Update app.py

app.py

@@ -10,12 +10,12 @@ from transformers import (
 )
 from ppt_parser import transfer_to_structure
 
-# ✅ Load Hugging Face token
+# ✅ Load Hugging Face token and model ID
 hf_token = os.getenv("HF_TOKEN")
 model_id = "meta-llama/Llama-4-Scout-17B-16E-Instruct"
 
-# ✅ Load image processor, tokenizer, and model manually
-image_processor = BlipImageProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
+# ✅ Load processor, tokenizer, and model
+processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
 tokenizer = AutoTokenizer.from_pretrained(model_id, token=hf_token)
 
 model = Llama4ForConditionalGeneration.from_pretrained(
@@ -30,6 +30,7 @@ model = Llama4ForConditionalGeneration.from_pretrained(
 extracted_text = ""
 image_paths = []
 
+# ✅ Extract all text from parsed PPTX JSON
 def extract_text_from_pptx_json(parsed_json: dict) -> str:
     text = ""
     for slide in parsed_json.values():
@@ -46,6 +47,7 @@ def extract_text_from_pptx_json(parsed_json: dict) -> str:
                         text += para.get("text", "") + "\n"
     return text.strip()
 
+# ✅ Handle uploaded .pptx files
 def handle_pptx_upload(pptx_file):
     global extracted_text, image_paths
     tmp_path = pptx_file.name
@@ -54,16 +56,18 @@ def handle_pptx_upload(pptx_file):
     extracted_text = extract_text_from_pptx_json(parsed_json)
     return extracted_text or "No readable text found in slides."
 
+# ✅ Ask Llama 4 Scout using image + text
 def ask_llama(question):
     global extracted_text, image_paths
 
     if not extracted_text and not image_paths:
         return "Please upload and extract a PPTX file first."
 
-    # ✅ Use the first image only (you can expand to multiple with batching)
+    # Use the first slide image
     image = Image.open(image_paths[0]).convert("RGB")
-    vision_inputs = image_processor(images=image, return_tensors="pt").to(model.device)
+    vision_inputs = processor(images=image, return_tensors="pt").to(model.device)
 
+    # Prepare text prompt
     prompt = f"<|user|>\n{extracted_text}\n\nQuestion: {question}<|end|>\n<|assistant|>\n"
     text_inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
 
@@ -74,7 +78,11 @@ def ask_llama(question):
             max_new_tokens=256,
         )
 
-    response = tokenizer.decode(output[0][text_inputs["input_ids"].shape[-1]:], skip_special_tokens=True)
+    # Decode response
+    response = tokenizer.decode(
+        output[0][text_inputs["input_ids"].shape[-1]:],
+        skip_special_tokens=True
+    )
     return response.strip()
 
 # ✅ Gradio UI