Update app.py

app.py

@@ -1,11 +1,10 @@
 import gradio as gr
 from transformers import PaliGemmaForConditionalGeneration, PaliGemmaProcessor
-from PIL import Image
 import spaces
 import torch
 import re
 
-model = PaliGemmaForConditionalGeneration.from_pretrained("gokaygokay/sd3-long-captioner").to("cpu").eval()
+model = PaliGemmaForConditionalGeneration.from_pretrained("gokaygokay/sd3-long-captioner").to("cuda").eval()
 processor = PaliGemmaProcessor.from_pretrained("gokaygokay/sd3-long-captioner")
 
 def modify_caption(caption: str) -> str:
@@ -26,55 +25,46 @@ def modify_caption(caption: str) -> str:
     
     def replace_fn(match):
         return replacers[match.group(0)]
-    
+
     return re.sub(pattern, replace_fn, caption, count=1, flags=re.IGNORECASE)
 
+@spaces.GPU
 def create_captions_rich(images):
-    # Debugging: Print out the type of 'images'
-    print(f"Type of 'images': {type(images)}")
-    if isinstance(images, tuple):
-        print("Received a tuple, expected a file-like object.")
-        # If it's a tuple, you can try accessing the first element as an example
-        print(f"Type of 'images[0]': {type(images[0])}")
-
     captions = []
-    for image_path in images:
-        try:
-            # If 'images' is a tuple, you might need to modify this part to extract the image file correctly
-            image = Image.open(image_path).convert("RGB")
-            prompt = "caption en"
-            model_inputs = processor(text=prompt, images=image, return_tensors="pt").to("cpu")
-            input_len = model_inputs["input_ids"].shape[-1]
-
-            with torch.inference_mode():
-                generation = model.generate(**model_inputs, max_new_tokens=256, do_sample=False)
-                generation = generation[0][input_len:]
-                decoded = processor.decode(generation, skip_special_tokens=True)
+    prompt = "caption en"
+    
+    for image in images:
+        model_inputs = processor(text=prompt, images=image, return_tensors="pt").to("cuda")
+        input_len = model_inputs["input_ids"].shape[-1]
+        
+        with torch.inference_mode():
+            generation = model.generate(**model_inputs, max_new_tokens=256, do_sample=False)
+            generation = generation[0][input_len:]
+            decoded = processor.decode(generation, skip_special_tokens=True)
 
-                modified_caption = modify_caption(decoded)
-                captions.append(modified_caption)
-        except Exception as e:
-            captions.append(f"Error processing image: {e}")
+            modified_caption = modify_caption(decoded)
+            captions.append(modified_caption)
+    
     return captions
 
-
 css = """
   #mkd {
     height: 500px; 
     overflow: auto; 
-    border: 8px solid #ccc; 
+    border: 16px solid #ccc; 
   }
 """
 
 with gr.Blocks(css=css) as demo:
-  gr.HTML("<h1><center>Finetuned PaliGemma for SD3 prompt generation.<center><h1>")
-  with gr.Tab(label="Image to Prompt for SD3"):
-    with gr.Row():
-      with gr.Column():
-        input_img = gr.Gallery(label="Input Images", type="pil", interactive=True)
-        submit_btn = gr.Button(value="Start")
-      output = gr.Textbox(label="Prompt", lines=10, interactive=True)
+    gr.HTML("<h1><center>Fine-tuned PaliGemma for SD3 Image Guided Prompt Generation.<center><h1>")
+    
+    with gr.Tab(label="Image to Prompt for SD3."):
+        with gr.Row():
+            with gr.Column():
+                input_imgs = gr.Image(label="Input Images", type="pil", tool="editor", interactive=True, multiple=True)
+                submit_btn = gr.Button(value="Start")
+            outputs = gr.Text(label="Prompts", interactive=False)
 
-    submit_btn.click(create_captions_rich, [input_img], [output])
+        submit_btn.click(create_captions_rich, [input_imgs], [outputs])
 
-demo.launch(debug=True) 
+demo.launch(debug=True)