Updated the prompt for reformulation and randomly giving 100 urls in the bottom

rag.py

@@ -25,11 +25,20 @@ CHUNK_OVERLAP = 128
 embedding_model_hf = "BAAI/bge-m3"
 # embedding_model_hf = "sentence-transformers/all-mpnet-base-v2"
 qa_model_name = "gpt-3.5-turbo"
-bestReformulationPrompt = "Given a chat history and the latest user question, which may reference context from the chat history, you must formulate a standalone question that can be understood without the chat history. You are strictly forbidden from using any outside knowledge. Do not, under any circumstances, answer the question. Reformulate it if necessary; otherwise, return it as is."
-bestSystemPrompt = "You're an assistant for question-answering tasks. Under absolutely no circumstances should you use external knowledge or go beyond the provided preknowledge. Your approach must be systematic and meticulous. First, identify CLUES such as keywords, phrases, contextual information, semantic relations, tones, and references that aid in determining the context of the input. Second, construct a concise diagnostic REASONING process (limiting to 130 words) based on premises supporting the INPUT relevance within the provided context. Third, utilizing the identified clues, reasoning, and input, furnish the pertinent answer for the question. Remember, you are required to use ONLY the provided context to answer the questions. If the question does not align with the preknowledge or if the preknowledge is absent, state that you don't know the answer. External knowledge is strictly prohibited. Failure to adhere will result in incorrect answers. The preknowledge is as follows:"
+bestReformulationPrompt = """Given a chat history and the latest user question, which may reference context from the chat history, you must formulate a standalone question that can be understood without the chat history. You are strictly forbidden from using any outside knowledge. You are strictly forbidden from adding extra things in the question if not required. Do not, under any circumstances, answer the question. Reformulate ONLY if it is necessary; otherwise, return it as is.
+
+Example: I am getting fat, how can I loose weight?
+Reformulated question: What are some effective strategies for losing weight in a healthy manner?
+This reformulation is BAD since it added extra thing "in a health manner". The question was understandable even without chat history.
+
+Example: When was he born
+Chat History: Who is brack obama
+Reformulated question: When was Barack obama born?
+This reformulation is good since I am able to understand the question which I earlier was not able to understand without chat history."""
+bestSystemPrompt = "You're an assistant for question-answering tasks. Under absolutely no circumstances should you use external knowledge or go beyond the provided preknowledge. Your approach must be systematic and meticulous. First, identify CLUES such as keywords, phrases, contextual information, semantic relations, tones, and references that aid in determining the context of the input. Second, construct a concise diagnostic REASONING process (limiting to 130 words) based on premises supporting the INPUT relevance within the provided preknowledge. Third, utilizing the identified clues, reasoning, and input, furnish the pertinent answer for the question. Remember, you are required to use ONLY the provided preknowledge to answer the questions. If the question does not align with the preknowledge or if the preknowledge is absent, state that you don't know the answer. External knowledge is strictly prohibited. Failure to adhere will result in incorrect answers. The preknowledge is as follows:"
 
 # embeddings_oa = OpenAIEmbeddings(model=embedding_model_oa)
-embeddings_hf = HuggingFaceEmbeddings(model_name=embedding_model_hf, show_progress=True)
+embeddings_hf = HuggingFaceEmbeddings(model_name=embedding_model_hf)
 
 
 def setupDb(data_path):
@@ -98,8 +107,8 @@ def chatWithRag(reformulationPrompt, QAPrompt, question, chat_history):
     if QAPrompt != None or len(QAPrompt):
         curr_question_prompt = QAPrompt
 
-    # reformulated_query = reformulate_question(chat_history, question, reformulationPrompt)
-    reformulated_query = question
+    reformulated_query = reformulate_question(chat_history, question, reformulationPrompt)
+    # reformulated_query = question
     retreived_documents = [
         doc
         for doc in db.similarity_search_with_score(reformulated_query)
@@ -144,7 +153,6 @@ with gr.Blocks() as demo:
         )
     output = gr.Textbox(label="Output")
     submit_btn = gr.Button("Submit")
-    selected_urls = random.sample(relevant_content, min(100, len(relevant_content)))
 
     chat_history = gr.State([])
     submit_btn.click(
@@ -158,7 +166,15 @@ with gr.Blocks() as demo:
         outputs=[output, chat_history],
     )
     with gr.Accordion("Urls", open=False):
-        gr.Markdown(", ".join(selected_urls))
+        urls = gr.Markdown()
+
+    demo.load(
+        lambda: ", ".join(
+            random.sample(relevant_content, min(100, len(relevant_content)))
+        ),
+        None,
+        urls,
+    )
 
 gr.close_all()
 demo.launch()