Update app.py

app.py

@@ -102,8 +102,10 @@ Your task:
 Now produce the final report only, without reiterating these instructions or the query."""
 
     messages = [
-        {"role": "system", "content": system_message},
-        {"role": "user", "content": user_content}
+        [
+            {"role": "system", "content": [{"type": "text", "text": system_message},]},
+            {"role": "user", "content": [{"type": "text", "text": user_content},]},
+        ],
     ]
 
     return messages
@@ -263,30 +265,15 @@ def main():
             else:
                 sampled_docs = scored_docs
             
-            prompt = [
-                {"role": "user", "content": f"""
-Generate a well-structured business report based on tweets from twitter/X with sentiment that answers Query Question and meets following Requirements. 
-**Requirements:**
-- Include an introduction, key insights, and a conclusion.
-- Ensure the analysis is concise and does not cut off abruptly. 
-- Summarize major findings without repeating verbatim.
-- Cover both positive and negative aspects, highlighting trends in user sentiment.
-**Query Question:**
-"{query_input}"
-**Tweets with sentiment score:**
-{sampled_docs}
-Please ensure the report is complete and reaches approximately 300 words.
-                """}
-            ]
+            prompt = build_prompt(query_input, sampled_docs)
     
             def process_with_gemma(prompt):
                 try:
-                    # tokenizer = AutoTokenizer.from_pretrained("unsloth/gemma-3-1b-it")
                     pipe = pipeline(
                         "text-generation",
                         model="unsloth/gemma-3-1b-it",
-                        device=0 if torch.cuda.is_available() else -1,
-                        # tokenizer=tokenizer,
+                        device="cuda" if torch.cuda.is_available() else -1,
+                        torch_dtype=torch.bfloat16,
                     )
                     result = pipe(prompt, max_new_tokens=256, repetition_penalty=1.2, do_sample=True, temperature=0.5, return_full_text=False)
                     return result, None