Replies: 4 comments 8 replies
-
|
To address the issues you're encountering with your RAG chatbot, let's tackle them one by one:
By addressing these specific areas, you should be able to improve the performance and accuracy of your chatbot.
|
Beta Was this translation helpful? Give feedback.
-
|
@dosu can you give a working example of context chat engine and also the storing of messages |
Beta Was this translation helpful? Give feedback.
-
|
To effectively use the For managing load and maintaining context relevance, especially to prevent errors related to message length, implement a strategy to manage the chat history efficiently. This involves limiting the number of messages stored and periodically pruning older messages. If a single message exceeds the token limit, consider splitting it into smaller parts to ensure each is within the limit. Here's a simplified approach to manage the chat history for context relevance: def get(self) -> List[ChatMessage]:
"""Get chat history, ensuring each message part is within the token limit."""
message_count = len(self.chat_history)
message_str = " ".join([str(m.content) for m in self.chat_history[-message_count:]])
token_count = len(self.tokenizer_fn(message_str))
while token_count > self.token_limit and message_count > 1:
message_count -= 1
message_str = " ".join([str(m.content) for m in self.chat_history[-message_count:]])
token_count = len(self.tokenizer_fn(message_str))
if token_count > self.token_limit:
# Split the message into smaller parts if it exceeds the token limit
message_parts = self.chat_history[-message_count].content.split(' ')
split_messages = []
for i in range(0, len(message_parts), self.token_limit):
split_message = ChatMessage(content=' '.join(message_parts[i:i+self.token_limit]))
split_messages.append(split_message)
return split_messages
return self.chat_history[-message_count:]This code snippet helps in managing the chat history effectively by ensuring that each message part is within the token limit, thus preventing errors related to message length. By following these guidelines, you should be able to address the issues with the
|
Beta Was this translation helpful? Give feedback.
-
|
@dosu can you write the whole code from the beginning like from storing on index and not using openai. and can you also code it with streamlit |
Beta Was this translation helpful? Give feedback.
-
|
To create a RAG chatbot for e-book summarization with Streamlit, you'll need to index your e-books, set up a custom language model to avoid using OpenAI, and manage chat history effectively. Here's a streamlined approach:
Here's a basic code structure to get you started: import streamlit as st
from my_custom_llm import MyCustomLLM # Your custom LLM implementation
from ebook_reader import EbookReader # Your custom e-book reader
# Load and index e-books
ebook_path = "path/to/ebook"
ebook_reader = EbookReader(ebook_path)
documents = ebook_reader.load_documents()
# Initialize your custom LLM
custom_llm = MyCustomLLM(model="custom-model-name", temperature=0.5)
# Streamlit UI setup
st.title("E-book Summarization Chatbot")
# Chat history management
if "messages" not in st.session_state:
st.session_state.messages = []
# Input from user
user_input = st.text_input("Ask a question about the e-book:")
# Generate and display response
if user_input:
response = custom_llm.generate_response(user_input, documents)
st.session_state.messages.append({"user": user_input, "assistant": response})
for message in st.session_state.messages:
st.write(f"{message['user']}: {message['assistant']}")This example assumes the existence of
|
Beta Was this translation helpful? Give feedback.
-
|
@dosu can you give me a simple example of context chat engine using huggingface interface API without it throwing a error of message should be of odd length |
Beta Was this translation helpful? Give feedback.
-
|
To effectively use the from llama_index.legacy.llms.types import ChatMessage, MessageRole
# Prepare a sequence of chat messages, ensuring an odd length
messages = [
ChatMessage(role=MessageRole.USER, content="What is the summary of the book?"),
ChatMessage(role=MessageRole.ASSISTANT, content="The book covers various topics including..."),
ChatMessage(role=MessageRole.USER, content="Can you provide more details on the topic of AI advancements?")
]
# Use the chat method to process these messages
response = chat(messages) # Assuming 'chat' is your method to process messages
print(response.message.content)This setup ensures that the sequence alternates correctly between user and assistant messages, starting and ending with a user message, thus maintaining an odd length and avoiding the error.
|
Beta Was this translation helpful? Give feedback.
-
|
@dosu cant you write the whole code in detail while using context chat engine from defaults |
Beta Was this translation helpful? Give feedback.
-
|
To create a detailed example using the
Here's a simplified code structure to illustrate these steps: import streamlit as st
from llama_index.core.chat_engine.condense_plus_context import CondensePlusContextChatEngine
from llama_index.core.memory import ChatMemoryBuffer
from llama_index.core.llms.huggingface import HuggingFaceLLM
from llama_index.core.indices.base_retriever import BaseRetriever
# Initialize your HuggingFace LLM
llm = HuggingFaceLLM(model_name="your-model-name", api_token="your-huggingface-api-token")
# Custom retriever for e-book content
class EbookRetriever(BaseRetriever):
def retrieve(self, message: str):
# Retrieval logic for e-book content
pass
retriever = EbookRetriever()
# Initialize the chat engine
chat_engine = CondensePlusContextChatEngine.from_defaults(retriever=retriever, llm=llm)
# Streamlit UI setup
st.title("E-book Summarizer")
user_input = st.text_input("Ask a question about the e-book:")
if user_input:
response = chat_engine.chat(user_input)
st.write(response.response)This example provides a foundation for building a chatbot that leverages e-book content to generate responses. Remember to adapt the LLM initialization for HuggingFace integration and implement the
|
Beta Was this translation helpful? Give feedback.
-
|
@dosu the above code is showing this error NotImplementedError: Messages passed in must be of odd length |
Beta Was this translation helpful? Give feedback.
-
|
To address the Here's a quick fix you can apply: def ensure_odd_length_messages(chat_history):
if len(chat_history) % 2 == 0:
# Append a system message to make the length odd
chat_history.append(ChatMessage(content="System placeholder message", role="system"))
return chat_history
# Use this function before invoking the chat method
chat_history = ensure_odd_length_messages(chat_history)
response = context_chat_engine.chat("Your message here", chat_history)This function checks the length of Implementing this solution should help you overcome the error and ensure smoother operation of your chatbot.
|
Beta Was this translation helpful? Give feedback.
-
|
@logan-markewich @Thomas-AH-Heller pls help just getting the same error of message should be in odd length when the same code is working for condense question chat engine but i want to use the context chat engine as i want the chatbot to only answer from the context. Please help I read all the documents even that is not helping i kindly help even a update document that shows an example for local LLM or HuggingfaceInterface API would work wonders for me PLS PLS PLS help |
Beta Was this translation helpful? Give feedback.
-
Question
I am trying to create a RAG chatbot that only answers from the content that i provide through pdf and never uses its own knowledge to answer or add information to the answer that is outside of the context. At this moment i am facing three issues:-
the answer length is quite small
the follow up question that it creates is totally of the mark
model overload issues
As i am creating a RAG chatbot for e-book summarization the answers are meant to be lengthy. I have used HuggingFace for embeddings as well as for the LLM. There was very less docs to refer to when using CondenseQuestionChatEngine so add to learn for myself. I thought i had solved it then i realize it was running on default llm which is openai, and after solving that issue all the prompts that i created for the Condensequestion started to fail, now if i run the question in query engine and print it i get the right answers. but the chat engine when trying to make sense using the chat history it completely goes of the grid and ask the wrong question all together and way it is setup the bot will revert with i don't know.
now i tried to switch to context chat engine just to compare the results its giving me an error:-
NotImplementedError("Messages passed in must be of odd length.")
here is my code for reference:-
`
import streamlit as st
from llama_index.core import Settings
import logging
import sys
import os.path
logging.basicConfig(stream=sys.stdout, level=logging.INFO)
logging.getLogger().addHandler(logging.StreamHandler(stream=sys.stdout))
from llama_index.core.service_context import set_global_service_context
from llama_index.llms.llama_cpp import LlamaCPP
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
from langchain.embeddings.huggingface import HuggingFaceBgeEmbeddings
from langchain_community.llms.huggingface_endpoint import HuggingFaceEndpoint
from llama_cpp import Llama
from langchain.llms import HuggingFaceHub
from llama_index.core.prompts.chat_prompts import ChatPromptTemplate, ChatMessage,MessageRole
from llama_index.core.chat_engine import CondenseQuestionChatEngine, ContextChatEngine
from llama_index.legacy.prompts import ChatPromptTemplate
from llama_index.core.base.llms.types import ChatMessage, MessageRole
try:
from llama_index import VectorStoreIndex, ServiceContext, Document, SimpleDirectoryReader, StorageContext, load_index_from_storage
except ImportError:
from llama_index.core import VectorStoreIndex, ServiceContext, Document, SimpleDirectoryReader, StorageContext, load_index_from_storage
from llama_index.llms.huggingface import (
HuggingFaceInferenceAPI,
HuggingFaceLLM,
)
from llama_index.llms.llama_cpp.llama_utils import (
messages_to_prompt,
completion_to_prompt,
)
from huggingface_hub import login
login("MY_HUGGINGFACE_API_KEY")
from transformers import AutoTokenizer
st.set_page_config(page_title="Chat with the Streamlit docs, powered by LlamaIndex", page_icon="🦙", layout="centered", initial_sidebar_state="auto", menu_items=None)
st.title("Chat with the Streamlit docs, powered by LlamaIndex 💬🦙")
st.info("Check out the full tutorial to build this app in our blog post", icon="📃")
from llama_index.core import PromptTemplate
prompt_template = """### System: Use the following pieces of information to answer the user's question.
If you don't know the answer, just say that you don't know, don't try to make up an answer.
Only return the helpful answer below and nothing else.
Helpful answer:
"""
if "messages" not in st.session_state.keys(): # Initialize the chat messages history
st.session_state.messages = [
{"role": "assistant", "content": "Ask me a question about the E-books!"}
]
PERSIST_DIR = "./storage"
@st.cache_resource(show_spinner=False)
def load_data():
with st.spinner(text="Loading and indexing the E-books – hang tight! This should take 1-2 minutes."):
reader = SimpleDirectoryReader(input_dir="./data", recursive=True)
docs = reader.load_data()
llm = HuggingFaceInferenceAPI(
generate_kwargs={"temperature": 0.0},
model_name="meta-llama/Llama-2-70b-chat-hf",
)
model_name = "BAAI/bge-large-en"
model_kwargs = {'device': 'cpu'}
encode_kwargs = {'normalize_embeddings': False}
embed_model = HuggingFaceBgeEmbeddings(
model_name=model_name,
model_kwargs=model_kwargs,
encode_kwargs=encode_kwargs
)
service_context=ServiceContext.from_defaults(
chunk_size=1000,
chunk_overlap=100,
embed_model=embed_model,
llm=llm
)
set_global_service_context(service_context)
if not os.path.exists(PERSIST_DIR):
reader = SimpleDirectoryReader(input_dir="./data", recursive=True)
Settings.llm = llm
Settings.embed_model = embed_model
docs = reader.load_data()
index = VectorStoreIndex.from_documents(documents=docs, service_context=service_context)
index.storage_context.persist(persist_dir=PERSIST_DIR)
st.write("LoadEmbedding>>>", index)
return index
else:
load the existing index
Settings.llm = llm
Settings.embed_model = embed_model
storage_context = StorageContext.from_defaults(persist_dir=PERSIST_DIR)
index = load_index_from_storage(storage_context)
st.write("StoredEmbedding>>>", index)
return index
index = load_data()
def generate_text(prompt):
question = ("tell me a story with a lesson?")
qa_prompt_str = (
"Context information is below.\n"
"---------------------\n"
"{context_str}\n"
"---------------------\n"
"Given only the context information and not prior knowledge, "
"answer the question: {query_str}\n"
)
refine_prompt_str = (
"We have the opportunity to refine the original answer "
"(only if needed) with some more context below.\n"
"------------\n"
"{context_msg}\n"
"------------\n"
"Given the new context, refine the original answer to better "
"answer the question: {query_str}. "
"If the context isn't useful, output the original answer again.\n"
"Original Answer: {existing_answer}"
)
chat_text_qa_msgs = [
ChatMessage(
role=MessageRole.SYSTEM,
content=(
prompt_template
),
),
ChatMessage(
role=MessageRole.USER,
content=(
qa_prompt_str
),
),
]
text_qa_template = ChatPromptTemplate(chat_text_qa_msgs)
Refine Prompt
chat_refine_msgs = [
ChatMessage(
role=MessageRole.SYSTEM,
content=(
"If the context isn't helpful, just say I don't know. Don't any add informtion into the answer that is not available in the context"
),
),
ChatMessage(
role=MessageRole.USER,
content=(
"New Context: {context_msg}\n"
"Query: {query_str}\n"
"Original Answer: {existing_answer}\n"
"New Answer: "
),
),
]
refine_template = ChatPromptTemplate(chat_refine_msgs)
custom_prompt = PromptTemplate(
"""
Given a conversation (between Human and Assistant) and a follow up message from Human,
rewrite the message to be a standalone question that captures all relevant context
from the conversation.
{chat_history} {question} """ )
list of ChatMessage objects
custom_chat_history = [
ChatMessage(
role=MessageRole.USER,
content="Hello assistant, we are having an insightful discussion about the given content and you are helping me understand the content by answering or summerizing and explaining me the content without changing its true meaning.",
),
ChatMessage(role=MessageRole.ASSISTANT, content="Okay, sounds good."),
]
if "chat_engine" not in st.session_state.keys(): # Initialize the chat engine
if prompt := st.chat_input("Your question"): # Prompt for user input and save to chat history
st.session_state.messages.append({"role": "user", "content": prompt})
for message in st.session_state.messages: # Display the prior chat messages
with st.chat_message(message["role"]):
st.write(message["content"])
If last message is not from assistant, generate a new response
if st.session_state.messages[-1]["role"] != "assistant":
with st.chat_message("assistant"):
with st.spinner("Thinking..."):
response = st.session_state.chat_engine.chat(prompt)
response = st.session_state.chat_engine.stream_chat(prompt,chat_engine.chat_history)
st.write(response.response)
message = {"role": "assistant", "content": response.response}
st.session_state.messages.append(message) # Add response to message history`
I am now sharing some of the examples:-
here i asked about Mumbai and a story knowledge about both of them is given by me in the form of pdf.
image
as you can see it answered both the 1 and 2 question correctly. But the for the 3rd question it was querring a wrong question all together. It was querring with this:-
image
and i don't know why it is spamming the queries as if they are in some kind of a loop.
I know there are lot of flaws in this code but please understand i am new to AI/ML so pls kindly help!!!
I think the project is near completion just few adjustment is needed just confused of what should i do your help would be much appreciated.
Beta Was this translation helpful? Give feedback.
All reactions