Update app.py

app.py

@@ -36,6 +36,16 @@ tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForCausalLM.from_pretrained(model_name)
 tokenizer.pad_token = tokenizer.eos_token
 
+# Connect to Qdrant + embedding
+embedding = HuggingFaceEmbeddings(model_name="Omartificial-Intelligence-Space/GATE-AraBert-v1")
+qdrant_client = QdrantClient(url=QDRANT_URL, api_key=QDRANT_API_KEY)
+
+vector_store = Qdrant(
+    client=qdrant_client,
+    collection_name=COLLECTION_NAME,
+    embeddings=embedding
+)
+
 # Generation settings
 generation_config = GenerationConfig(
     max_new_tokens=150,
@@ -56,16 +66,6 @@ llm_pipeline = pipeline(
 )
 llm = HuggingFacePipeline(pipeline=llm_pipeline)
 
-# Connect to Qdrant + embedding
-embedding = HuggingFaceEmbeddings(model_name="Omartificial-Intelligence-Space/GATE-AraBert-v1")
-qdrant_client = QdrantClient(url=QDRANT_URL, api_key=QDRANT_API_KEY)
-
-vector_store = Qdrant(
-    client=qdrant_client,
-    collection_name=COLLECTION_NAME,
-    embeddings=embedding
-)
-
 retriever = vector_store.as_retriever(search_kwargs={"k": 3})
 
 # Set up RAG QA chain
@@ -110,6 +110,26 @@ def generate_prompt(question: str) -> str:
 Question: {question}
 Answer:"""
 
+
+# def generate_prompt(question: str) -> str:
+#     lang = detect(question)
+#     if lang == "ar":
+#         return (
+#             "أجب على السؤال الطبي التالي بلغة عربية فصحى، بإجابة دقيقة ومفصلة. إذا لم تجد معلومات كافية في السياق، استخدم معرفتك الطبية السابقة. \n"
+#             "- عدم تكرار أي نقطة أو عبارة أو كلمة\n"
+#             "- وضوح وسلاسة كل نقطة\n"
+#             "- تجنب الحشو والعبارات الزائدة\n"
+#             f"\nالسؤال: {question}\nالإجابة:"
+#         )
+#     else:
+#         return (
+#             "Answer the following medical question in clear English with a detailed, non-redundant response. "
+#             "Do not repeat ideas, phrases, or restate the question in the answer. If the context lacks relevant "
+#             "information, rely on your prior medical knowledge. If the answer involves multiple points, list them "
+#             "in concise and distinct bullet points:\n"
+#             f"Question: {question}\nAnswer:"
+#         )
+        
 # Input schema
 # class ChatRequest(BaseModel):
 #     message: str