modify word doc chunking

.gitattributes

@@ -38,3 +38,6 @@ faiss_word_doc_index/index.faiss filter=lfs diff=lfs merge=lfs -text
 faiss_excel_doc_index/index.pkl filter=lfs diff=lfs merge=lfs -text
 faiss_excel_doc_index/index.faiss filter=lfs diff=lfs merge=lfs -text
 combined_keyword_retriever.pkl filter=lfs diff=lfs merge=lfs -text
+faiss_recursive_split_word_doc_index/index.faiss filter=lfs diff=lfs merge=lfs -text
+faiss_recursive_split_word_doc_index/index.pkl filter=lfs diff=lfs merge=lfs -text
+combined_recursive_keyword_retriever.pkl filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,7 @@
+faiss_word_doc_index/*
+recursice_word_keyword_retriever.pkl
+word_keyword_retriever.pkl
+excel_keyword_retriever.pkl
+rag_pipeline.ipynb
+data_loader.ipynb
+combined_keyword_retriever.pkl

app.py

@@ -1,4 +1,5 @@
 from langchain_core.output_parsers import StrOutputParser
+from langchain_core.prompts import ChatPromptTemplate,MessagesPlaceholder
 from langchain_core.runnables import RunnablePassthrough
 from langchain_huggingface.embeddings import HuggingFaceEmbeddings
 from langchain.retrievers.document_compressors import EmbeddingsFilter
@@ -16,7 +17,7 @@ GROQ_API_KEY="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"
 os.environ['GROQ_API_KEY'] = GROQ_API_KEY
 
 
-embed_model = HuggingFaceEmbeddings(model_name="Alibaba-NLP/gte-multilingual-base", model_kwargs={"trust_remote_code":True})
+embed_model = HuggingFaceEmbeddings(model_name="Alibaba-NLP/gte-multilingual-base", model_kwargs={"trust_remote_code":True, "device": "cuda"})
 llm = ChatGroq(
     model="llama-3.1-8b-instant",  
     temperature=0.0,
@@ -25,11 +26,11 @@ llm = ChatGroq(
 )
 
 excel_vectorstore = FAISS.load_local(folder_path="./faiss_excel_doc_index", embeddings=embed_model, allow_dangerous_deserialization=True)
-word_vectorstore = FAISS.load_local(folder_path="./faiss_word_doc_index", embeddings=embed_model, allow_dangerous_deserialization=True)
+word_vectorstore = FAISS.load_local(folder_path="./faiss_recursive_split_word_doc_index", embeddings=embed_model, allow_dangerous_deserialization=True)
 excel_vectorstore.merge_from(word_vectorstore)
 combined_vectorstore = excel_vectorstore
 
-with open('combined_keyword_retriever.pkl', 'rb') as f:
+with open('combined_recursive_keyword_retriever.pkl', 'rb') as f:
     combined_keyword_retriever = pickle.load(f)
     combined_keyword_retriever.k = 10
 
@@ -47,18 +48,31 @@ compression_retriever = ContextualCompressionRetriever(
     base_compressor=embeddings_filter, base_retriever=ensemble_retriever
 )
 
-prompt = hub.pull("rlm/rag-prompt")
+template = """
+User: You are an AI Assistant that follows instructions extremely well.
+Please be truthful and give direct answers. Please tell 'I don't know' if user query is not in CONTEXT
 
+Keep in mind, you will lose the job, if you answer out of CONTEXT questions
+
+CONTEXT: {context}
+Query: {question}
+
+Remember only return AI answer
+Assistant:
+"""
+
+prompt = ChatPromptTemplate.from_template(template)
+output_parser = StrOutputParser()
 
 def format_docs(docs):
     return "\n\n".join(doc.page_content for doc in docs)
 
 
 rag_chain = (
-    {"context": compression_retriever | format_docs, "question": RunnablePassthrough()}
+    {"context": compression_retriever.with_config(run_name="Docs") | format_docs, "question": RunnablePassthrough()}
     | prompt
     | llm
-    | StrOutputParser()
+    | output_parser
 )
 
 
@@ -67,19 +81,38 @@ rag_chain = (
 
 # zero = torch.Tensor([0]).cuda()
 
-@spaces.GPU
+# @spaces.GPU
 def get_response(question, history):
-    print(question)
-    # for chunk in rag_chain.stream(question):
-    #     yield chunk
-    respose =  rag_chain.invoke(question)
-    print(respose)
-    
-    return respose
-
+    # print(question)
+    curr_ans = ""
+    for chunk in rag_chain.stream(question):
+        curr_ans += chunk
+        yield curr_ans
+
+example_questions = [
+    "الموسم المناسب لزراعة الذرة العلفية ؟",
+    "ما هي الاحتياجات المائية لتربية الحيوانات؟",
+    "ما هي خطوات إنتاج الشتلات؟",
+    "الموسم المناسب لزراعة الطماطم في الحقل المكشوف بدولة الإمارات؟",
+    "شروط اختيار مكان منحل العسل؟",
+    "ما هو تقييم مطعم قصر نجد؟",
+    "ما كمية أعلاف الجت المستلمة في منطقة الظفرة عام 2022",
+    "ما مساحات المزارع المروية بالتنقيط في منطقة الرحبة عام 2020",
+    "في إمارة أبوظبي في عام 2022، هل نسبة العينات الغذائية الغير مطابقة من إجمالي العينات أعلى في العينات المحلية أم العينات المستوردة"
+]
 with gr.Blocks() as demo:
+    gr.Markdown(
+        """
+        # ADAFSA RAG Chatbot Demo
+        """
+    )
     chatbot = gr.Chatbot(placeholder="<strong>ADAFSA-RAG Chatbot</strong>")
-    gr.ChatInterface(fn=get_response, chatbot=chatbot)
+    gr.ChatInterface(
+        title="",
+        fn=get_response, 
+        chatbot=chatbot,
+        examples=example_questions,
+    )
     
 demo.launch()
 

combined_recursive_keyword_retriever.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71c816aa5e0cb849c3c9f36ca72ecf7b0968d0fd5ab5a63a3316223e68d5398d
+size 8449174

faiss_recursive_split_word_doc_index/index.faiss

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e21d5d78d4acf373e94ae40d43fcad7b724207b7b4c18455cc1fc613b6c01f5
+size 14736429

faiss_recursive_split_word_doc_index/index.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:27889ba1e7400d896ad677b1e545fd7a01ee16b8d2dbd3c2b9c6431d5b0ff50d
+size 4029431