Retriever Made

app.py

@@ -1,64 +1,91 @@
-import gradio as gr
-from huggingface_hub import InferenceClient
+import streamlit as st
+from langchain_community.document_loaders import HuggingFaceDatasetLoader
+from langchain_text_splitters import CharacterTextSplitter
+from langchain.vectorstores import Chroma
+from langchain_aws import BedrockEmbeddings
+from langchain.chat_models import ChatBedrock
+from langchain.schema import HumanMessage
+import os
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+# Optional: For Cohere reranking
+import cohere
 
+# --- Load and Prepare Dataset (run once and cache)
+@st.cache_resource
+def load_and_index_data():
+    loader = HuggingFaceDatasetLoader(
+        dataset_name="Cohere/wikipedia-22-12-simple-embeddings",
+        page_content_column="text",
+        name="train",
+        load_max_docs=100
+    )
+    documents = loader.load()
+    
+    splitter = CharacterTextSplitter(separator="\n", chunk_size=500, chunk_overlap=50)
+    chunks = splitter.split_documents(documents)
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+    embedding = BedrockEmbeddings(
+        model_id="amazon.titan-embed-text-v1",
+        region_name="us-east-1"
+    )
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+    vectordb = Chroma.from_documents(
+        documents=chunks,
+        embedding=embedding,
+        persist_directory="./chromadb"
+    )
+    vectordb.persist()
+    return vectordb
 
-    messages.append({"role": "user", "content": message})
+# --- Re-rank using Claude 3.5 via Bedrock
+def rerank_with_claude(query, docs):
+    claude = ChatBedrock(
+        model_id="anthropic.claude-3-sonnet-20240229-v1:0",
+        region_name="us-east-1"
+    )
+    context = "\n\n".join([f"Document {i+1}:\n{doc.page_content}" for i, doc in enumerate(docs)])
+    prompt = f"""You are a helpful assistant tasked with re-ranking search results based on their relevance to a user query.
+    
+Query: {query}
 
-    response = ""
+Documents:
+{context}
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+Please rank the documents in order of relevance to the query and explain briefly."""
+    response = claude([HumanMessage(content=prompt)])
+    return response.content
 
-        response += token
-        yield response
+# --- Re-rank using Cohere
+def rerank_with_cohere(query, docs):
+    co = cohere.Client(st.secrets["COHERE_API_KEY"])
+    documents = [doc.page_content for doc in docs]
+    results = co.rerank(query=query, documents=documents, top_n=5)
+    return results
 
+# --- Streamlit UI
+st.set_page_config(page_title="Re-ranking Demo", layout="wide")
+st.title("🔎 Wikipedia Search with Re-ranking")
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
+query = st.text_input("Enter your question:")
+rerank_method = st.selectbox("Choose re-ranking method:", ["None (Baseline)", "Claude 3.5 (Bedrock)", "Cohere Rerank"])
 
+if query:
+    vectordb = load_and_index_data()
+    retriever = vectordb.as_retriever(search_kwargs={"k": 5})
+    baseline_results = retriever.get_relevant_documents(query)
+
+    st.subheader("🔍 Baseline Results")
+    for i, doc in enumerate(baseline_results):
+        st.markdown(f"**Doc {i+1}:** {doc.page_content[:300]}...")
+
+    if rerank_method == "Claude 3.5 (Bedrock)":
+        st.subheader("✨ Re-ranked Results (Claude 3.5)")
+        ranked_text = rerank_with_claude(query, baseline_results)
+        st.text(ranked_text)
+
+    elif rerank_method == "Cohere Rerank":
+        st.subheader("✨ Re-ranked Results (Cohere)")
+        reranked = rerank_with_cohere(query, baseline_results)
+        for i, result in enumerate(reranked.results):
+            st.markdown(f"**Doc {i+1}** (score: {result.relevance_score:.2f}):\n\n{result.document[:300]}...")
 
-if __name__ == "__main__":
-    demo.launch()

requirements.txt

@@ -1 +1,8 @@
-huggingface_hub==0.25.2
+huggingface_hub==0.25.2
+streamlit
+langchain
+langchain-community
+langchain-aws
+chromadb
+cohere
+boto3