Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,155 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import torch
|
3 |
+
from accelerate import Accelerator
|
4 |
+
from PIL import Image
|
5 |
+
import random
|
6 |
+
import requests
|
7 |
+
import streamlit as st
|
8 |
+
from transformers import BlipProcessor, BlipForConditionalGeneration
|
9 |
+
from langchain_huggingface import HuggingFaceEndpoint
|
10 |
+
from langchain_core.prompts import PromptTemplate
|
11 |
+
from langchain_core.output_parsers import StrOutputParser
|
12 |
+
|
13 |
+
# Define the model IDs
|
14 |
+
llm_model_id = "mistralai/Mistral-7B-Instruct-v0.3"
|
15 |
+
blip_model_id = "Salesforce/blip-image-captioning-large"
|
16 |
+
|
17 |
+
# Initialize BLIP processor and model
|
18 |
+
processor = BlipProcessor.from_pretrained(blip_model_id)
|
19 |
+
model = BlipForConditionalGeneration.from_pretrained(blip_model_id)
|
20 |
+
|
21 |
+
# Initialize the accelerator
|
22 |
+
accelerator = Accelerator()
|
23 |
+
|
24 |
+
def get_llm_hf_inference(model_id=llm_model_id, max_new_tokens=128, temperature=0.1):
|
25 |
+
try:
|
26 |
+
llm = HuggingFaceEndpoint(
|
27 |
+
repo_id=model_id,
|
28 |
+
max_new_tokens=max_new_tokens,
|
29 |
+
temperature=temperature,
|
30 |
+
token=os.getenv("HF_TOKEN")
|
31 |
+
)
|
32 |
+
except Exception as e:
|
33 |
+
st.error(f"Error loading model: {e}")
|
34 |
+
llm = None
|
35 |
+
return llm
|
36 |
+
|
37 |
+
def generate_caption(image, min_len=30, max_len=100):
|
38 |
+
try:
|
39 |
+
inputs = processor(image, return_tensors="pt")
|
40 |
+
out = model.generate(**inputs, min_length=min_len, max_length=max_len)
|
41 |
+
caption = processor.decode(out[0], skip_special_tokens=True)
|
42 |
+
return caption
|
43 |
+
except Exception as e:
|
44 |
+
st.error(f"Error generating caption: {e}")
|
45 |
+
return 'Unable to generate caption.'
|
46 |
+
|
47 |
+
# Configure the Streamlit app
|
48 |
+
st.set_page_config(page_title="HuggingFace ChatBot", page_icon="π€")
|
49 |
+
st.title("Personal HuggingFace ChatBot")
|
50 |
+
st.markdown(f"*This is a simple chatbot using the HuggingFace transformers library with {llm_model_id}.*")
|
51 |
+
|
52 |
+
# Initialize session state
|
53 |
+
if "avatars" not in st.session_state:
|
54 |
+
st.session_state.avatars = {'user': None, 'assistant': None}
|
55 |
+
|
56 |
+
if 'user_text' not in st.session_state:
|
57 |
+
st.session_state.user_text = None
|
58 |
+
|
59 |
+
if "max_response_length" not in st.session_state:
|
60 |
+
st.session_state.max_response_length = 256
|
61 |
+
|
62 |
+
if "system_message" not in st.session_state:
|
63 |
+
st.session_state.system_message = "friendly AI conversing with a human user"
|
64 |
+
|
65 |
+
if "starter_message" not in st.session_state:
|
66 |
+
st.session_state.starter_message = "Hello, there! How can I help you today?"
|
67 |
+
|
68 |
+
if "uploaded_image_path" not in st.session_state:
|
69 |
+
st.session_state.uploaded_image_path = None
|
70 |
+
|
71 |
+
# Sidebar for settings
|
72 |
+
with st.sidebar:
|
73 |
+
st.header("System Settings")
|
74 |
+
st.session_state.system_message = st.text_area(
|
75 |
+
"System Message", value="You are a friendly AI conversing with a human user."
|
76 |
+
)
|
77 |
+
st.session_state.starter_message = st.text_area(
|
78 |
+
'First AI Message', value="Hello, there! How can I help you today?"
|
79 |
+
)
|
80 |
+
st.session_state.max_response_length = st.number_input(
|
81 |
+
"Max Response Length", value=128
|
82 |
+
)
|
83 |
+
st.markdown("*Select Avatars:*")
|
84 |
+
col1, col2 = st.columns(2)
|
85 |
+
with col1:
|
86 |
+
st.session_state.avatars['assistant'] = st.selectbox(
|
87 |
+
"AI Avatar", options=["π€", "π¬", "π€"], index=0
|
88 |
+
)
|
89 |
+
with col2:
|
90 |
+
st.session_state.avatars['user'] = st.selectbox(
|
91 |
+
"User Avatar", options=["π€", "π±ββοΈ", "π¨πΎ", "π©", "π§πΎ"], index=0
|
92 |
+
)
|
93 |
+
reset_history = st.button("Reset Chat History")
|
94 |
+
|
95 |
+
# Initialize or reset chat history
|
96 |
+
if "chat_history" not in st.session_state or reset_history:
|
97 |
+
st.session_state.chat_history = [{"role": "assistant", "content": st.session_state.starter_message}]
|
98 |
+
|
99 |
+
def get_response(system_message, chat_history, user_text, max_new_tokens=256):
|
100 |
+
hf = get_llm_hf_inference(max_new_tokens=max_new_tokens, temperature=0.1)
|
101 |
+
if hf is None:
|
102 |
+
return "Error with model inference.", chat_history
|
103 |
+
|
104 |
+
prompt = PromptTemplate.from_template(
|
105 |
+
"[INST] {system_message}\nCurrent Conversation:\n{chat_history}\n\nUser: {user_text}.\n [/INST]\nAI:"
|
106 |
+
)
|
107 |
+
chat = prompt | hf.bind(skip_prompt=True) | StrOutputParser(output_key='content')
|
108 |
+
response = chat.invoke(input=dict(system_message=system_message, user_text=user_text, chat_history=chat_history))
|
109 |
+
response = response.split("AI:")[-1]
|
110 |
+
|
111 |
+
chat_history.append({'role': 'user', 'content': user_text})
|
112 |
+
chat_history.append({'role': 'assistant', 'content': response})
|
113 |
+
return response, chat_history
|
114 |
+
|
115 |
+
# Chat interface
|
116 |
+
chat_interface = st.container()
|
117 |
+
with chat_interface:
|
118 |
+
output_container = st.container()
|
119 |
+
|
120 |
+
# Image upload and captioning
|
121 |
+
uploaded_image = st.file_uploader("Upload an image", type=["jpg", "jpeg", "png"])
|
122 |
+
if uploaded_image and st.session_state.uploaded_image_path is None:
|
123 |
+
# Save the uploaded image to a session-local directory
|
124 |
+
with st.spinner("Processing image... 0%"):
|
125 |
+
image = Image.open(uploaded_image).convert("RGB")
|
126 |
+
|
127 |
+
# Create a directory for session images if not exists
|
128 |
+
if not os.path.exists("session_images"):
|
129 |
+
os.makedirs("session_images")
|
130 |
+
|
131 |
+
# Save image to local session directory
|
132 |
+
image_path = os.path.join("session_images", uploaded_image.name)
|
133 |
+
image.save(image_path)
|
134 |
+
|
135 |
+
# Generate and save caption
|
136 |
+
caption = generate_caption(image)
|
137 |
+
st.session_state.chat_history.append({'role': 'user', 'content': f'![uploaded image]({image_path})'})
|
138 |
+
st.session_state.chat_history.append({'role': 'assistant', 'content': caption})
|
139 |
+
st.spinner("Processing image... 100%")
|
140 |
+
|
141 |
+
st.session_state.user_text = st.chat_input(placeholder="Enter your text here.")
|
142 |
+
|
143 |
+
if st.session_state.user_text:
|
144 |
+
with st.chat_message("user", avatar=st.session_state.avatars['user']):
|
145 |
+
st.markdown(st.session_state.user_text)
|
146 |
+
with st.chat_message("assistant", avatar=st.session_state.avatars['assistant']):
|
147 |
+
response, st.session_state.chat_history = get_response(
|
148 |
+
system_message=st.session_state.system_message,
|
149 |
+
chat_history=st.session_state.chat_history,
|
150 |
+
user_text=st.session_state.user_text,
|
151 |
+
max_new_tokens=st.session_state.max_response_length
|
152 |
+
)
|
153 |
+
st.markdown(response)
|
154 |
+
st.spinner("Thinking... 100%")
|
155 |
+
|