Spaces:
Sleeping
Sleeping
File size: 1,836 Bytes
77f3883 aacac9b 77f3883 75ebc6b aacac9b 77f3883 75ebc6b 77f3883 aacac9b 77f3883 75ebc6b 77f3883 75ebc6b 77f3883 75ebc6b 77f3883 75ebc6b 77f3883 75ebc6b 77f3883 75ebc6b 77f3883 aacac9b |
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 |
from pathlib import Path
from langchain.chains import RetrievalQA
from transformers import pipeline, AutoTokenizer
from langchain_community.vectorstores import Chroma
from langchain_community.document_loaders import PyMuPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEmbeddings, HuggingFacePipeline
# Load all PDFs from the data folder
def load_documents(pdf_dir):
docs = []
for pdf_file in Path(pdf_dir).glob("*.pdf"):
loader = PyMuPDFLoader(str(pdf_file))
docs.extend(loader.load())
return docs
def load_rag_chain():
# Make sure the data directory exists
pdf_dir = Path("data")
pdf_dir.mkdir(parents=True, exist_ok=True)
# Load and split PDFs
raw_docs = load_documents(pdf_dir)
splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
pages = splitter.split_documents(raw_docs)
# Load sentence transformer for embeddings
embeddings = HuggingFaceEmbeddings(
model_name="sentence-transformers/all-MiniLM-L6-v2",
model_kwargs={"device": "cpu"},
)
# Vector store
vectordb_dir = "chroma_db"
vectordb = Chroma.from_documents(pages, embeddings, persist_directory=vectordb_dir)
retriever = vectordb.as_retriever(search_type="mmr", search_kwargs={"k": 5})
# Load a completely free and CPU-compatible model
hf_pipeline = pipeline(
"text2text-generation",
model="google/flan-t5-base",
tokenizer=AutoTokenizer.from_pretrained("google/flan-t5-base"),
max_new_tokens=512,
temperature=0.3,
device=-1 # -1 means CPU
)
llm = HuggingFacePipeline(pipeline=hf_pipeline)
# Build RetrievalQA chain
qa_chain = RetrievalQA.from_llm(llm=llm, retriever=retriever)
return qa_chain
|