initial commit

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.sqlite3 filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,96 @@
+import streamlit as st
+from langchain import memory as lc_memory
+from langsmith import Client
+from streamlit_feedback import streamlit_feedback
+from utils import get_expression_chain, retriever
+from langchain_core.tracers.context import collect_runs
+from dotenv import load_dotenv
+
+load_dotenv()
+client = Client()
+
+st.set_page_config(page_title = "SUP'ASSISTANT")
+st.subheader("Hey there! How can I help you today!")
+
+memory = lc_memory.ConversationBufferMemory(
+    chat_memory=lc_memory.StreamlitChatMessageHistory(key="langchain_messages"),
+    return_messages=True,
+    memory_key="chat_history",
+)
+st.sidebar.markdown("## Feedback Scale")
+feedback_option = (
+    "thumbs" if st.sidebar.toggle(label="`Faces` ⇄ `Thumbs`", value=False) else "faces"
+)
+with st.sidebar:
+    model_name = st.selectbox("**Model**", options=["llama-3.1-70b-versatile","gemma2-9b-it","gemma-7b-it","llama-3.2-3b-preview", "llama3-70b-8192", "mixtral-8x7b-32768"])
+    temp = st.slider("**Temperature**", min_value=0.0, max_value=1.0, step=0.001)
+    n_docs = st.number_input("**Number of retireved documents**", min_value=0, max_value=10, value=5, step=1)
+if st.sidebar.button("Clear message history"):
+    print("Clearing message history")
+    memory.clear()
+
+retriever = retriever(n_docs=n_docs)
+# Create Chain
+chain = get_expression_chain(retriever,model_name,temp)
+
+for msg in st.session_state.langchain_messages:
+    avatar = "🦜" if msg.type == "ai" else None
+    with st.chat_message(msg.type, avatar=avatar):
+        st.markdown(msg.content)
+
+if prompt := st.chat_input(placeholder="Ask me a question!"):
+    st.chat_message("user").write(prompt)
+    with st.chat_message("assistant", avatar="🦜"):
+        message_placeholder = st.empty()
+        full_response = ""
+        # Define the basic input structure for the chains
+        input_dict = {"input": prompt}
+
+        with collect_runs() as cb:
+            for chunk in chain.stream(input_dict, config={"tags": ["Streamlit Chat"]}):
+                full_response += chunk.content
+                message_placeholder.markdown(full_response + "▌")
+            memory.save_context(input_dict, {"output": full_response})
+            st.session_state.run_id = cb.traced_runs[0].id
+        message_placeholder.markdown(full_response)
+
+if st.session_state.get("run_id"):
+    run_id = st.session_state.run_id
+    feedback = streamlit_feedback(
+        feedback_type=feedback_option,
+        optional_text_label="[Optional] Please provide an explanation",
+        key=f"feedback_{run_id}",
+    )
+
+    # Define score mappings for both "thumbs" and "faces" feedback systems
+    score_mappings = {
+        "thumbs": {"👍": 1, "👎": 0},
+        "faces": {"😀": 1, "🙂": 0.75, "😐": 0.5, "🙁": 0.25, "😞": 0},
+    }
+
+    # Get the score mapping based on the selected feedback option
+    scores = score_mappings[feedback_option]
+
+    if feedback:
+        # Get the score from the selected feedback option's score mapping
+        score = scores.get(feedback["score"])
+
+        if score is not None:
+            # Formulate feedback type string incorporating the feedback option
+            # and score value
+            feedback_type_str = f"{feedback_option} {feedback['score']}"
+
+            # Record the feedback with the formulated feedback type string
+            # and optional comment
+            feedback_record = client.create_feedback(
+                run_id,
+                feedback_type_str,
+                score=score,
+                comment=feedback.get("text"),
+            )
+            st.session_state.feedback = {
+                "feedback_id": str(feedback_record.id),
+                "score": score,
+            }
+        else:
+            st.warning("Invalid feedback score.")

requirements.txt

@@ -0,0 +1,13 @@
+langchain-groq
+langchain-core
+streamlit
+langchain-chroma
+langchain-nomic
+langchain
+nomic
+python-dotenv
+langchain-community
+rank_bm25
+cohere
+nomic[local]
+streamlit-feedback

sup-knowledge-eng-nomic/chroma.sqlite3

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:112a7ee3f7fb675803ed49ffe7901311156373f8ba3142c3a3026b2f3936d633
+size 7704576

sup-knowledge-eng-nomic/ec6754ec-5fa6-4b04-bfb6-d2f052cd81fe/data_level0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a13e72541800c513c73dccea69f79e39cf4baef4fa23f7e117c0d6b0f5f99670
+size 3212000

sup-knowledge-eng-nomic/ec6754ec-5fa6-4b04-bfb6-d2f052cd81fe/header.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ec6df10978b056a10062ed99efeef2702fa4a1301fad702b53dd2517103c746
+size 100

sup-knowledge-eng-nomic/ec6754ec-5fa6-4b04-bfb6-d2f052cd81fe/length.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5fb64b021f47ff585087f63e019088911fa892704ffa3e9506f3a120d807cfa
+size 4000

utils.py

@@ -0,0 +1,93 @@
+from langchain_chroma import Chroma
+from langchain_nomic.embeddings import NomicEmbeddings
+from langchain_core.documents import Document
+from langchain.retrievers.document_compressors import CohereRerank
+from langchain.retrievers import ContextualCompressionRetriever
+from langchain.retrievers import BM25Retriever, EnsembleRetriever
+from langchain_groq import ChatGroq
+from dotenv import load_dotenv
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.runnables import Runnable, RunnableMap
+from langchain.schema import BaseRetriever
+load_dotenv()
+
+def retriever(n_docs=5):
+    vector_database_path = "sup-knowledge-eng-nomic"
+
+    embeddings_model = NomicEmbeddings(model="nomic-embed-text-v1.5", inference_mode="local")
+
+
+    vectorstore = Chroma(collection_name="sup-store-eng-nomic",
+                        persist_directory=vector_database_path,
+                    embedding_function=embeddings_model)
+
+    vs_retriever = vectorstore.as_retriever(k=n_docs)
+
+    texts = vectorstore.get()['documents']
+    metadatas = vectorstore.get()["metadatas"]
+
+    documents = []
+    for i in range(len(texts)):
+        doc = Document(page_content=texts[i], metadata=metadatas[i])
+        documents.append(doc)
+
+    keyword_retriever = BM25Retriever.from_documents(documents)
+    keyword_retriever.k =  n_docs
+
+    ensemble_retriever = EnsembleRetriever(retrievers=[vs_retriever,keyword_retriever],
+                                       weights=[0.5, 0.5])
+    
+    compressor = CohereRerank(model="rerank-english-v3.0")
+    retriever = ContextualCompressionRetriever(
+        base_compressor=compressor, base_retriever=ensemble_retriever
+    )
+
+    return retriever
+
+rag_prompt = """You are an assistant for question-answering tasks.
+The questions that you will be asked will mainly be about SUP'COM (also known as Higher School Of Communication Of Tunis).
+
+Here is the context to use to answer the question:
+
+{context} 
+
+Think carefully about the above context. 
+
+Now, review the user question:
+
+{input}
+
+Provide an answer to this questions using only the above context. 
+
+Answer:"""
+
+# Post-processing
+def format_docs(docs):
+    return "\n\n".join(doc.page_content for doc in docs)
+
+def get_expression_chain(retriever: BaseRetriever, model_name="llama-3.1-70b-versatile", temp=0
+) -> Runnable:
+    """Return a chain defined primarily in LangChain Expression Language"""
+    def retrieve_context(input_text):
+        # Use the retriever to fetch relevant documents
+        docs = retriever.get_relevant_documents(input_text)
+        return format_docs(docs)
+    
+    ingress = RunnableMap(
+        {
+            "input": lambda x: x["input"],
+            "context": lambda x: retrieve_context(x["input"]),
+        }
+    )
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            (
+                "system",
+                rag_prompt
+            )
+        ]
+    )
+    llm = ChatGroq(model=model_name, temperature=temp)
+
+    chain = ingress | prompt | llm
+    return chain