added feature llm response text generation

app.py

@@ -8,6 +8,8 @@ import PyPDF2
 from tqdm.auto import tqdm
 import math
 from transformers import pipeline
+from langchain.prompts import ChatPromptTemplate
+import re
 # import json
 
 # st.config(PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION="python")
@@ -76,8 +78,10 @@ def get_pinecone_semantic_index(pinecone):
     # st.text(f"Succesfully connected to the pinecone index")
     return index
 
-def promt_engineer(text):
-    prompt_template = """
+
+
+def promt_engineer(text, query):
+    summary_prompt_template = """
     write a concise summary of the following text delimited by triple backquotes.
     return your response in bullet points which convers the key points of the text.
 
@@ -85,11 +89,11 @@ def promt_engineer(text):
 
     BULLET POINT SUMMARY:
     """
-    # Load the summarization pipeline with the specified model
+    Load the summarization pipeline with the specified model
     summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
 
     # Generate the prompt
-    prompt = prompt_template.format(text=text)
+    prompt = summary_prompt_template.format(text=text)
 
     # Generate the summary
     summary = summarizer(prompt, max_length=1024, min_length=50)[0]["summary_text"]
@@ -100,7 +104,38 @@ def promt_engineer(text):
         st.write(summary)
         st.divider()
 
-    return summary
+    GENERATION_PROMPT_TEMPLATE = """
+    Instructions:
+    -------------------------------------------------------------------------------------------------------------------------------
+    Answer the question only based on the below context:
+    - You're a Research AI expert in the explaining and reading the research papers.
+    - Questions with out-of-context replay with The question is out of context. 
+    - Always try to provide Keep it simple answers in nice format without incomplete sentence.
+    - Give the answer atleast 5 seperate lines addition to the title info.
+    - Only If question is relevent to context provide Doc Title: <title> Paragraph: <Paragraph> Page No: <pagenumber> 
+    -------------------------------------------------------------------------------------------------------------------------------
+    {context}
+    -------------------------------------------------------------------------------------------------------------------------------
+    Answer the question based on the above context: {question}
+    """
+
+    prompt_template = ChatPromptTemplate.from_template(GENERATION_PROMPT_TEMPLATE)
+    prompt = prompt_template.format(context=text, question=query)
+    response_text = ""
+    result = ""
+    
+    try:
+        llm = HuggingFaceHub(
+            repo_id="meta-llama/Meta-Llama-3-8B-Instruct", model_kwargs={"temperature": 0.1, "max_new_tokens": 256, "task":"text-generation"}
+        )
+        response_text = llm.invoke(prompt)
+        escaped_query = re.escape(query)
+        result = re.split(f'Answer the question based on the above context: {escaped_query}\n',response_text)[-1]
+        st.error(f"Error invoke: {e}")
+    except Exception as e:
+        st.error(f"Error invoke: {e}")
+
+    return summary, result
 
 def chat_actions():
     
@@ -111,7 +146,8 @@ def chat_actions():
         {"role": "user", "content": st.session_state["chat_input"]},
     )
 
-    query_embedding = model.encode(st.session_state["chat_input"])
+    query = st.session_state["chat_input"]
+    query_embedding = model.encode(query)
     # create the query vector
     query_vector = query_embedding.tolist()
     # now query vector database
@@ -136,13 +172,13 @@ def chat_actions():
         p = math.pow(1024, 2)
         mbsize = round(len(bytesize) / p, 2)
         st.write(f"Text lenth of {len(consolidated_text)} characters with {mbsize}MB size")
-        summary = promt_engineer(consolidated_text[:1024])
+        summary, response = promt_engineer(consolidated_text[:1024], query)
 
     for res in result['matches']:
         st.session_state["chat_history"].append(
             {
                 "role": "assistant",
-                "content": f"{summary}",
+                "content": f"{response}",
             },  # This can be replaced with your chat response logic
         )
         break;