1.5 and denoise_timesteps 32
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 15.438105583190918
-(512, 256, 256, 3)
-Calc FID for CFG 1.5 and denoise_timesteps 16
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 19.319408416748047
-(512, 256, 256, 3)
-Calc FID for CFG 1.5 and denoise_timesteps 8
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 35.212379455566406
-(512, 256, 256, 3)
-Calc FID for CFG 1.5 and denoise_timesteps 4
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 106.40361785888672
-(512, 256, 256, 3)
-Calc FID for CFG 1.5 and denoise_timesteps 2
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 302.155517578125
-(512, 256, 256, 3)
-Calc FID for CFG 1.5 and denoise_timesteps 1
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 249.88137817382812
+FID is 14.05518913269043
 (512, 256, 256, 3)
 Calc FID for CFG 1.75 and denoise_timesteps 128
 DiT: Input of shape (512, 32, 32, 4) dtype float32
 DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
 DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
 DiT: Conditioning of shape (512, 768) dtype float32
-FID is 9.794225692749023
-(512, 256, 256, 3)
-Calc FID for CFG 1.75 and denoise_timesteps 64
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 9.958918571472168
-(512, 256, 256, 3)
-Calc FID for CFG 1.75 and denoise_timesteps 32
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 10.63227367401123
-(512, 256, 256, 3)
-Calc FID for CFG 1.75 and denoise_timesteps 16
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 13.343984603881836
-(512, 256, 256, 3)
-Calc FID for CFG 1.75 and denoise_timesteps 8
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 25.57964324951172
-(512, 256, 256, 3)
-Calc FID for CFG 1.75 and denoise_timesteps 4
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 88.0321044921875
-(512, 256, 256, 3)
-Calc FID for CFG 1.75 and denoise_timesteps 2
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 291.9643249511719
-(512, 256, 256, 3)
-Calc FID for CFG 1.75 and denoise_timesteps 1
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 245.0141143798828
+FID is 9.713033676147461
 (512, 256, 256, 3)
 Calc FID for CFG 2.0 and denoise_timesteps 128
 DiT: Input of shape (512, 32, 32, 4) dtype float32
 DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
 DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
 DiT: Conditioning of shape (512, 768) dtype float32
-FID is 7.921169281005859
-(512, 256, 256, 3)
-Calc FID for CFG 2.0 and denoise_timesteps 64
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 8.00857162475586
-(512, 256, 256, 3)
-Calc FID for CFG 2.0 and denoise_timesteps 32
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 8.395737648010254
-(512, 256, 256, 3)
-Calc FID for CFG 2.0 and denoise_timesteps 16
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 10.18696403503418
-(512, 256, 256, 3)
-Calc FID for CFG 2.0 and denoise_timesteps 8
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 19.315452575683594
-(512, 256, 256, 3)
-Calc FID for CFG 2.0 and denoise_timesteps 4
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 73.04430389404297
-(512, 256, 256, 3)
-Calc FID for CFG 2.0 and denoise_timesteps 2
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 282.81884765625
-(512, 256, 256, 3)
-Calc FID for CFG 2.0 and denoise_timesteps 1
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 241.4156036376953
+FID is 7.850697994232178
 (512, 256, 256, 3)
 Calc FID for CFG 2.25 and denoise_timesteps 128
 DiT: Input of shape (512, 32, 32, 4) dtype float32
 DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
 DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
 DiT: Conditioning of shape (512, 768) dtype float32
-FID is 7.42802619934082
-(512, 256, 256, 3)
-Calc FID for CFG 2.25 and denoise_timesteps 64
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 7.460850715637207
-(512, 256, 256, 3)
-Calc FID for CFG 2.25 and denoise_timesteps 32
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 7.677089691162109
-(512, 256, 256, 3)
-Calc FID for CFG 2.25 and denoise_timesteps 16
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 8.823616027832031
-(512, 256, 256, 3)
-Calc FID for CFG 2.25 and denoise_timesteps 8
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 15.475367546081543
-(512, 256, 256, 3)
-Calc FID for CFG 2.25 and denoise_timesteps 4
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 61.213165283203125
-(512, 256, 256, 3)
-Calc FID for CFG 2.25 and denoise_timesteps 2
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 274.59478759765625
-(512, 256, 256, 3)
-Calc FID for CFG 2.25 and denoise_timesteps 1
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 238.5747833251953
+FID is 7.359264373779297
 (512, 256, 256, 3)
 Calc FID for CFG 2.5 and denoise_timesteps 128
 DiT: Input of shape (512, 32, 32, 4) dtype float32
 DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
 DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
 DiT: Conditioning of shape (512, 768) dtype float32
-FID is 7.72978401184082
-(512, 256, 256, 3)
-Calc FID for CFG 2.5 and denoise_timesteps 64
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 7.7087931632995605
-(512, 256, 256, 3)
-Calc FID for CFG 2.5 and denoise_timesteps 32
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 7.7968902587890625
-(512, 256, 256, 3)
-Calc FID for CFG 2.5 and denoise_timesteps 16
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 8.493557929992676
-(512, 256, 256, 3)
-Calc FID for CFG 2.5 and denoise_timesteps 8
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 13.25935173034668
-(512, 256, 256, 3)
-Calc FID for CFG 2.5 and denoise_timesteps 4
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 51.91324996948242
-(512, 256, 256, 3)
-Calc FID for CFG 2.5 and denoise_timesteps 2
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 267.32489013671875
-(512, 256, 256, 3)
-Calc FID for CFG 2.5 and denoise_timesteps 1
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 236.34414672851562
+FID is 7.6529860496521
 (512, 256, 256, 3)
 Calc FID for CFG 2.75 and denoise_timesteps 128
 DiT: Input of shape (512, 32, 32, 4) dtype float32
 DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
 DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
 DiT: Conditioning of shape (512, 768) dtype float32
-FID is 8.462559700012207
-(512, 256, 256, 3)
-Calc FID for CFG 2.75 and denoise_timesteps 64
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 8.41374683380127
-(512, 256, 256, 3)
-Calc FID for CFG 2.75 and denoise_timesteps 32
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 8.378111839294434
-(512, 256, 256, 3)
-Calc FID for CFG 2.75 and denoise_timesteps 16
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 8.747322082519531
-(512, 256, 256, 3)
-Calc FID for CFG 2.75 and denoise_timesteps 8
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 12.060856819152832
-(512, 256, 256, 3)
-Calc FID for CFG 2.75 and denoise_timesteps 4
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 44.55529022216797
-(512, 256, 256, 3)
-Calc FID for CFG 2.75 and denoise_timesteps 2
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 260.9886474609375
-(512, 256, 256, 3)
-Calc FID for CFG 2.75 and denoise_timesteps 1
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 234.50088500976562
+FID is 8.407513618469238
 (512, 256, 256, 3)
 Calc FID for CFG 3.0 and denoise_timesteps 128
 DiT: Input of shape (512, 32, 32, 4) dtype float32
 DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
 DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
 DiT: Conditioning of shape (512, 768) dtype float32
-FID is 9.388092041015625
-(512, 256, 256, 3)
-Calc FID for CFG 3.0 and denoise_timesteps 64
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 9.323236465454102
-(512, 256, 256, 3)
-Calc FID for CFG 3.0 and denoise_timesteps 32
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 9.24086856842041
-(512, 256, 256, 3)
-Calc FID for CFG 3.0 and denoise_timesteps 16
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 9.349201202392578
-(512, 256, 256, 3)
-Calc FID for CFG 3.0 and denoise_timesteps 8
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 11.579026222229004
-(512, 256, 256, 3)
-Calc FID for CFG 3.0 and denoise_timesteps 4
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 38.7520751953125
-(512, 256, 256, 3)
-Calc FID for CFG 3.0 and denoise_timesteps 2
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 255.22378540039062
-(512, 256, 256, 3)
-Calc FID for CFG 3.0 and denoise_timesteps 1
-DiT: Input of shape (512, 32, 32, 4) dtype float32
-DiT: After patch embed, shape is (512, 256, 768) dtype bfloat16
-DiT: Patch Embed of shape (512, 256, 768) dtype bfloat16
-DiT: Conditioning of shape (512, 768) dtype float32
-FID is 232.8525390625
+FID is 9.344593048095703