Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -8,6 +8,8 @@ from llama_index import ServiceContext, VectorStoreIndex, Document, StorageConte
|
|
8 |
from llama_index.memory import ChatMemoryBuffer
|
9 |
import os
|
10 |
import datetime
|
|
|
|
|
11 |
|
12 |
#imports for resnet
|
13 |
from transformers import AutoFeatureExtractor, ResNetForImageClassification
|
@@ -45,8 +47,7 @@ This application, titled 'AInimal Go!', is a conceptual prototype designed to de
|
|
45 |
cookie_manager = stx.CookieManager()
|
46 |
|
47 |
#Function to init resnet
|
48 |
-
|
49 |
-
@st.cache_resource()
|
50 |
def load_model_and_labels():
|
51 |
# Load animal labels as a dictionary
|
52 |
animal_labels_dict = {}
|
@@ -81,9 +82,11 @@ def get_image_caption(image_data):
|
|
81 |
return predicted_label_name, predicted_label_id
|
82 |
|
83 |
|
84 |
-
@st.cache_resource
|
85 |
def init_llm(api_key):
|
86 |
-
llm = PaLM(api_key=api_key)
|
|
|
|
|
87 |
service_context = ServiceContext.from_defaults(llm=llm, embed_model="local")
|
88 |
|
89 |
storage_context = StorageContext.from_defaults(persist_dir="storage")
|
@@ -92,27 +95,34 @@ def init_llm(api_key):
|
|
92 |
|
93 |
return llm, service_context, storage_context, index, chatmemory
|
94 |
|
95 |
-
llm, service_context, storage_context, index, chatmemory = init_llm(
|
96 |
|
97 |
def is_animal(predicted_label_id):
|
98 |
# Check if the predicted label ID is within the animal classes range
|
99 |
return 0 <= predicted_label_id <= 398
|
100 |
|
101 |
-
|
102 |
# Function to create the chat engine.
|
103 |
@st.cache_resource
|
104 |
def create_chat_engine(img_desc, api_key):
|
|
|
|
|
|
|
105 |
doc = Document(text=img_desc)
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
|
|
|
|
|
|
|
|
|
|
111 |
)
|
112 |
|
113 |
-
return
|
|
|
114 |
|
115 |
-
|
116 |
# Clear chat function
|
117 |
def clear_chat():
|
118 |
if "messages" in st.session_state:
|
@@ -149,7 +159,7 @@ else:
|
|
149 |
|
150 |
col1, col2, col3 = st.columns([1, 2, 1])
|
151 |
with col2: # Camera input will be in the middle column
|
152 |
-
camera_image = st.camera_input("Take a picture")
|
153 |
|
154 |
|
155 |
# Determine the source of the image (upload or camera)
|
@@ -162,17 +172,20 @@ else:
|
|
162 |
|
163 |
if image_data:
|
164 |
# Display the uploaded image at a standard width.
|
|
|
165 |
st.image(image_data, caption='Uploaded Image.', width=200)
|
166 |
|
167 |
# Process the uploaded image to get a caption.
|
|
|
168 |
img_desc, label_id = get_image_caption(image_data)
|
169 |
|
170 |
if not (is_animal(label_id)):
|
|
|
171 |
st.error("Please upload image of an animal!")
|
172 |
st.stop()
|
173 |
|
174 |
# Initialize the chat engine with the image description.
|
175 |
-
chat_engine = create_chat_engine(img_desc,
|
176 |
st.write("Image Uploaded Successfully. Ask me anything about it.")
|
177 |
|
178 |
|
@@ -182,8 +195,9 @@ else:
|
|
182 |
|
183 |
# Display previous messages
|
184 |
for message in st.session_state.messages:
|
185 |
-
|
186 |
-
|
|
|
187 |
|
188 |
# Handle new user input
|
189 |
user_input = st.chat_input("Ask me about the image:", key="chat_input")
|
@@ -193,27 +207,38 @@ else:
|
|
193 |
|
194 |
# Display user message immediately
|
195 |
with st.chat_message("user"):
|
196 |
-
st.
|
197 |
|
198 |
# Call the chat engine to get the response if an image has been uploaded
|
199 |
if image_data and user_input:
|
200 |
try:
|
201 |
with st.spinner('Waiting for the chat engine to respond...'):
|
202 |
# Get the response from your chat engine
|
203 |
-
|
204 |
-
You
|
205 |
-
|
206 |
-
{
|
207 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
208 |
# Append assistant message to the session state
|
209 |
-
st.session_state.messages.append({"role": "assistant", "content": response})
|
210 |
|
211 |
# Display the assistant message
|
212 |
with st.chat_message("assistant"):
|
213 |
-
st.
|
|
|
214 |
|
215 |
except Exception as e:
|
216 |
st.error(f'An error occurred.')
|
|
|
|
|
217 |
|
218 |
# Increment the message count and update the cookie
|
219 |
message_count += 1
|
|
|
8 |
from llama_index.memory import ChatMemoryBuffer
|
9 |
import os
|
10 |
import datetime
|
11 |
+
from llama_index.llms import Cohere
|
12 |
+
from llama_index.query_engine import CitationQueryEngine
|
13 |
|
14 |
#imports for resnet
|
15 |
from transformers import AutoFeatureExtractor, ResNetForImageClassification
|
|
|
47 |
cookie_manager = stx.CookieManager()
|
48 |
|
49 |
#Function to init resnet
|
50 |
+
@st.cache_resource(show_spinner="Initializing ResNet model for image classification. Please wait...")
|
|
|
51 |
def load_model_and_labels():
|
52 |
# Load animal labels as a dictionary
|
53 |
animal_labels_dict = {}
|
|
|
82 |
return predicted_label_name, predicted_label_id
|
83 |
|
84 |
|
85 |
+
@st.cache_resource(show_spinner="Initializing LLM and setting up service context. Please wait...")
|
86 |
def init_llm(api_key):
|
87 |
+
# llm = PaLM(api_key=api_key)
|
88 |
+
llm = Cohere(model="command", api_key=st.secrets['COHERE_API_TOKEN'])
|
89 |
+
|
90 |
service_context = ServiceContext.from_defaults(llm=llm, embed_model="local")
|
91 |
|
92 |
storage_context = StorageContext.from_defaults(persist_dir="storage")
|
|
|
95 |
|
96 |
return llm, service_context, storage_context, index, chatmemory
|
97 |
|
98 |
+
llm, service_context, storage_context, index, chatmemory = init_llm(os.environ["GOOGLE_API_KEY"])
|
99 |
|
100 |
def is_animal(predicted_label_id):
|
101 |
# Check if the predicted label ID is within the animal classes range
|
102 |
return 0 <= predicted_label_id <= 398
|
103 |
|
|
|
104 |
# Function to create the chat engine.
|
105 |
@st.cache_resource
|
106 |
def create_chat_engine(img_desc, api_key):
|
107 |
+
|
108 |
+
#llm = PaLM(api_key=api_key)
|
109 |
+
#service_context = ServiceContext.from_defaults(llm=llm,embed_model="local")
|
110 |
doc = Document(text=img_desc)
|
111 |
+
|
112 |
+
# Now is_animal is a boolean indicating whether the image is of an animal
|
113 |
+
print("Is the image of an animal:", is_animal)
|
114 |
+
|
115 |
+
query_engine = CitationQueryEngine.from_args(
|
116 |
+
index,
|
117 |
+
similarity_top_k=3,
|
118 |
+
# here we can control how granular citation sources are, the default is 512
|
119 |
+
citation_chunk_size=512,
|
120 |
+
verbose=True
|
121 |
)
|
122 |
|
123 |
+
return query_engine
|
124 |
+
|
125 |
|
|
|
126 |
# Clear chat function
|
127 |
def clear_chat():
|
128 |
if "messages" in st.session_state:
|
|
|
159 |
|
160 |
col1, col2, col3 = st.columns([1, 2, 1])
|
161 |
with col2: # Camera input will be in the middle column
|
162 |
+
camera_image = st.camera_input("Take a picture", on_change=on_image_upload)
|
163 |
|
164 |
|
165 |
# Determine the source of the image (upload or camera)
|
|
|
172 |
|
173 |
if image_data:
|
174 |
# Display the uploaded image at a standard width.
|
175 |
+
st.session_state['assistant_avatar'] = image_data
|
176 |
st.image(image_data, caption='Uploaded Image.', width=200)
|
177 |
|
178 |
# Process the uploaded image to get a caption.
|
179 |
+
#img_desc = get_image_caption(image_data)
|
180 |
img_desc, label_id = get_image_caption(image_data)
|
181 |
|
182 |
if not (is_animal(label_id)):
|
183 |
+
#st.error("Please upload image of an animal!")
|
184 |
st.error("Please upload image of an animal!")
|
185 |
st.stop()
|
186 |
|
187 |
# Initialize the chat engine with the image description.
|
188 |
+
chat_engine = create_chat_engine(img_desc, os.environ["GOOGLE_API_KEY"])
|
189 |
st.write("Image Uploaded Successfully. Ask me anything about it.")
|
190 |
|
191 |
|
|
|
195 |
|
196 |
# Display previous messages
|
197 |
for message in st.session_state.messages:
|
198 |
+
avatar = st.session_state['assistant_avatar'] if message["role"] == "assistant" else None
|
199 |
+
with st.chat_message(message["role"], avatar = avatar):
|
200 |
+
st.write(message["content"])
|
201 |
|
202 |
# Handle new user input
|
203 |
user_input = st.chat_input("Ask me about the image:", key="chat_input")
|
|
|
207 |
|
208 |
# Display user message immediately
|
209 |
with st.chat_message("user"):
|
210 |
+
st.write(user_input)
|
211 |
|
212 |
# Call the chat engine to get the response if an image has been uploaded
|
213 |
if image_data and user_input:
|
214 |
try:
|
215 |
with st.spinner('Waiting for the chat engine to respond...'):
|
216 |
# Get the response from your chat engine
|
217 |
+
system_prompt=f"""
|
218 |
+
You are a chatbot, able to have normal interactions. Do not make up information.
|
219 |
+
You always answer in great detail and are polite. Your job is to roleplay as an {img_desc}.
|
220 |
+
Remember to make {img_desc} sounds while talking but dont overdo it.
|
221 |
+
"""
|
222 |
+
|
223 |
+
response = chat_engine.query(f"{system_prompt}. {user_input}")
|
224 |
+
|
225 |
+
#response = chat_engine.chat(f"""You are a chatbot that roleplays as an animal and also makes animal sounds when chatting.
|
226 |
+
#You always answer in great detail and are polite. Your responses always descriptive.
|
227 |
+
#Your job is to rolelpay as the animal that is mentioned in the image the user has uploaded. Image description: {img_desc}. User question
|
228 |
+
#{user_input}""")
|
229 |
+
|
230 |
# Append assistant message to the session state
|
231 |
+
st.session_state.messages.append({"role": "assistant", "content": response.response})
|
232 |
|
233 |
# Display the assistant message
|
234 |
with st.chat_message("assistant"):
|
235 |
+
st.write(response.response)
|
236 |
+
st.expander("hello")
|
237 |
|
238 |
except Exception as e:
|
239 |
st.error(f'An error occurred.')
|
240 |
+
# Optionally, you can choose to break the flow here if a critical error happens
|
241 |
+
# return
|
242 |
|
243 |
# Increment the message count and update the cookie
|
244 |
message_count += 1
|