Update app.py

app.py

@@ -101,13 +101,13 @@ def inference( prompt, negative_prompt, guidance_scale, ddim_steps, seed):
 
     text_input = tokenizer.value(prompt, padding="max_length", max_length=tokenizer.value.model_max_length, truncation=True, return_tensors="pt")
 
-    text_embeddings = text_encoder.value(text_input.input_ids.to(device))[0]
+    text_embeddings = text_encoder.value(text_input.input_ids.to(device.value))[0]
 
     max_length = text_input.input_ids.shape[-1]
     uncond_input = tokenizer.value(
                             [negative_prompt], padding="max_length", max_length=max_length, return_tensors="pt"
                         )
-    uncond_embeddings = text_encoder.value(uncond_input.input_ids.to(device))[0]
+    uncond_embeddings = text_encoder.value(uncond_input.input_ids.to(device.value))[0]
     text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
     noise_scheduler.value.set_timesteps(ddim_steps) 
     latents = latents * noise_scheduler.value.init_noise_sigma
@@ -141,7 +141,7 @@ def edit_inference(prompt, negative_prompt, guidance_scale, ddim_steps, seed, st
     #pad to same number of PCs
     pcs_original = original_weights.shape[1]
     pcs_edits = young.value.shape[1]
-    padding =  torch.zeros((1,pcs_original-pcs_edits)).to(device)
+    padding =  torch.zeros((1,pcs_original-pcs_edits)).to(device.value)
     young_pad = torch.cat((young.value, padding), 1)
     pointy_pad = torch.cat((pointy.value, padding), 1)
     wavy_pad = torch.cat((wavy.value, padding), 1)
@@ -160,13 +160,13 @@ def edit_inference(prompt, negative_prompt, guidance_scale, ddim_steps, seed, st
 
     text_input = tokenizer.value(prompt, padding="max_length", max_length=tokenizer.value.model_max_length, truncation=True, return_tensors="pt")
 
-    text_embeddings = text_encoder.value(text_input.input_ids.to(device))[0]
+    text_embeddings = text_encoder.value(text_input.input_ids.to(device.value))[0]
 
     max_length = text_input.input_ids.shape[-1]
     uncond_input = tokenizer.value(
                             [negative_prompt], padding="max_length", max_length=max_length, return_tensors="pt"
                         )
-    uncond_embeddings = text_encoder.value(uncond_input.input_ids.to(device))[0]
+    uncond_embeddings = text_encoder.value(uncond_input.input_ids.to(device.value))[0]
     text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
     noise_scheduler.value.set_timesteps(ddim_steps) 
     latents = latents * noise_scheduler.value.init_noise_sigma