Hugging Face's logo

Spaces:
1inkusFace
/
RealVis_v5.0_BF16_F
Running on Zero

App Files Community
1inkusFace commited on
Commit
d358d91
·
verified ·
1 Parent(s): fd9365d

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +19 -22
app.py CHANGED
@@ -237,28 +237,27 @@ def generate_30(
237
 
238
  # 2. Encode with the two text encoders
239
  prompt_embeds_a = pipe.text_encoder(text_input_ids1.to(torch.device('cuda')), output_hidden_states=True)
240
- pooled_prompt_embeds_a = prompt_embeds_a[0] # Pooled output from encoder 1
241
  print('pooled shape 1: ', pooled_prompt_embeds_a.shape)
242
  prompt_embeds_a = prompt_embeds_a.hidden_states[-2] # Penultimate hidden state from encoder 1
243
  print('encoder shape: ', prompt_embeds_a.shape)
244
  prompt_embeds_b = pipe.text_encoder(text_input_ids2.to(torch.device('cuda')), output_hidden_states=True)
245
- pooled_prompt_embeds_b = prompt_embeds_b[0] # Pooled output from encoder 2
246
- prompt_embeds_b = prompt_embeds_b.hidden_states[-2] # Penultimate hidden state from encoder 2
247
 
248
  prompt_embeds_a2 = pipe.text_encoder_2(text_input_ids1b.to(torch.device('cuda')), output_hidden_states=True)
249
- pooled_prompt_embeds_a2 = prompt_embeds_a2[0] # Pooled output from encoder 1
250
  print('pooled shape 2: ', pooled_prompt_embeds_a2.shape)
251
- prompt_embeds_a2 = prompt_embeds_a2.hidden_states[-2] # Penultimate hidden state from encoder 1
252
  print('encoder shape2: ', prompt_embeds_a2.shape)
253
  prompt_embeds_b2 = pipe.text_encoder_2(text_input_ids2b.to(torch.device('cuda')), output_hidden_states=True)
254
  pooled_prompt_embeds_b2 = prompt_embeds_b2[0] # Pooled output from encoder 2
255
  prompt_embeds_b2 = prompt_embeds_b2.hidden_states[-2] # Penultimate hidden state from encoder 2
256
 
257
-
258
  # 3. Concatenate the embeddings
259
  prompt_embeds = torch.cat([prompt_embeds_a, prompt_embeds_b])
260
  print('catted shape: ', prompt_embeds.shape)
261
- pooled_prompt_embeds = torch.cat([pooled_prompt_embeds_a, pooled_prompt_embeds_a])
262
  print('catted pooled shape: ', pooled_prompt_embeds.shape)
263
  pooled_prompt_embeds = torch.mean(pooled_prompt_embeds,dim=0,keepdim=True)
264
  print('meaned pooled shape: ', pooled_prompt_embeds.shape)
@@ -274,7 +273,7 @@ def generate_30(
274
  print('catted pooled shape 2: ', pooled_prompt_embeds2.shape)
275
  pooled_prompt_embeds2 = torch.mean(pooled_prompt_embeds2,dim=0,keepdim=True)
276
  print('pooled meaned shape 2: ', pooled_prompt_embeds2.shape)
277
- pooled_prompt_embeds = torch.cat([pooled_prompt_embeds, pooled_prompt_embeds2],dim=0)
278
  print('catted combined meaned pooled shape: ', pooled_prompt_embeds.shape)
279
 
280
  options = {
@@ -367,24 +366,23 @@ def generate_60(
367
 
368
  # 2. Encode with the two text encoders
369
  prompt_embeds_a = pipe.text_encoder(text_input_ids1.to(torch.device('cuda')), output_hidden_states=True)
370
- pooled_prompt_embeds_a = prompt_embeds_a[0] # Pooled output from encoder 1
371
  print('pooled shape 1: ', pooled_prompt_embeds_a.shape)
372
  prompt_embeds_a = prompt_embeds_a.hidden_states[-2] # Penultimate hidden state from encoder 1
373
  print('encoder shape: ', prompt_embeds_a.shape)
374
  prompt_embeds_b = pipe.text_encoder(text_input_ids2.to(torch.device('cuda')), output_hidden_states=True)
375
- pooled_prompt_embeds_b = prompt_embeds_b[0] # Pooled output from encoder 2
376
- prompt_embeds_b = prompt_embeds_b.hidden_states[-2] # Penultimate hidden state from encoder 2
377
 
378
  prompt_embeds_a2 = pipe.text_encoder_2(text_input_ids1b.to(torch.device('cuda')), output_hidden_states=True)
379
- pooled_prompt_embeds_a2 = prompt_embeds_a2[0] # Pooled output from encoder 1
380
  print('pooled shape 2: ', pooled_prompt_embeds_a2.shape)
381
- prompt_embeds_a2 = prompt_embeds_a2.hidden_states[-2] # Penultimate hidden state from encoder 1
382
  print('encoder shape2: ', prompt_embeds_a2.shape)
383
  prompt_embeds_b2 = pipe.text_encoder_2(text_input_ids2b.to(torch.device('cuda')), output_hidden_states=True)
384
  pooled_prompt_embeds_b2 = prompt_embeds_b2[0] # Pooled output from encoder 2
385
  prompt_embeds_b2 = prompt_embeds_b2.hidden_states[-2] # Penultimate hidden state from encoder 2
386
 
387
-
388
  # 3. Concatenate the embeddings
389
  prompt_embeds = torch.cat([prompt_embeds_a, prompt_embeds_b])
390
  print('catted shape: ', prompt_embeds.shape)
@@ -404,7 +402,7 @@ def generate_60(
404
  print('catted pooled shape 2: ', pooled_prompt_embeds2.shape)
405
  pooled_prompt_embeds2 = torch.mean(pooled_prompt_embeds2,dim=0,keepdim=True)
406
  print('pooled meaned shape 2: ', pooled_prompt_embeds2.shape)
407
- pooled_prompt_embeds = torch.cat([pooled_prompt_embeds, pooled_prompt_embeds2],dim=0)
408
  print('catted combined meaned pooled shape: ', pooled_prompt_embeds.shape)
409
 
410
  options = {
@@ -497,24 +495,23 @@ def generate_90(
497
 
498
  # 2. Encode with the two text encoders
499
  prompt_embeds_a = pipe.text_encoder(text_input_ids1.to(torch.device('cuda')), output_hidden_states=True)
500
- pooled_prompt_embeds_a = prompt_embeds_a[0] # Pooled output from encoder 1
501
  print('pooled shape 1: ', pooled_prompt_embeds_a.shape)
502
  prompt_embeds_a = prompt_embeds_a.hidden_states[-2] # Penultimate hidden state from encoder 1
503
  print('encoder shape: ', prompt_embeds_a.shape)
504
  prompt_embeds_b = pipe.text_encoder(text_input_ids2.to(torch.device('cuda')), output_hidden_states=True)
505
- pooled_prompt_embeds_b = prompt_embeds_b[0] # Pooled output from encoder 2
506
- prompt_embeds_b = prompt_embeds_b.hidden_states[-2] # Penultimate hidden state from encoder 2
507
 
508
  prompt_embeds_a2 = pipe.text_encoder_2(text_input_ids1b.to(torch.device('cuda')), output_hidden_states=True)
509
- pooled_prompt_embeds_a2 = prompt_embeds_a2[0] # Pooled output from encoder 1
510
  print('pooled shape 2: ', pooled_prompt_embeds_a2.shape)
511
- prompt_embeds_a2 = prompt_embeds_a2.hidden_states[-2] # Penultimate hidden state from encoder 1
512
  print('encoder shape2: ', prompt_embeds_a2.shape)
513
  prompt_embeds_b2 = pipe.text_encoder_2(text_input_ids2b.to(torch.device('cuda')), output_hidden_states=True)
514
  pooled_prompt_embeds_b2 = prompt_embeds_b2[0] # Pooled output from encoder 2
515
  prompt_embeds_b2 = prompt_embeds_b2.hidden_states[-2] # Penultimate hidden state from encoder 2
516
 
517
-
518
  # 3. Concatenate the embeddings
519
  prompt_embeds = torch.cat([prompt_embeds_a, prompt_embeds_b])
520
  print('catted shape: ', prompt_embeds.shape)
@@ -534,7 +531,7 @@ def generate_90(
534
  print('catted pooled shape 2: ', pooled_prompt_embeds2.shape)
535
  pooled_prompt_embeds2 = torch.mean(pooled_prompt_embeds2,dim=0,keepdim=True)
536
  print('pooled meaned shape 2: ', pooled_prompt_embeds2.shape)
537
- pooled_prompt_embeds = torch.cat([pooled_prompt_embeds, pooled_prompt_embeds2],dim=0)
538
  print('catted combined meaned pooled shape: ', pooled_prompt_embeds.shape)
539
 
540
  options = {
 
237
 
238
  # 2. Encode with the two text encoders
239
  prompt_embeds_a = pipe.text_encoder(text_input_ids1.to(torch.device('cuda')), output_hidden_states=True)
240
+ pooled_prompt_embeds_a = prompt_embeds_a[0][:, -1, :] # Pooled output from encoder 1
241
  print('pooled shape 1: ', pooled_prompt_embeds_a.shape)
242
  prompt_embeds_a = prompt_embeds_a.hidden_states[-2] # Penultimate hidden state from encoder 1
243
  print('encoder shape: ', prompt_embeds_a.shape)
244
  prompt_embeds_b = pipe.text_encoder(text_input_ids2.to(torch.device('cuda')), output_hidden_states=True)
245
+ pooled_prompt_embeds_b = prompt_embeds_b[0][:, -1, :] # Pooled output from encoder 1
246
+ prompt_embeds_b = prompt_embeds_b.hidden_states[-2] # Penultimate hidden state from encoder 1
247
 
248
  prompt_embeds_a2 = pipe.text_encoder_2(text_input_ids1b.to(torch.device('cuda')), output_hidden_states=True)
249
+ pooled_prompt_embeds_a2 = prompt_embeds_a2[0] # Pooled output from encoder 2
250
  print('pooled shape 2: ', pooled_prompt_embeds_a2.shape)
251
+ prompt_embeds_a2 = prompt_embeds_a2.hidden_states[-2] # Penultimate hidden state from encoder 2
252
  print('encoder shape2: ', prompt_embeds_a2.shape)
253
  prompt_embeds_b2 = pipe.text_encoder_2(text_input_ids2b.to(torch.device('cuda')), output_hidden_states=True)
254
  pooled_prompt_embeds_b2 = prompt_embeds_b2[0] # Pooled output from encoder 2
255
  prompt_embeds_b2 = prompt_embeds_b2.hidden_states[-2] # Penultimate hidden state from encoder 2
256
 
 
257
  # 3. Concatenate the embeddings
258
  prompt_embeds = torch.cat([prompt_embeds_a, prompt_embeds_b])
259
  print('catted shape: ', prompt_embeds.shape)
260
+ pooled_prompt_embeds = torch.cat([pooled_prompt_embeds_a, pooled_prompt_embeds_b])
261
  print('catted pooled shape: ', pooled_prompt_embeds.shape)
262
  pooled_prompt_embeds = torch.mean(pooled_prompt_embeds,dim=0,keepdim=True)
263
  print('meaned pooled shape: ', pooled_prompt_embeds.shape)
 
273
  print('catted pooled shape 2: ', pooled_prompt_embeds2.shape)
274
  pooled_prompt_embeds2 = torch.mean(pooled_prompt_embeds2,dim=0,keepdim=True)
275
  print('pooled meaned shape 2: ', pooled_prompt_embeds2.shape)
276
+ pooled_prompt_embeds = torch.cat([pooled_prompt_embeds, pooled_prompt_embeds2],dim=1)
277
  print('catted combined meaned pooled shape: ', pooled_prompt_embeds.shape)
278
 
279
  options = {
 
366
 
367
  # 2. Encode with the two text encoders
368
  prompt_embeds_a = pipe.text_encoder(text_input_ids1.to(torch.device('cuda')), output_hidden_states=True)
369
+ pooled_prompt_embeds_a = prompt_embeds_a[0][:, -1, :] # Pooled output from encoder 1
370
  print('pooled shape 1: ', pooled_prompt_embeds_a.shape)
371
  prompt_embeds_a = prompt_embeds_a.hidden_states[-2] # Penultimate hidden state from encoder 1
372
  print('encoder shape: ', prompt_embeds_a.shape)
373
  prompt_embeds_b = pipe.text_encoder(text_input_ids2.to(torch.device('cuda')), output_hidden_states=True)
374
+ pooled_prompt_embeds_b = prompt_embeds_b[0][:, -1, :] # Pooled output from encoder 1
375
+ prompt_embeds_b = prompt_embeds_b.hidden_states[-2] # Penultimate hidden state from encoder 1
376
 
377
  prompt_embeds_a2 = pipe.text_encoder_2(text_input_ids1b.to(torch.device('cuda')), output_hidden_states=True)
378
+ pooled_prompt_embeds_a2 = prompt_embeds_a2[0] # Pooled output from encoder 2
379
  print('pooled shape 2: ', pooled_prompt_embeds_a2.shape)
380
+ prompt_embeds_a2 = prompt_embeds_a2.hidden_states[-2] # Penultimate hidden state from encoder 2
381
  print('encoder shape2: ', prompt_embeds_a2.shape)
382
  prompt_embeds_b2 = pipe.text_encoder_2(text_input_ids2b.to(torch.device('cuda')), output_hidden_states=True)
383
  pooled_prompt_embeds_b2 = prompt_embeds_b2[0] # Pooled output from encoder 2
384
  prompt_embeds_b2 = prompt_embeds_b2.hidden_states[-2] # Penultimate hidden state from encoder 2
385
 
 
386
  # 3. Concatenate the embeddings
387
  prompt_embeds = torch.cat([prompt_embeds_a, prompt_embeds_b])
388
  print('catted shape: ', prompt_embeds.shape)
 
402
  print('catted pooled shape 2: ', pooled_prompt_embeds2.shape)
403
  pooled_prompt_embeds2 = torch.mean(pooled_prompt_embeds2,dim=0,keepdim=True)
404
  print('pooled meaned shape 2: ', pooled_prompt_embeds2.shape)
405
+ pooled_prompt_embeds = torch.cat([pooled_prompt_embeds, pooled_prompt_embeds2],dim=1)
406
  print('catted combined meaned pooled shape: ', pooled_prompt_embeds.shape)
407
 
408
  options = {
 
495
 
496
  # 2. Encode with the two text encoders
497
  prompt_embeds_a = pipe.text_encoder(text_input_ids1.to(torch.device('cuda')), output_hidden_states=True)
498
+ pooled_prompt_embeds_a = prompt_embeds_a[0][:, -1, :] # Pooled output from encoder 1
499
  print('pooled shape 1: ', pooled_prompt_embeds_a.shape)
500
  prompt_embeds_a = prompt_embeds_a.hidden_states[-2] # Penultimate hidden state from encoder 1
501
  print('encoder shape: ', prompt_embeds_a.shape)
502
  prompt_embeds_b = pipe.text_encoder(text_input_ids2.to(torch.device('cuda')), output_hidden_states=True)
503
+ pooled_prompt_embeds_b = prompt_embeds_b[0][:, -1, :] # Pooled output from encoder 1
504
+ prompt_embeds_b = prompt_embeds_b.hidden_states[-2] # Penultimate hidden state from encoder 1
505
 
506
  prompt_embeds_a2 = pipe.text_encoder_2(text_input_ids1b.to(torch.device('cuda')), output_hidden_states=True)
507
+ pooled_prompt_embeds_a2 = prompt_embeds_a2[0] # Pooled output from encoder 2
508
  print('pooled shape 2: ', pooled_prompt_embeds_a2.shape)
509
+ prompt_embeds_a2 = prompt_embeds_a2.hidden_states[-2] # Penultimate hidden state from encoder 2
510
  print('encoder shape2: ', prompt_embeds_a2.shape)
511
  prompt_embeds_b2 = pipe.text_encoder_2(text_input_ids2b.to(torch.device('cuda')), output_hidden_states=True)
512
  pooled_prompt_embeds_b2 = prompt_embeds_b2[0] # Pooled output from encoder 2
513
  prompt_embeds_b2 = prompt_embeds_b2.hidden_states[-2] # Penultimate hidden state from encoder 2
514
 
 
515
  # 3. Concatenate the embeddings
516
  prompt_embeds = torch.cat([prompt_embeds_a, prompt_embeds_b])
517
  print('catted shape: ', prompt_embeds.shape)
 
531
  print('catted pooled shape 2: ', pooled_prompt_embeds2.shape)
532
  pooled_prompt_embeds2 = torch.mean(pooled_prompt_embeds2,dim=0,keepdim=True)
533
  print('pooled meaned shape 2: ', pooled_prompt_embeds2.shape)
534
+ pooled_prompt_embeds = torch.cat([pooled_prompt_embeds, pooled_prompt_embeds2],dim=1)
535
  print('catted combined meaned pooled shape: ', pooled_prompt_embeds.shape)
536
 
537
  options = {