Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -301,17 +301,19 @@ def generate_60(
|
|
| 301 |
prompt_embeds_a = pipe.text_encoder(text_input_ids1.to(torch.device('cuda')), output_hidden_states=True)
|
| 302 |
pooled_prompt_embeds_a = prompt_embeds_a[0] # Pooled output from encoder 1
|
| 303 |
prompt_embeds_a = prompt_embeds_a.hidden_states[-2] # Penultimate hidden state from encoder 1
|
| 304 |
-
|
| 305 |
prompt_embeds_b = pipe.text_encoder(text_input_ids2.to(torch.device('cuda')), output_hidden_states=True)
|
| 306 |
pooled_prompt_embeds_b = prompt_embeds_b[0] # Pooled output from encoder 2
|
| 307 |
prompt_embeds_b = prompt_embeds_b.hidden_states[-2] # Penultimate hidden state from encoder 2
|
| 308 |
|
| 309 |
# 3. Concatenate the embeddings along the sequence dimension (dim=1)
|
| 310 |
prompt_embeds = torch.cat([prompt_embeds_a, prompt_embeds_b], dim=1)
|
|
|
|
| 311 |
pooled_prompt_embeds = torch.cat([pooled_prompt_embeds_a, pooled_prompt_embeds_b], dim=1)
|
| 312 |
-
|
| 313 |
# 4. (Optional) Average the pooled embeddings
|
| 314 |
prompt_embeds = prompt_embeds.mean(dim=1, keepdim=True)
|
|
|
|
| 315 |
pooled_prompt_embeds = pooled_prompt_embeds.mean(dim=1, keepdim=True)
|
| 316 |
|
| 317 |
|
|
@@ -384,17 +386,19 @@ def generate_90(
|
|
| 384 |
prompt_embeds_a = pipe.text_encoder(text_input_ids1.to(torch.device('cuda')), output_hidden_states=True)
|
| 385 |
pooled_prompt_embeds_a = prompt_embeds_a[0] # Pooled output from encoder 1
|
| 386 |
prompt_embeds_a = prompt_embeds_a.hidden_states[-2] # Penultimate hidden state from encoder 1
|
| 387 |
-
|
| 388 |
prompt_embeds_b = pipe.text_encoder(text_input_ids2.to(torch.device('cuda')), output_hidden_states=True)
|
| 389 |
pooled_prompt_embeds_b = prompt_embeds_b[0] # Pooled output from encoder 2
|
| 390 |
prompt_embeds_b = prompt_embeds_b.hidden_states[-2] # Penultimate hidden state from encoder 2
|
| 391 |
|
| 392 |
# 3. Concatenate the embeddings along the sequence dimension (dim=1)
|
| 393 |
prompt_embeds = torch.cat([prompt_embeds_a, prompt_embeds_b], dim=1)
|
|
|
|
| 394 |
pooled_prompt_embeds = torch.cat([pooled_prompt_embeds_a, pooled_prompt_embeds_b], dim=1)
|
| 395 |
-
|
| 396 |
# 4. (Optional) Average the pooled embeddings
|
| 397 |
prompt_embeds = prompt_embeds.mean(dim=1, keepdim=True)
|
|
|
|
| 398 |
pooled_prompt_embeds = pooled_prompt_embeds.mean(dim=1, keepdim=True)
|
| 399 |
|
| 400 |
|
|
|
|
| 301 |
prompt_embeds_a = pipe.text_encoder(text_input_ids1.to(torch.device('cuda')), output_hidden_states=True)
|
| 302 |
pooled_prompt_embeds_a = prompt_embeds_a[0] # Pooled output from encoder 1
|
| 303 |
prompt_embeds_a = prompt_embeds_a.hidden_states[-2] # Penultimate hidden state from encoder 1
|
| 304 |
+
print('encoder shape: ', prompt_embeds_a.shape)
|
| 305 |
prompt_embeds_b = pipe.text_encoder(text_input_ids2.to(torch.device('cuda')), output_hidden_states=True)
|
| 306 |
pooled_prompt_embeds_b = prompt_embeds_b[0] # Pooled output from encoder 2
|
| 307 |
prompt_embeds_b = prompt_embeds_b.hidden_states[-2] # Penultimate hidden state from encoder 2
|
| 308 |
|
| 309 |
# 3. Concatenate the embeddings along the sequence dimension (dim=1)
|
| 310 |
prompt_embeds = torch.cat([prompt_embeds_a, prompt_embeds_b], dim=1)
|
| 311 |
+
print('catted shape: ', prompt_embeds.shape)
|
| 312 |
pooled_prompt_embeds = torch.cat([pooled_prompt_embeds_a, pooled_prompt_embeds_b], dim=1)
|
| 313 |
+
|
| 314 |
# 4. (Optional) Average the pooled embeddings
|
| 315 |
prompt_embeds = prompt_embeds.mean(dim=1, keepdim=True)
|
| 316 |
+
print('averaged shape: ', prompt_embeds.shape)
|
| 317 |
pooled_prompt_embeds = pooled_prompt_embeds.mean(dim=1, keepdim=True)
|
| 318 |
|
| 319 |
|
|
|
|
| 386 |
prompt_embeds_a = pipe.text_encoder(text_input_ids1.to(torch.device('cuda')), output_hidden_states=True)
|
| 387 |
pooled_prompt_embeds_a = prompt_embeds_a[0] # Pooled output from encoder 1
|
| 388 |
prompt_embeds_a = prompt_embeds_a.hidden_states[-2] # Penultimate hidden state from encoder 1
|
| 389 |
+
print('encoder shape: ', prompt_embeds_a.shape)
|
| 390 |
prompt_embeds_b = pipe.text_encoder(text_input_ids2.to(torch.device('cuda')), output_hidden_states=True)
|
| 391 |
pooled_prompt_embeds_b = prompt_embeds_b[0] # Pooled output from encoder 2
|
| 392 |
prompt_embeds_b = prompt_embeds_b.hidden_states[-2] # Penultimate hidden state from encoder 2
|
| 393 |
|
| 394 |
# 3. Concatenate the embeddings along the sequence dimension (dim=1)
|
| 395 |
prompt_embeds = torch.cat([prompt_embeds_a, prompt_embeds_b], dim=1)
|
| 396 |
+
print('catted shape: ', prompt_embeds.shape)
|
| 397 |
pooled_prompt_embeds = torch.cat([pooled_prompt_embeds_a, pooled_prompt_embeds_b], dim=1)
|
| 398 |
+
|
| 399 |
# 4. (Optional) Average the pooled embeddings
|
| 400 |
prompt_embeds = prompt_embeds.mean(dim=1, keepdim=True)
|
| 401 |
+
print('averaged shape: ', prompt_embeds.shape)
|
| 402 |
pooled_prompt_embeds = pooled_prompt_embeds.mean(dim=1, keepdim=True)
|
| 403 |
|
| 404 |
|