File size: 5,169 Bytes
0ff7c7e 5b00cc2 38a4055 0ff7c7e 5b00cc2 0ff7c7e c3d7129 0ff7c7e c3d7129 4dc8cac c3d7129 5b00cc2 c3d7129 5b00cc2 38a4055 5b00cc2 c3d7129 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 |
import time
import streamlit as st
from llama_index import ServiceContext, StorageContext, set_global_service_context, VectorStoreIndex, Document
from llama_index.prompts import PromptTemplate
from llama_index.embeddings import LangchainEmbedding
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
from llama_index.chat_engine.condense_question import CondenseQuestionChatEngine
from llama_index.llms import LlamaCPP
from llama_index.llms.llama_utils import messages_to_prompt, completion_to_prompt
from PyPDF2 import PdfReader
# Model Specific Prompt
def modelspecific_prompt(promptmessage):
# Model Specific Prompt
# As per https://huggingface.co/TheBloke/phi-2-GGUF
return f"Instruct: {promptmessage}\nOutput:"
# PDF to Text
def extract_text_from_pdf(pdf):
pdf_reader = PdfReader(pdf)
data = ''.join(page.extract_text() for page in pdf_reader.pages)
data = data.split()
return data
def main():
# LLM Intialization
llm = LlamaCPP(
model_url=None, # We'll load locally.
# Trying small version of an already small model
model_path='models/phi-2.Q4_K_M.gguf',
temperature=0.1,
max_new_tokens=512,
context_window=2048, # Phi-2 2K context window - this could be a limitation for RAG as it has to put the content into this context window
generate_kwargs={},
# set to at least 1 to use GPU
# This is small model and there's no indication of layers offloaded to the GPU
model_kwargs={"n_gpu_layers": 32},
messages_to_prompt=messages_to_prompt,
completion_to_prompt=completion_to_prompt,
verbose=True
)
# Embedding Initialization
embed_model = LangchainEmbedding(
HuggingFaceEmbeddings(model_name="BAAI/bge-small-en-v1.5")
)
# Service Context
service_context = ServiceContext.from_defaults(
chunk_size=128, # Number of tokens in each chunk
chunk_overlap=20,
# This should be automatically set with the model metadata but we'll force it to ensure wit is
context_window=2048,
num_output=768, # Maximum output from the LLM, let's put this at 512 to ensure LlamaIndex saves that "space" for the output
llm=llm,
embed_model=embed_model
)
set_global_service_context(service_context)
# Storage Context
storage_context = StorageContext.from_defaults()
st.title("Llama-CPP Local LLM with RAG (Phi-2 RAG)")
# Credits
st.markdown(
"Made with ❤️️ By Danyaal Majid & Muhammad Bin Asif Using [HF Spaces](https://huggingface.co/spaces/DanyaalMajid/NLP-Final-LocalLLM-RAG)")
pdf = st.file_uploader("Upload a PDF file", type=["pdf"])
if pdf is not None:
text_list = extract_text_from_pdf(pdf)
documents = [Document(text=t) for t in text_list]
nodes = (service_context.node_parser.get_nodes_from_documents(documents))
storage_context.docstore.add_documents(nodes)
index = (VectorStoreIndex.from_documents(
documents, service_context=service_context, storage_context=storage_context, llm=llm))
# chat_engine = index.as_chat_engine(chat_mode="simple", verbose=True)
custom_prompt = PromptTemplate("")
query_engine = index.as_query_engine()
chat_engine = CondenseQuestionChatEngine.from_defaults(
query_engine=query_engine,
condense_question_prompt=custom_prompt,
verbose=True,
)
# Initialize chat history
if "messages" not in st.session_state:
st.session_state.messages = []
# Display chat messages from history on app rerun
for message in st.session_state.messages:
with st.chat_message(message["role"]):
st.markdown(message["content"])
# Accept user input
if prompt := st.chat_input("What is up?"):
# Add user message to chat history
st.session_state.messages.append(
{"role": "user", "content": prompt})
# Display user message in chat message container
with st.chat_message("user"):
st.markdown(prompt)
# Display assistant response in chat message container
with st.chat_message("assistant"):
message_placeholder = st.empty()
full_response = ""
assistant_response = chat_engine.chat(
modelspecific_prompt(str(prompt)))
assistant_response = str(assistant_response)
# Simulate stream of response with milliseconds delay
for chunk in assistant_response.split():
full_response += chunk + " "
time.sleep(0.05)
# Add a blinking cursor to simulate typing
message_placeholder.markdown(full_response + "▌")
message_placeholder.markdown(full_response)
# Add assistant response to chat history
st.session_state.messages.append(
{"role": "assistant", "content": full_response})
if __name__ == "__main__":
main()
|