'expected string or buffer' or 'AssertionError' when it comes to embedding

[VadeanFlaviuAlexandru]

Long story short: i made my own 'chat with your PDF' demo with streamlit and if i use huggingFace embedding i get the following error: 'AssertionError'. If i use openAI embedding i get: "expected string or buffer". Im also trying for it to be automatic, so i just upload the file, click a button and wait for the output. Can someone tell me what the problem is:

if pdf is not None:
    buttonPressed = st.button("Fa-mi un rezumat!")
    if buttonPressed:
        with st.spinner("Se citeste... 📄🤔... o secunda, te rog..."):
            rawText = get_pdf_text(pdf)
            textChunks = get_text_chunks(rawText)
            vectorstore = get_vectorstore(textChunks, option)
            docs = vectorstore.similarity_search(questions)
            llm = ChatOpenAI(model_name='gpt-3.5-turbo', temperature=0.1)
            chain = load_qa_chain(llm, chain_type="stuff", prompt=prompt)
            with get_openai_callback() as cb:
                response = chain({"context": docs, "input_documents": questions}, return_only_outputs=True)
                print(cb)
            st.write(response["output_text"])

,

def get_text_chunks(text):
    text_splitter = CharacterTextSplitter(
        separator="\n",
        chunk_size=1000,
        chunk_overlap=200,
        length_function=len
    )
    chunks = text_splitter.create_documents(text)
    return chunks

,

def get_vectorstore(text_chunks, freeEmbedding):
    embeddings = OpenAIEmbeddings()
    if freeEmbedding == "Gratis (dar incet)":
        embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-xl")
    vectorstoreDB = FAISS.from_texts(texts=text_chunks, embedding=embeddings)
    return vectorstoreDB

and

prompt_template = """..."""
questions = """ ... """
prompt = PromptTemplate(template=prompt_template, input_variables=["input_documents", "context"])

[dosubot[bot]]

🤖

Hello,

From the code and context you've provided, it seems like the errors might be related to the data types being passed to the embedding functions. However, without the exact error messages and tracebacks, it's hard to pinpoint the exact issue.

For the 'AssertionError' with HuggingFace embeddings, it's possible that the model you're trying to use ('hkunlp/instructor-xl') is not compatible with the function or the data you're passing to it. It would be helpful if you could provide the exact error message and traceback for this issue.

For the 'expected string or buffer' error with OpenAI embeddings, this error typically occurs when a function that expects a string (or buffer) as input is given a different data type. It's possible that the get_pdf_text function is returning a data type that the OpenAI embedding function cannot process. Could you please provide more information about the get_pdf_text function and the data it returns?

Also, it would be helpful to know more about the get_text_chunks and get_vectorstore functions. The expected inputs and outputs for these functions are not clear from the provided context. Could you please provide more information about these functions?

Once we have more information, we should be able to help you resolve these issues. Thank you for your patience.

Sources

• libs/langchain/langchain/vectorstores/singlestoredb.py
• libs/langchain/langchain/text_splitter.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[VadeanFlaviuAlexandru]

def get_pdf_text(pdf_docs):
    pdf_reader = PdfReader(pdf)
    text = ""
    for page in pdf_reader.pages:
        text += page.extract_text()
    return text

[VadeanFlaviuAlexandru]

got it working for who ever is going to stumble upon this post:

prompt_template="""...{summaries}...{question}..."""
questions="""...."""

def get_pdf_text(pdf_docs):
    pdf_reader = PdfReader(pdf_docs)
    text = ""
    for page in pdf_reader.pages:
        text += page.extract_text()
    return text


def get_text_chunks(text):
    text_splitter = CharacterTextSplitter(
        separator="\n",
        chunk_size=1000,
        chunk_overlap=200,
        length_function=len
    )
    chunks = text_splitter.split_text(text)
    return chunks


def get_vectorstore(text_chunks, freeEmbedding):
    metadata_list = [{'source': "local"}] * len(text_chunks)
    embeddings = OpenAIEmbeddings()
    if freeEmbedding == "Gratis (dar incet)":
        embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-xl")
    vectorstoreDB = FAISS.from_texts(texts=text_chunks, embedding=embeddings, metadatas=metadata_list)
    return vectorstoreDB


load_dotenv()
st.set_page_config(page_title="ExtragereCicada - PDF", page_icon="📄")
st.header("Extragere Conținut - PDF")

pdf = st.file_uploader("Incarca PDF:", type="pdf")
option = st.selectbox("Modul de incorporare (cat de rapid se citeste PDF-u):",
                      ("Gratis (dar incet)", "Platit (dar rapid)"))

if pdf is not None:
    buttonPressed = st.button("Fa-mi un rezumat!")
    st.write("-------------------------------")
    if buttonPressed:
        with st.spinner("Se citeste... 📄🤔... o secunda, te rog..."):
            rawText = get_pdf_text(pdf)
            textChunks = get_text_chunks(rawText)
            print('---------TEXTCHUNKS', textChunks)
            vectorstore = get_vectorstore(textChunks, option)
            print('=========vectorSTORE', vectorstore)
            docs = vectorstore.similarity_search(questions)
            print('=========DOCS', docs)
            doc = Document(page_content=prompt_template,
                           metadata={"summaries": docs, "question": questions, "source": pdf.name})
            prompt = PromptTemplate.from_template("{source} {page_content} {summaries} {question}")
            format_document(doc, prompt)
            print('=========DOC', doc)
            llm = ChatOpenAI(model_name='gpt-3.5-turbo', temperature=0.1)
            chain = RetrievalQAWithSourcesChain.from_chain_type(llm=llm,
                                                                chain_type="stuff",
                                                                retriever=vectorstore.as_retriever(),
                                                                chain_type_kwargs={
                                                                    "prompt": PromptTemplate(
                                                                        template=prompt_template,
                                                                        input_variables=["summaries", "question"]
                                                                    )
                                                                })
            st.write(chain({"question": questions, "summaries": doc}, return_only_outputs=True)["answer"])