reqs

app.py

@@ -134,7 +134,7 @@ pali = PaliGemmaForConditionalGeneration.from_pretrained('google/paligemma-3b-pt
 processor = AutoProcessor.from_pretrained('google/paligemma-3b-pt-224')
 
 
-
+@spaces.GPU()
 def to_wanted_embs(image_outputs, input_ids, attention_mask, cache_position=None):
     inputs_embeds = pali.get_input_embeddings()(input_ids)
     selected_image_feature = image_outputs.to(dtype).to(device)
@@ -148,7 +148,7 @@ def to_wanted_embs(image_outputs, input_ids, attention_mask, cache_position=None
     return inputs_embeds
     
 
-
+@spaces.GPU()
 def generate_pali(user_emb):
     prompt = 'caption en'
     model_inputs = processor(text=prompt, images=torch.zeros(1, 3, 224, 224), return_tensors="pt")
@@ -540,8 +540,7 @@ def encode_space(x):
     im_emb, _ = pipe.encode_image(
                 image, DEVICE, 1, output_hidden_state
             )
-            
-            
+
     im = torchvision.transforms.ToTensor()(x).unsqueeze(0)
     im = torch.nn.functional.interpolate(im, (224, 224))
     im = (im - .5) * 2