Spaces:
Runtime error
Runtime error
Update worker.py
Browse files
worker.py
CHANGED
@@ -1,11 +1,12 @@
|
|
1 |
import os
|
2 |
import torch
|
3 |
-
from langchain.chains import
|
4 |
from langchain_community.embeddings import HuggingFaceEmbeddings
|
5 |
from langchain_community.document_loaders import PyPDFLoader
|
6 |
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
7 |
from langchain_community.vectorstores import Chroma
|
8 |
-
from langchain_community.llms import
|
|
|
9 |
|
10 |
# Check for GPU availability
|
11 |
DEVICE = "cuda:0" if torch.cuda.is_available() else "cpu"
|
@@ -13,11 +14,12 @@ DEVICE = "cuda:0" if torch.cuda.is_available() else "cpu"
|
|
13 |
# Global variables
|
14 |
conversation_retrieval_chain = None
|
15 |
chat_history = []
|
16 |
-
|
17 |
embeddings = None
|
18 |
|
|
|
19 |
def init_llm():
|
20 |
-
global
|
21 |
|
22 |
# Ensure API key is set in Hugging Face Spaces
|
23 |
hf_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
|
@@ -25,17 +27,19 @@ def init_llm():
|
|
25 |
raise ValueError("HUGGINGFACEHUB_API_TOKEN is not set in environment variables.")
|
26 |
|
27 |
model_id = "tiiuae/falcon-7b-instruct"
|
28 |
-
|
|
|
29 |
|
30 |
embeddings = HuggingFaceEmbeddings(
|
31 |
model_name="sentence-transformers/all-MiniLM-L6-v2", model_kwargs={"device": DEVICE}
|
32 |
)
|
33 |
|
|
|
34 |
def process_document(document_path):
|
35 |
global conversation_retrieval_chain
|
36 |
|
37 |
# Ensure LLM and embeddings are initialized
|
38 |
-
if not
|
39 |
init_llm()
|
40 |
|
41 |
loader = PyPDFLoader(document_path)
|
@@ -44,22 +48,27 @@ def process_document(document_path):
|
|
44 |
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=64)
|
45 |
texts = text_splitter.split_documents(documents)
|
46 |
|
47 |
-
|
|
|
|
|
|
|
|
|
|
|
48 |
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
return_source_documents=False
|
54 |
)
|
55 |
|
|
|
56 |
def process_prompt(prompt):
|
57 |
global conversation_retrieval_chain, chat_history
|
58 |
|
59 |
if not conversation_retrieval_chain:
|
60 |
return "No document has been processed yet. Please upload a PDF first."
|
61 |
|
62 |
-
output = conversation_retrieval_chain({"
|
63 |
answer = output["answer"]
|
64 |
|
65 |
chat_history.append((prompt, answer))
|
|
|
1 |
import os
|
2 |
import torch
|
3 |
+
from langchain.chains import ConversationalRetrievalChain
|
4 |
from langchain_community.embeddings import HuggingFaceEmbeddings
|
5 |
from langchain_community.document_loaders import PyPDFLoader
|
6 |
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
7 |
from langchain_community.vectorstores import Chroma
|
8 |
+
from langchain_community.llms import HuggingFacePipeline
|
9 |
+
from transformers import pipeline
|
10 |
|
11 |
# Check for GPU availability
|
12 |
DEVICE = "cuda:0" if torch.cuda.is_available() else "cpu"
|
|
|
14 |
# Global variables
|
15 |
conversation_retrieval_chain = None
|
16 |
chat_history = []
|
17 |
+
llm_pipeline = None
|
18 |
embeddings = None
|
19 |
|
20 |
+
|
21 |
def init_llm():
|
22 |
+
global llm_pipeline, embeddings
|
23 |
|
24 |
# Ensure API key is set in Hugging Face Spaces
|
25 |
hf_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
|
|
|
27 |
raise ValueError("HUGGINGFACEHUB_API_TOKEN is not set in environment variables.")
|
28 |
|
29 |
model_id = "tiiuae/falcon-7b-instruct"
|
30 |
+
hf_pipeline = pipeline("text-generation", model=model_id, device=DEVICE)
|
31 |
+
llm_pipeline = HuggingFacePipeline(pipeline=hf_pipeline)
|
32 |
|
33 |
embeddings = HuggingFaceEmbeddings(
|
34 |
model_name="sentence-transformers/all-MiniLM-L6-v2", model_kwargs={"device": DEVICE}
|
35 |
)
|
36 |
|
37 |
+
|
38 |
def process_document(document_path):
|
39 |
global conversation_retrieval_chain
|
40 |
|
41 |
# Ensure LLM and embeddings are initialized
|
42 |
+
if not llm_pipeline or not embeddings:
|
43 |
init_llm()
|
44 |
|
45 |
loader = PyPDFLoader(document_path)
|
|
|
48 |
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1024, chunk_overlap=64)
|
49 |
texts = text_splitter.split_documents(documents)
|
50 |
|
51 |
+
# Load or create ChromaDB
|
52 |
+
persist_directory = "./chroma_db"
|
53 |
+
if os.path.exists(persist_directory):
|
54 |
+
db = Chroma(persist_directory=persist_directory, embedding_function=embeddings)
|
55 |
+
else:
|
56 |
+
db = Chroma.from_documents(texts, embedding=embeddings, persist_directory=persist_directory)
|
57 |
|
58 |
+
retriever = db.as_retriever(search_type="similarity", search_kwargs={'k': 6})
|
59 |
+
|
60 |
+
conversation_retrieval_chain = ConversationalRetrievalChain.from_llm(
|
61 |
+
llm=llm_pipeline, retriever=retriever
|
|
|
62 |
)
|
63 |
|
64 |
+
|
65 |
def process_prompt(prompt):
|
66 |
global conversation_retrieval_chain, chat_history
|
67 |
|
68 |
if not conversation_retrieval_chain:
|
69 |
return "No document has been processed yet. Please upload a PDF first."
|
70 |
|
71 |
+
output = conversation_retrieval_chain({"question": prompt, "chat_history": chat_history})
|
72 |
answer = output["answer"]
|
73 |
|
74 |
chat_history.append((prompt, answer))
|