updated libs , fixed breaking changes and added code for the community

.gitignore

@@ -164,3 +164,11 @@ data/chroma/863ec92d-3170-4dcc-b3ac-7b99a5b05b5e/data_level0.bin
 data/chroma/863ec92d-3170-4dcc-b3ac-7b99a5b05b5e/header.bin
 data/chroma/863ec92d-3170-4dcc-b3ac-7b99a5b05b5e/length.bin
 data/chroma/863ec92d-3170-4dcc-b3ac-7b99a5b05b5e/link_lists.bin
+.env_example
+docs/Perizia.pdf
+docs/detrazioni.txt
+sample-docs/detrazioni-miste.txt
+data/chroma/e2c49fd5-6e9e-4f7f-a342-50ec3a0e0cda/data_level0.bin
+data/chroma/e2c49fd5-6e9e-4f7f-a342-50ec3a0e0cda/header.bin
+data/chroma/e2c49fd5-6e9e-4f7f-a342-50ec3a0e0cda/length.bin
+data/chroma/e2c49fd5-6e9e-4f7f-a342-50ec3a0e0cda/link_lists.bin

Pipfile

@@ -21,6 +21,8 @@ azure-identity = "==1.15.0"
 chromadb = "*"
 llama-index = "*"
 traceloop-sdk = "*"
+langchain-openai = "*"
+sentence-transformers = "*"
 
 [dev-packages]
 ipykernel = "*"

community/import-data.py

@@ -0,0 +1,176 @@
+from langchain.chains import ConversationalRetrievalChain
+from langchain_openai import AzureChatOpenAI
+from langchain_openai import AzureOpenAIEmbeddings
+from langchain_community.vectorstores.chroma import Chroma
+from langchain.text_splitter import (
+    RecursiveCharacterTextSplitter,
+    SentenceTransformersTokenTextSplitter,
+)
+from pypdf import PdfReader
+from langchain.document_loaders import PyPDFLoader
+
+import os
+from dotenv import load_dotenv
+
+load_dotenv()
+
+persist_directory = "data/chroma"
+collection_name = "book_eng"
+
+# embedding
+embedding = AzureOpenAIEmbeddings(
+    deployment=os.getenv("OPENAI_EMBEDDING_DEPLOYMENT_NAME"),
+    chunk_size=1,
+    embedding_ctx_length=1000,
+)
+
+# llm
+llm = AzureChatOpenAI(
+    azure_deployment=os.getenv("OPENAI_CHAT_DEPLOYMENT_NAME"),
+    openai_api_type="azure",
+    temperature=0.0,
+)
+
+vector_db = Chroma(
+    collection_name=collection_name,
+    persist_directory=persist_directory,
+    embedding_function=embedding,
+)
+
+
+def load_pdf(pdf_path: str):
+    from io import BytesIO
+
+    # Rec char Splitter
+    character_splitter = RecursiveCharacterTextSplitter(
+        chunk_size=1000,
+        chunk_overlap=0,
+        separators=["\n\n", "\n", ". ", " ", ""],
+    )
+
+    # sentence token splitter
+    token_splitter = SentenceTransformersTokenTextSplitter(
+        chunk_overlap=0, tokens_per_chunk=256
+    )
+
+    pdf_bytes = None
+    with open(pdf_path, "rb") as f:
+        pdf_bytes = f.read()
+
+    doc = PdfReader(BytesIO(pdf_bytes))
+    docs = []
+    for page_num in range(len(doc.pages)):
+        pdf_page = doc.pages[page_num]
+        pdf_page_text = pdf_page.extract_text()
+
+        # skip empty pages
+        if not pdf_page_text:
+            continue
+
+        # split text
+        character_split_texts = character_splitter.split_text(pdf_page_text)
+
+        token_split_texts = []
+        for text in character_split_texts:
+            token_split_texts += token_splitter.split_text(text)
+
+        # create metadata from token split
+        page_nr = int(page_num + 1)
+
+        # set metadata for each split
+        metadatas = [{"source": pdf_path, "page": page_nr} for _ in token_split_texts]
+
+        # convert to document
+        documents = character_splitter.create_documents(
+            texts=token_split_texts, metadatas=metadatas
+        )
+
+        docs.extend(documents)
+
+    db = Chroma.from_documents(
+        collection_name=collection_name,
+        documents=docs,
+        embedding=embedding,
+        persist_directory=persist_directory,
+    )
+
+
+# just for testing
+def get_relevant_documents(question: str):
+    docs = vector_db.similarity_search(question)
+    return docs
+
+
+def ask(
+    query: str,
+) -> str:
+    """Ask to LLM and return the answer"""
+    from langchain.retrievers import MultiQueryRetriever
+    from langchain.prompts import PromptTemplate
+
+    # Continue with the question using doc retriever
+    # get the prompt
+    final_prompt = """
+        You are a friendly helpful assistant to help and maintain polite conversation. 
+        Your users are asking questions about information retrieved from a book.
+        Answer the user's question using only these information.
+        Remember to be polite and friendly.
+        
+        Context: 
+        {context}
+
+        Question: 
+        {question}
+
+        Answer:
+        """
+
+    PROMPT = PromptTemplate(
+        template=final_prompt,
+        input_variables=["context", "question"],
+    )
+
+    # see on langsmith
+    retriever = MultiQueryRetriever.from_llm(
+        vector_db.as_retriever(
+            search_type="mmr",
+            search_kwargs={"k": 3, "score_threshold": 0.9},
+        ),
+        llm=llm,
+    )
+
+    chain_type_kwargs = {"prompt": PROMPT}
+    qa = ConversationalRetrievalChain.from_llm(
+        llm,
+        retriever,
+        return_source_documents=True,
+        verbose=True,
+        combine_docs_chain_kwargs=chain_type_kwargs,
+        rephrase_question=False,
+    )
+
+    llm_result = qa.invoke({"question": query, "chat_history": []})
+    return llm_result
+
+
+if __name__ == "__main__":
+    ### code examples based on https://amzn.eu/d/5LZBc6p
+    import logging
+
+    # FOR LOGGING MULTIQUERY
+    logging.basicConfig()
+    logging.getLogger("langchain.retrievers.multi_query").setLevel(logging.INFO)
+
+    # LOAD PDF
+    pdf_path = os.getenv("FILE_PATH_ENG")
+    print(pdf_path)
+
+    # load_pdf(pdf_path)
+    # print("pdf loaded")
+
+    # result = get_relevant_documents("What's the name of the singer?")
+    # print(result)
+
+    # result = ask("What's the name of the singer?")
+    result = ask("What is the instrument played by Fred?")
+    print(result["answer"])

lcel/rag-sample.py

@@ -30,7 +30,7 @@
 
 def _get_retriever():
     vectorstore = Chroma.from_texts(
-        ["Luca is passionate about LLMs and GenerativeAi"], embedding=embedding
+        ["Sam is passionate about LLMs and GenerativeAi"], embedding=embedding
     )
     retriever = vectorstore.as_retriever()
     return retriever
@@ -87,7 +87,7 @@ def rag_with_different_inputs(question: str, style: str):
 
 
 if __name__ == "__main__":
-    question = "What is Luca passionate about?"
+    question = "What is Sam passionate about?"
 
     # result = simple_rag(question=question)
     style = "Heavy Metal Fan"

rag/rag-lama.py

@@ -39,8 +39,8 @@
 documents = [Document(text=file_content)]
 
 
-api_key = os.getenv("OPENAI_API_KEY")
-azure_endpoint = os.getenv("OPENAI_API_BASE")
+api_key = os.getenv("AZURE_OPENAI_API_KEY")
+azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT")
 api_version = os.getenv("OPENAI_API_VERSION")
 
 llm = AzureOpenAI(

rag/rag-llama-window-retrieval.py

@@ -40,8 +40,8 @@
 
 
 ### THE LLM
-api_key = os.getenv("OPENAI_API_KEY")
-azure_endpoint = os.getenv("OPENAI_API_BASE")
+api_key = os.getenv("AZURE_OPENAI_API_KEY")
+azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT")
 api_version = os.getenv("OPENAI_API_VERSION")
 
 llm = AzureOpenAI(