Create main.py

main.py

@@ -0,0 +1,103 @@
+import os
+import streamlit as st
+import pickle
+import time
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from langchain.llms.base import LLM
+from langchain.chains import RetrievalQAWithSourcesChain
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.document_loaders import UnstructuredURLLoader
+from langchain.vectorstores import FAISS
+from secret_key import google_genai_api_key
+from langchain.embeddings import HuggingFaceEmbeddings
+
+
+class CustomHuggingFaceLLM(LLM):
+    def __init__(self, model_name, temperature=0.7):
+        self.model = AutoModelForCausalLM.from_pretrained(model_name)
+        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
+        self.temperature = temperature
+
+    def _call(self, prompt, stop=None):
+        input_ids = self.tokenizer.encode(prompt, return_tensors="pt")
+        output = self.model.generate(
+            input_ids,
+            max_length=512,
+            temperature=self.temperature,
+            do_sample=True,
+            top_p=0.95,
+            top_k=3
+        )
+        generated_text = self.tokenizer.decode(output[0], skip_special_tokens=True)
+        return generated_text
+
+    @property
+    def _identifying_params(self):
+        return {"model_name": self.model.config._name_or_path, "temperature": self.temperature}
+
+    @property
+    def _llm_type(self):
+        return "custom_huggingface"
+
+
+
+main_directory = os.path.dirname(os.path.abspath(__file__))
+
+st.title("Web Page search Bot: Research Tool 📈")
+st.sidebar.title("Article URLs")
+
+urls = []
+for i in range(3):
+    url = st.sidebar.text_input(f"URL {i+1}")
+    urls.append(url)
+
+process_url_clicked = st.sidebar.button("Process URLs")
+file_path_faiss = "faiss_store.pkl"
+
+main_placeholder = st.empty()
+
+# Load a pre-trained embedding model
+embedding_model = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L6-v2')
+llm = CustomHuggingFaceLLM(model_name="meta-llama/Meta-Llama-3.1-8B", temperature=0.6)
+if process_url_clicked:
+    # load data
+    loader = UnstructuredURLLoader(urls=urls)
+    main_placeholder.text("Data Loading...Started...✅✅✅")
+    data = loader.load()
+    # split data
+    # Do not include unnecessary separators like , and . It will reduce chunks too small.
+    text_splitter = RecursiveCharacterTextSplitter(
+        separators=['\n\n'],
+        chunk_size=1000,
+        chunk_overlap=100
+    )
+    main_placeholder.text("Text Splitter...Started...✅✅✅")
+    docs = text_splitter.split_documents(data)
+
+    # create embeddings and save it to FAISS index
+    vectorstore_faiss = FAISS.from_documents(documents=docs,embedding=embedding_model)
+    main_placeholder.text("Embedding Vector Started Building...✅✅✅")
+    time.sleep(2)
+    
+    # Save the FAISS index to a pickle file
+    with open(file_path_faiss, "wb") as f:
+        pickle.dump(vectorstore_faiss, f)
+
+query = main_placeholder.text_input("Question: ")
+if query:
+    if os.path.exists(file_path_faiss):
+        with open(file_path_faiss, "rb") as f:
+            vectorstore = pickle.load(f)
+            chain = RetrievalQAWithSourcesChain.from_llm(llm=llm, retriever=vectorstore.as_retriever(), verbose=True) # type: ignore
+            result = chain({"question": query}, return_only_outputs=True)
+            # result will be a dictionary of this format --> {"answer": "", "sources": [] }
+            st.header("Answer")
+            st.write(result["answer"])
+
+            # Display sources, if available
+            sources = result.get("sources", "")
+            if sources:
+                st.subheader("Sources:")
+                sources_list = sources.split("\n")  # Split the sources by newline
+                for source in sources_list:
+                    st.write(source)