File size: 8,701 Bytes
7e40afd |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 |
import torch # Add missing import
import streamlit as st
import os
import tempfile
from langchain_community.document_loaders import (
TextLoader,
CSVLoader,
UnstructuredFileLoader
)
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.retrievers import BM25Retriever
from langchain.retrievers import EnsembleRetriever
from transformers import (
AutoTokenizer,
AutoModelForCausalLM,
BitsAndBytesConfig,
pipeline
)
# Configuration
MODEL_NAME = "mistralai/Mistral-7B-Instruct-v0.3"
EMBEDDING_MODEL = "thenlper/gte-large"
CHUNK_SIZE = 1024
CHUNK_OVERLAP = 128
MAX_NEW_TOKENS = 2048
# Initialize session state
if "messages" not in st.session_state:
st.session_state.messages = []
@st.cache_resource
def initialize_model():
quantization_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_compute_dtype=torch.float16,
bnb_4bit_quant_type="nf4",
bnb_4bit_use_double_quant=True
)
# Load config first to modify RoPE params
from transformers import AutoConfig
config = AutoConfig.from_pretrained(
MODEL_NAME,
trust_remote_code=True
)
# Fix RoPE scaling configuration
if hasattr(config, "rope_scaling"):
config.rope_scaling = {
"type": config.rope_scaling.get("rope_type", "linear"),
"factor": config.rope_scaling.get("factor", 8.0)
}
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(
MODEL_NAME,
config=config,
quantization_config=quantization_config,
device_map="auto",
trust_remote_code=True
)
return pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
device_map="auto",
max_new_tokens=MAX_NEW_TOKENS,
temperature=0.1
)
def process_uploaded_files(uploaded_files):
documents = []
with tempfile.TemporaryDirectory() as temp_dir:
for file in uploaded_files:
temp_path = os.path.join(temp_dir, file.name)
with open(temp_path, "wb") as f:
f.write(file.getbuffer())
try:
if file.name.endswith(".txt"):
loader = TextLoader(temp_path)
elif file.name.endswith(".csv"):
loader = CSVLoader(temp_path)
else:
loader = UnstructuredFileLoader(temp_path)
documents.extend(loader.load())
except Exception as e:
st.error(f"Error loading {file.name}: {str(e)}")
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=CHUNK_SIZE,
chunk_overlap=CHUNK_OVERLAP,
length_function=len
)
return text_splitter.split_documents(documents)
def create_retriever(documents):
embeddings = HuggingFaceEmbeddings(
model_name=EMBEDDING_MODEL,
model_kwargs={'device': 'cuda'},
encode_kwargs={'normalize_embeddings': True}
)
bm25_retriever = BM25Retriever.from_documents(documents)
bm25_retriever.k = st.session_state.get("top_k", 5)
return EnsembleRetriever(
retrievers=[bm25_retriever],
weights=[0.5]
)
def generate_response(query, retriever, generator):
docs = retriever.get_relevant_documents(query)
context = "\n\n".join(
f"[Doc{i+1}] {doc.page_content}\nSource: {doc.metadata.get('source', 'unknown')}"
for i, doc in enumerate(docs)
)
prompt = f"""<s>[INST] You are a precision-focused research assistant tasked with answering queries based solely on the provided context.
**Context:**
{context}
**Query:**
{query}
**Response Instructions:**
- Write a detailed, coherent, and insightful article that fully addresses the query based on the provided context.
- Adhere to the following principles:
1. **Define the Core Subject**: Introduce and build the discussion logically around the main topic.
2. **Establish Connections**: Highlight relationships between ideas and concepts with reasoning and examples.
3. **Elaborate on Key Points**: Provide in-depth explanations and emphasize the significance of concepts.
4. **Maintain Objectivity**: Use only the context provided, avoiding speculation or external knowledge.
5. **Ensure Structure and Clarity**: Present information sequentially for a smooth narrative flow.
6. **Engage with Content**: Explore implicit meanings, resolve doubts, and address counterpoints logically.
7. **Provide Examples and Insights**: Use examples to clarify abstract ideas and offer actionable steps if applicable.
8. **Logical Depth**: Draw inferences, explain purposes, and refute opposing ideas when necessary.
- Cite sources explicitly as [Doc1], [Doc2], etc.
- If uncertain, state: "I cannot determine from the provided context."
Craft the response as a seamless, thorough, and authoritative explanation that naturally integrates all aspects of the query. [/INST]"""
response = generator(
prompt,
pad_token_id=generator.tokenizer.eos_token_id,
do_sample=True
)[0]['generated_text']
return response.split("[/INST]")[-1].strip(), docs
# def generate_response(query, retriever, generator):
# docs = retriever.get_relevant_documents(query)
# context = "\n\n".join(
# f"[Doc{i+1}] {doc.page_content}\nSource: {doc.metadata.get('source', 'unknown')}"
# for i, doc in enumerate(docs)
# )
# prompt = f"""<s>[INST] You are a precise research assistant. Use ONLY the provided context:
# {context}
# Question: {query}
# Answer with:
# 1. Direct facts from context
# 2. NO speculation
# 3. Cite sources like [Doc1]
# 4. If unsure, say "I cannot determine this from the provided data" [/INST]"""
# response = generator(
# prompt,
# pad_token_id=generator.tokenizer.eos_token_id,
# do_sample=True
# )[0]['generated_text']
# return response.split("[/INST]")[-1].strip(), docs
# Streamlit UI
st.title("π Document-Based QA Assistant")
st.markdown("Upload your documents and ask questions!")
# Sidebar controls
with st.sidebar:
st.header("Configuration")
uploaded_files = st.file_uploader(
"Upload documents (TXT)",
type=["txt", "csv"],
accept_multiple_files=True
)
st.session_state.top_k = st.slider("Number of documents to retrieve", 3, 10, 5)
st.markdown("---")
st.markdown("Powered by Mistral-7B and LangChain")
# Main chat interface
for message in st.session_state.messages:
with st.chat_message(message["role"]):
st.markdown(message["content"])
if "sources" in message:
with st.expander("View Sources"):
for i, doc in enumerate(message["sources"]):
st.markdown(f"**Doc{i+1}** ({doc.metadata.get('source', 'unknown')})")
st.info(doc.page_content)
# Process documents
if uploaded_files and "retriever" not in st.session_state:
with st.spinner("Processing documents..."):
documents = process_uploaded_files(uploaded_files)
st.session_state.retriever = create_retriever(documents)
st.session_state.generator = initialize_model()
if prompt := st.chat_input("Ask a question about your documents"):
# Add user message
st.session_state.messages.append({"role": "user", "content": prompt})
with st.chat_message("user"):
st.markdown(prompt)
# Generate response
if "retriever" not in st.session_state:
st.error("Please upload documents first!")
st.stop()
with st.spinner("Analyzing documents..."):
try:
response, sources = generate_response(
prompt,
st.session_state.retriever,
st.session_state.generator
)
# Add assistant response
st.session_state.messages.append({
"role": "assistant",
"content": response,
"sources": sources
})
# Display response
with st.chat_message("assistant"):
st.markdown(response)
with st.expander("View Document Sources"):
for i, doc in enumerate(sources):
st.markdown(f"**Doc{i+1}** ({doc.metadata.get('source', 'unknown')})")
st.info(doc.page_content)
except Exception as e:
st.error(f"Error generating response: {str(e)}") |