Upload folder using huggingface_hub

1e-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 13345880 seconds ago.
+Loaded checkpoint from 14090812 seconds ago.
 Loaded model with step 474001
 ┌──────────────────────────────────────────────────────────────────────────────┐
 │                                    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 5202 seconds ago.
+Loaded checkpoint from 750135 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 36.60750961303711
-(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 37.717193603515625
-(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 40.62632369995117
-(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 48.93124771118164
-(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 73.71484375
-(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 158.0609130859375
-(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 313.73358154296875
-(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 286.5404357910156
+FID is 36.649375915527344
 (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 22.644939422607422
-(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 23.565139770507812
-(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 25.94041633605957
-(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 32.80158996582031
-(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 55.220787048339844
-(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 133.41140747070312
-(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 302.35345458984375
-(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 275.65960693359375
+FID is 22.582571029663086
 (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 14.374984741210938
-(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 15.053552627563477
-(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 16.86142921447754
-(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 22.118249893188477
-(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 40.5958137512207
-(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 111.90969848632812
-(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 291.7471923828125
-(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 268.52325439453125
+FID is 14.302921295166016
 (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 9.992362976074219
-(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 10.469978332519531
-(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 11.733261108398438
-(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 15.620132446289062
-(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 30.157955169677734
-(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 93.65843200683594
-(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 282.39471435546875
-(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 263.8368835449219
+FID is 9.92587661743164
 (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 8.087844848632812
-(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 8.410569190979004
-(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 9.256352424621582
-(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 11.975582122802734
-(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 23.16156005859375
-(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 78.73834228515625
-(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 274.115966796875
-(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 260.2587585449219
+FID is 8.037654876708984
 (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 7.667808532714844
-(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 7.854188919067383
-(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 8.360774993896484
-(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 10.235084533691406
-(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 18.630203247070312
-(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 66.66618347167969
-(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 266.68902587890625
-(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 257.45001220703125
+FID is 7.633633613586426
 (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 8.030121803283691
-(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 8.140653610229492
-(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 8.436915397644043
-(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 9.632453918457031
-(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 15.819036483764648
-(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 57.212120056152344
-(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 260.07421875
-(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 255.27415466308594
+FID is 8.01058292388916
 (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 8.7904691696167
-(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 8.839826583862305
-(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 8.988865852355957
-(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 9.728584289550781
-(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 14.18990707397461
-(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 49.663368225097656
-(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 253.92054748535156
-(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 253.68734741210938
+FID is 8.780667304992676
 (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 9.772045135498047
-(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 9.767073631286621
-(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 9.813874244689941
-(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 10.203750610351562
-(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 13.351149559020996
-(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 43.611480712890625
-(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 248.34854125976562
-(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 252.22958374023438
+FID is 9.77159309387207