Spaces:
Running
Running
halimbahae
commited on
Commit
•
a881166
1
Parent(s):
1335f41
Update app.py
Browse files
app.py
CHANGED
@@ -1,30 +1,20 @@
|
|
1 |
import os
|
2 |
import gradio as gr
|
3 |
from huggingface_hub import InferenceClient
|
|
|
|
|
|
|
4 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
5 |
-
from langchain.vectorstores import Chroma
|
6 |
-
from langchain.embeddings import HuggingFaceBgeEmbeddings
|
7 |
-
from langchain.document_loaders import PyPDFLoader, UnstructuredFileLoader, CSVLoader
|
8 |
-
from langchain.chains import RetrievalQA
|
9 |
-
from langchain.prompts import PromptTemplate
|
10 |
|
11 |
-
#
|
12 |
-
# client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
13 |
-
|
14 |
-
from huggingface_hub import InferenceClient
|
15 |
-
|
16 |
-
# Access the Hugging Face token from environment variables
|
17 |
HF_API_TOKEN = os.getenv("HF_API_TOKEN")
|
18 |
-
|
19 |
if not HF_API_TOKEN:
|
20 |
raise ValueError("Hugging Face API token is not set in environment variables.")
|
21 |
|
22 |
-
# Initialize
|
23 |
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta", token=HF_API_TOKEN)
|
24 |
|
25 |
-
|
26 |
-
|
27 |
-
# Function to load documents based on file type
|
28 |
def load_documents(file_path):
|
29 |
if file_path.endswith(".pdf"):
|
30 |
loader = PyPDFLoader(file_path)
|
@@ -34,54 +24,50 @@ def load_documents(file_path):
|
|
34 |
loader = CSVLoader(file_path)
|
35 |
else:
|
36 |
raise ValueError("Unsupported file format")
|
37 |
-
|
38 |
documents = loader.load()
|
39 |
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
|
40 |
return text_splitter.split_documents(documents)
|
41 |
|
42 |
-
#
|
43 |
def create_vector_store(documents, persist_dir="vector_db"):
|
44 |
embeddings = HuggingFaceBgeEmbeddings(
|
45 |
-
model_name="BAAI/bge-large-en",
|
46 |
-
model_kwargs={"device": "cpu"}
|
47 |
)
|
48 |
vector_store = Chroma.from_documents(documents, embeddings, persist_directory=persist_dir)
|
49 |
return vector_store
|
50 |
|
51 |
-
#
|
52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
# Retrieve relevant context
|
54 |
relevant_docs = retriever.get_relevant_documents(message)
|
55 |
context = "\n".join([doc.page_content for doc in relevant_docs])
|
56 |
-
|
57 |
-
#
|
58 |
-
|
59 |
-
Use the following context to answer the user's question.
|
60 |
-
If you don't know the answer, say "I don't know."
|
61 |
-
|
62 |
Context:
|
63 |
{context}
|
64 |
-
|
65 |
Question:
|
66 |
-
{
|
67 |
-
|
68 |
-
|
69 |
-
"""
|
70 |
-
formatted_prompt = prompt_template.format(context=context, question=message)
|
71 |
-
|
72 |
-
# Build conversational history
|
73 |
-
messages = [{"role": "system", "content": system_message}]
|
74 |
-
for val in history:
|
75 |
-
if val[0]:
|
76 |
-
messages.append({"role": "user", "content": val[0]})
|
77 |
-
if val[1]:
|
78 |
-
messages.append({"role": "assistant", "content": val[1]})
|
79 |
-
messages.append({"role": "user", "content": formatted_prompt})
|
80 |
-
|
81 |
-
# Stream response from Zephyr
|
82 |
response = ""
|
83 |
for msg in client.chat_completion(
|
84 |
-
messages=
|
85 |
max_tokens=max_tokens,
|
86 |
stream=True,
|
87 |
temperature=temperature,
|
@@ -91,37 +77,9 @@ def respond(message, history, system_message, max_tokens, temperature, top_p, re
|
|
91 |
response += token
|
92 |
yield response
|
93 |
|
94 |
-
# Initialize the vector store
|
95 |
-
persist_dir = "vector_db"
|
96 |
-
retriever = None # Will be initialized dynamically
|
97 |
-
|
98 |
-
def handle_query(message, history, system_message, max_tokens, temperature, top_p, file=None):
|
99 |
-
global retriever
|
100 |
-
if file: # Process uploaded file
|
101 |
-
documents = load_documents(file.name)
|
102 |
-
vector_store = create_vector_store(documents, persist_dir)
|
103 |
-
retriever = vector_store.as_retriever()
|
104 |
-
if not retriever:
|
105 |
-
return "No documents have been uploaded yet. Please upload a file to provide context."
|
106 |
-
return respond(message, history, system_message, max_tokens, temperature, top_p, retriever)
|
107 |
-
|
108 |
# Gradio app setup
|
109 |
-
# demo = gr.ChatInterface(
|
110 |
-
# fn=handle_query,
|
111 |
-
# additional_inputs=[
|
112 |
-
# gr.File(label="Upload File", type="file"),
|
113 |
-
# gr.Textbox(value="You are a knowledgeable assistant.", label="System Message"),
|
114 |
-
# gr.Slider(1, 2048, step=1, value=512, label="Max Tokens"),
|
115 |
-
# gr.Slider(0.1, 4.0, step=0.1, value=0.7, label="Temperature"),
|
116 |
-
# gr.Slider(0.1, 1.0, step=0.05, value=0.95, label="Top-p"),
|
117 |
-
# ],
|
118 |
-
# outputs="text",
|
119 |
-
# title="RAG with Zephyr-7B",
|
120 |
-
# description="A Retrieval-Augmented Generation chatbot powered by Zephyr-7B and Chroma vector database.",
|
121 |
-
# )
|
122 |
-
|
123 |
demo = gr.Interface(
|
124 |
-
fn=
|
125 |
inputs=[
|
126 |
gr.File(label="Upload Document"),
|
127 |
gr.Textbox(value="You are a knowledgeable assistant.", label="System Message"),
|
@@ -135,7 +93,5 @@ demo = gr.Interface(
|
|
135 |
description="Upload documents and ask questions using RAG.",
|
136 |
)
|
137 |
|
138 |
-
|
139 |
-
|
140 |
if __name__ == "__main__":
|
141 |
demo.launch()
|
|
|
1 |
import os
|
2 |
import gradio as gr
|
3 |
from huggingface_hub import InferenceClient
|
4 |
+
from langchain_community.vectorstores import Chroma
|
5 |
+
from langchain_community.embeddings import HuggingFaceBgeEmbeddings
|
6 |
+
from langchain_community.document_loaders import PyPDFLoader, UnstructuredFileLoader, CSVLoader
|
7 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
+
# Load Hugging Face API token
|
|
|
|
|
|
|
|
|
|
|
10 |
HF_API_TOKEN = os.getenv("HF_API_TOKEN")
|
|
|
11 |
if not HF_API_TOKEN:
|
12 |
raise ValueError("Hugging Face API token is not set in environment variables.")
|
13 |
|
14 |
+
# Initialize Zephyr client
|
15 |
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta", token=HF_API_TOKEN)
|
16 |
|
17 |
+
# Load documents based on file type
|
|
|
|
|
18 |
def load_documents(file_path):
|
19 |
if file_path.endswith(".pdf"):
|
20 |
loader = PyPDFLoader(file_path)
|
|
|
24 |
loader = CSVLoader(file_path)
|
25 |
else:
|
26 |
raise ValueError("Unsupported file format")
|
|
|
27 |
documents = loader.load()
|
28 |
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
|
29 |
return text_splitter.split_documents(documents)
|
30 |
|
31 |
+
# Create vector store
|
32 |
def create_vector_store(documents, persist_dir="vector_db"):
|
33 |
embeddings = HuggingFaceBgeEmbeddings(
|
34 |
+
model_name="BAAI/bge-large-en",
|
35 |
+
model_kwargs={"device": "cpu"},
|
36 |
)
|
37 |
vector_store = Chroma.from_documents(documents, embeddings, persist_directory=persist_dir)
|
38 |
return vector_store
|
39 |
|
40 |
+
# Initialize retriever and vector store
|
41 |
+
persist_dir = "vector_db"
|
42 |
+
retriever = None # Will be dynamically updated
|
43 |
+
|
44 |
+
# Handle queries and uploads
|
45 |
+
def handle_query(message, history, system_message, max_tokens, temperature, top_p, file=None):
|
46 |
+
global retriever
|
47 |
+
if file: # If a file is uploaded, process it
|
48 |
+
documents = load_documents(file.name)
|
49 |
+
vector_store = create_vector_store(documents, persist_dir)
|
50 |
+
retriever = vector_store.as_retriever()
|
51 |
+
if not retriever:
|
52 |
+
return "No documents uploaded yet. Please upload a file first."
|
53 |
+
|
54 |
# Retrieve relevant context
|
55 |
relevant_docs = retriever.get_relevant_documents(message)
|
56 |
context = "\n".join([doc.page_content for doc in relevant_docs])
|
57 |
+
|
58 |
+
# Build the prompt
|
59 |
+
prompt = f"""
|
60 |
+
Use the following context to answer the user's question.
|
|
|
|
|
61 |
Context:
|
62 |
{context}
|
63 |
+
|
64 |
Question:
|
65 |
+
{message}
|
66 |
+
Answer:"""
|
67 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
response = ""
|
69 |
for msg in client.chat_completion(
|
70 |
+
messages=[{"role": "system", "content": system_message}, {"role": "user", "content": prompt}],
|
71 |
max_tokens=max_tokens,
|
72 |
stream=True,
|
73 |
temperature=temperature,
|
|
|
77 |
response += token
|
78 |
yield response
|
79 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
80 |
# Gradio app setup
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
81 |
demo = gr.Interface(
|
82 |
+
fn=handle_query,
|
83 |
inputs=[
|
84 |
gr.File(label="Upload Document"),
|
85 |
gr.Textbox(value="You are a knowledgeable assistant.", label="System Message"),
|
|
|
93 |
description="Upload documents and ask questions using RAG.",
|
94 |
)
|
95 |
|
|
|
|
|
96 |
if __name__ == "__main__":
|
97 |
demo.launch()
|