Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -623,13 +623,14 @@ def calculate_relevance_score(summary, model):
|
|
623 |
print(f"Error parsing relevance score: {e}")
|
624 |
return 0.00
|
625 |
|
|
|
626 |
def rephrase_for_search(query, model):
|
627 |
rephrase_prompt = PromptTemplate(
|
628 |
input_variables=["query"],
|
629 |
template="""
|
630 |
Rephrase the following conversational query into a concise, search-engine-friendly format.
|
631 |
Remove any conversational elements and focus on the core information need.
|
632 |
-
Provide
|
633 |
|
634 |
Conversational query: {query}
|
635 |
|
@@ -639,8 +640,22 @@ def rephrase_for_search(query, model):
|
|
639 |
chain = LLMChain(llm=model, prompt=rephrase_prompt)
|
640 |
response = chain.run(query=query).strip()
|
641 |
|
642 |
-
# Extract only the rephrased query
|
643 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
644 |
|
645 |
return rephrased_query
|
646 |
|
@@ -658,12 +673,16 @@ def ask_question(question, temperature, top_p, repetition_penalty, web_search, g
|
|
658 |
else:
|
659 |
database = None
|
660 |
|
|
|
661 |
if web_search:
|
662 |
-
|
663 |
-
rephrased_query = rephrase_for_search(
|
664 |
-
print(f"Original query: {
|
665 |
print(f"Rephrased query: {rephrased_query}")
|
666 |
|
|
|
|
|
|
|
667 |
search_results = google_search(rephrased_query)
|
668 |
web_docs = [Document(page_content=result["text"], metadata={"source": result["link"]}) for result in search_results if result["text"]]
|
669 |
|
|
|
623 |
print(f"Error parsing relevance score: {e}")
|
624 |
return 0.00
|
625 |
|
626 |
+
|
627 |
def rephrase_for_search(query, model):
|
628 |
rephrase_prompt = PromptTemplate(
|
629 |
input_variables=["query"],
|
630 |
template="""
|
631 |
Rephrase the following conversational query into a concise, search-engine-friendly format.
|
632 |
Remove any conversational elements and focus on the core information need.
|
633 |
+
Provide ONLY the rephrased query without any explanation or additional text.
|
634 |
|
635 |
Conversational query: {query}
|
636 |
|
|
|
640 |
chain = LLMChain(llm=model, prompt=rephrase_prompt)
|
641 |
response = chain.run(query=query).strip()
|
642 |
|
643 |
+
# Extract only the rephrased query using regex
|
644 |
+
match = re.search(r'^(.*?)(?:\n|$)', response)
|
645 |
+
if match:
|
646 |
+
rephrased_query = match.group(1).strip()
|
647 |
+
else:
|
648 |
+
rephrased_query = response.strip()
|
649 |
+
|
650 |
+
# Remove any "Rephrased query:" prefix if present
|
651 |
+
rephrased_query = re.sub(r'^Rephrased query:\s*', '', rephrased_query, flags=re.IGNORECASE)
|
652 |
+
|
653 |
+
# Check if the rephrased query is actually a rephrasing and not the original prompt or instructions
|
654 |
+
if (rephrased_query.lower().startswith(("rephrase", "your task")) or
|
655 |
+
len(rephrased_query.split()) > len(query.split()) * 2):
|
656 |
+
# If it's not a proper rephrasing, use a simple keyword extraction
|
657 |
+
keywords = ' '.join(word for word in query.lower().split() if word not in {'how', 'did', 'the', 'in', 'a', 'an', 'and', 'or', 'but', 'is', 'are', 'was', 'were'})
|
658 |
+
return keywords
|
659 |
|
660 |
return rephrased_query
|
661 |
|
|
|
673 |
else:
|
674 |
database = None
|
675 |
|
676 |
+
# In the ask_question function:
|
677 |
if web_search:
|
678 |
+
original_query = question
|
679 |
+
rephrased_query = rephrase_for_search(original_query, model)
|
680 |
+
print(f"Original query: {original_query}")
|
681 |
print(f"Rephrased query: {rephrased_query}")
|
682 |
|
683 |
+
if rephrased_query == original_query:
|
684 |
+
print("Warning: Query was not rephrased. Using original query for search.")
|
685 |
+
|
686 |
search_results = google_search(rephrased_query)
|
687 |
web_docs = [Document(page_content=result["text"], metadata={"source": result["link"]}) for result in search_results if result["text"]]
|
688 |
|