Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,17 +1,24 @@
|
|
1 |
import streamlit as st
|
2 |
from openai import OpenAI
|
3 |
-
import torch
|
4 |
import os
|
5 |
-
import
|
6 |
-
from dotenv import load_dotenv, dotenv_values
|
7 |
import numpy as np
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
load_dotenv()
|
10 |
|
11 |
-
# Initialize the client
|
12 |
client = OpenAI(
|
13 |
base_url="https://api-inference.huggingface.co/v1",
|
14 |
-
api_key=os.environ.get('HUGGINGFACEHUB_API_TOKEN')
|
15 |
)
|
16 |
|
17 |
# Create supported models
|
@@ -21,7 +28,6 @@ model_links = {
|
|
21 |
"gemma-2-2b": "google/gemma-2-2b",
|
22 |
}
|
23 |
|
24 |
-
# Pull info about the model to display
|
25 |
model_info = {
|
26 |
"Meta-Llama-3.1-8B": {
|
27 |
'description': """The Llama (3.1) model is a **Large Language Model (LLM)** that's able to have question and answer interactions.
|
@@ -49,78 +55,117 @@ models = [key for key in model_links.keys()]
|
|
49 |
# Create the sidebar with the dropdown for model selection
|
50 |
selected_model = st.sidebar.selectbox("Select Model", models)
|
51 |
|
52 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
temp_values = st.sidebar.slider('Select a temperature value', 0.0, 1.0, 0.5)
|
54 |
|
55 |
-
#
|
56 |
st.sidebar.write(f"You're now chatting with **{selected_model}**")
|
57 |
st.sidebar.markdown(model_info[selected_model]['description'])
|
58 |
st.sidebar.image(model_info[selected_model]['logo'])
|
59 |
st.sidebar.markdown("*Generated content may be inaccurate or false.*")
|
60 |
|
61 |
-
if "prev_option" not in st.session_state:
|
62 |
-
st.session_state.prev_option = selected_model
|
63 |
-
|
64 |
-
if st.session_state.prev_option != selected_model:
|
65 |
-
st.session_state.messages = []
|
66 |
-
st.session_state.prev_option = selected_model
|
67 |
-
|
68 |
-
# Pull in the model we want to use
|
69 |
-
repo_id = model_links[selected_model]
|
70 |
-
|
71 |
-
st.header('Liahona.AI')
|
72 |
-
st.markdown(f'_powered_ by ***:violet[{selected_model}]***')
|
73 |
-
|
74 |
-
# Set a default model
|
75 |
-
if selected_model not in st.session_state:
|
76 |
-
st.session_state[selected_model] = model_links[selected_model]
|
77 |
-
|
78 |
# Initialize chat history
|
79 |
if "messages" not in st.session_state:
|
80 |
st.session_state.messages = []
|
81 |
|
82 |
-
# Display chat messages from history
|
83 |
for message in st.session_state.messages:
|
84 |
with st.chat_message(message["role"]):
|
85 |
st.markdown(message["content"])
|
86 |
|
87 |
-
#
|
88 |
-
|
89 |
|
90 |
-
|
|
|
|
|
91 |
with st.chat_message("user"):
|
92 |
st.markdown(prompt)
|
93 |
-
# Add user message to chat history
|
94 |
st.session_state.messages.append({"role": "user", "content": prompt})
|
95 |
|
96 |
-
#
|
97 |
with st.chat_message("assistant"):
|
98 |
-
|
99 |
try:
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
)
|
110 |
-
|
111 |
-
response = st.write_stream(stream)
|
112 |
-
|
113 |
except Exception as e:
|
114 |
response = """π΅βπ« Looks like someone unplugged something!
|
115 |
\n Either the model space is being updated or something is down.
|
116 |
-
\n
|
117 |
-
\n Try again later.
|
118 |
-
\n
|
119 |
-
\n Here's a random pic of a πΆ:"""
|
120 |
st.write(response)
|
121 |
random_dog_pick = 'https://random.dog/' + random_dog[np.random.randint(len(random_dog))]
|
122 |
st.image(random_dog_pick)
|
123 |
st.write("This was the error message:")
|
124 |
-
st.write(e)
|
125 |
|
126 |
st.session_state.messages.append({"role": "assistant", "content": response})
|
|
|
1 |
import streamlit as st
|
2 |
from openai import OpenAI
|
|
|
3 |
import os
|
4 |
+
from dotenv import load_dotenv
|
|
|
5 |
import numpy as np
|
6 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
7 |
+
from langchain.schema import Document
|
8 |
+
from langchain_community.llms import HuggingFaceHub
|
9 |
+
from langchain.chains import RetrievalQA, LLMChain
|
10 |
+
from langchain.prompts import PromptTemplate
|
11 |
+
from langchain_community.embeddings import HuggingFaceInstructEmbeddings
|
12 |
+
from langchain_community.vectorstores import Chroma
|
13 |
+
from langchain.retrievers import ContextualCompressionRetriever
|
14 |
+
from langchain.retrievers.document_compressors import LLMChainExtractor
|
15 |
|
16 |
load_dotenv()
|
17 |
|
18 |
+
# Initialize the OpenAI client for Hugging Face
|
19 |
client = OpenAI(
|
20 |
base_url="https://api-inference.huggingface.co/v1",
|
21 |
+
api_key=os.environ.get('HUGGINGFACEHUB_API_TOKEN')
|
22 |
)
|
23 |
|
24 |
# Create supported models
|
|
|
28 |
"gemma-2-2b": "google/gemma-2-2b",
|
29 |
}
|
30 |
|
|
|
31 |
model_info = {
|
32 |
"Meta-Llama-3.1-8B": {
|
33 |
'description': """The Llama (3.1) model is a **Large Language Model (LLM)** that's able to have question and answer interactions.
|
|
|
55 |
# Create the sidebar with the dropdown for model selection
|
56 |
selected_model = st.sidebar.selectbox("Select Model", models)
|
57 |
|
58 |
+
# Function to load and process documents
|
59 |
+
def load_and_process_documents(file_path):
|
60 |
+
with open(file_path, 'r') as file:
|
61 |
+
content = file.read()
|
62 |
+
|
63 |
+
doc = Document(page_content=content, metadata={"source": file_path})
|
64 |
+
|
65 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=8192, chunk_overlap=200)
|
66 |
+
splits = text_splitter.split_documents([doc])
|
67 |
+
|
68 |
+
return splits
|
69 |
+
|
70 |
+
# Function to set up the advanced RAG pipeline
|
71 |
+
@st.cache_resource
|
72 |
+
def setup_advanced_rag_pipeline(model_name):
|
73 |
+
# Load and process documents
|
74 |
+
splits = load_and_process_documents("index_training.json") # Replace with your document path
|
75 |
+
|
76 |
+
# Set up InstructorEmbeddings
|
77 |
+
embeddings = HuggingFaceInstructEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
78 |
+
|
79 |
+
# Create vectorstore
|
80 |
+
vectorstore = Chroma.from_documents(documents=splits, embedding=embeddings)
|
81 |
+
|
82 |
+
# Set up language model
|
83 |
+
llm = HuggingFaceHub(repo_id=model_links[model_name], model_kwargs={"temperature": 0.5, "max_length": 4000})
|
84 |
+
|
85 |
+
# Set up HyDE
|
86 |
+
hyde_prompt = PromptTemplate(
|
87 |
+
input_variables=["question"],
|
88 |
+
template="Please write a passage to answer the question\nQuestion: {question}\nPassage:"
|
89 |
+
)
|
90 |
+
hyde_chain = LLMChain(llm=llm, prompt=hyde_prompt)
|
91 |
+
|
92 |
+
def hyde_retriever(query):
|
93 |
+
hypothetical_doc = hyde_chain.run(query)
|
94 |
+
hyde_embedding = embeddings.embed_query(hypothetical_doc)
|
95 |
+
return vectorstore.similarity_search_by_vector(hyde_embedding, k=3)
|
96 |
+
|
97 |
+
# Set up ContextualCompressionRetriever
|
98 |
+
compressor = LLMChainExtractor.from_llm(llm)
|
99 |
+
compression_retriever = ContextualCompressionRetriever(
|
100 |
+
base_compressor=compressor,
|
101 |
+
base_retriever=hyde_retriever
|
102 |
+
)
|
103 |
+
|
104 |
+
# Create RetrievalQA chain
|
105 |
+
qa_chain = RetrievalQA.from_chain_type(
|
106 |
+
llm=llm,
|
107 |
+
chain_type="stuff",
|
108 |
+
retriever=compression_retriever,
|
109 |
+
return_source_documents=True
|
110 |
+
)
|
111 |
+
|
112 |
+
return qa_chain
|
113 |
+
|
114 |
+
# Streamlit app
|
115 |
+
st.header('Liahona.AI')
|
116 |
+
|
117 |
+
# Sidebar for model selection
|
118 |
+
selected_model = st.sidebar.selectbox("Select Model", list(model_links.keys()))
|
119 |
+
st.markdown(f'_powered_ by ***:violet[{selected_model}]***')
|
120 |
+
|
121 |
+
# Temperature slider
|
122 |
temp_values = st.sidebar.slider('Select a temperature value', 0.0, 1.0, 0.5)
|
123 |
|
124 |
+
# Display model info
|
125 |
st.sidebar.write(f"You're now chatting with **{selected_model}**")
|
126 |
st.sidebar.markdown(model_info[selected_model]['description'])
|
127 |
st.sidebar.image(model_info[selected_model]['logo'])
|
128 |
st.sidebar.markdown("*Generated content may be inaccurate or false.*")
|
129 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
130 |
# Initialize chat history
|
131 |
if "messages" not in st.session_state:
|
132 |
st.session_state.messages = []
|
133 |
|
134 |
+
# Display chat messages from history
|
135 |
for message in st.session_state.messages:
|
136 |
with st.chat_message(message["role"]):
|
137 |
st.markdown(message["content"])
|
138 |
|
139 |
+
# Set up advanced RAG pipeline
|
140 |
+
qa_chain = setup_advanced_rag_pipeline(selected_model)
|
141 |
|
142 |
+
# Chat input
|
143 |
+
if prompt := st.chat_input("Type message here..."):
|
144 |
+
# Display user message
|
145 |
with st.chat_message("user"):
|
146 |
st.markdown(prompt)
|
|
|
147 |
st.session_state.messages.append({"role": "user", "content": prompt})
|
148 |
|
149 |
+
# Generate and display assistant response
|
150 |
with st.chat_message("assistant"):
|
|
|
151 |
try:
|
152 |
+
result = qa_chain({"query": prompt})
|
153 |
+
response = result["result"]
|
154 |
+
st.write(response)
|
155 |
+
|
156 |
+
# Optionally, display source documents
|
157 |
+
st.expander("View Source Documents"):
|
158 |
+
for doc in result["source_documents"]:
|
159 |
+
st.write(doc.page_content)
|
160 |
+
st.write("---")
|
|
|
|
|
|
|
|
|
161 |
except Exception as e:
|
162 |
response = """π΅βπ« Looks like someone unplugged something!
|
163 |
\n Either the model space is being updated or something is down.
|
164 |
+
\n"""
|
|
|
|
|
|
|
165 |
st.write(response)
|
166 |
random_dog_pick = 'https://random.dog/' + random_dog[np.random.randint(len(random_dog))]
|
167 |
st.image(random_dog_pick)
|
168 |
st.write("This was the error message:")
|
169 |
+
st.write(str(e))
|
170 |
|
171 |
st.session_state.messages.append({"role": "assistant", "content": response})
|