Spaces:
Sleeping
Sleeping
removed punkt
Browse files
app.py
CHANGED
@@ -1,10 +1,6 @@
|
|
1 |
import os
|
2 |
import logging
|
3 |
import re
|
4 |
-
import nltk
|
5 |
-
import spacy
|
6 |
-
import traceback
|
7 |
-
from nltk.tokenize import sent_tokenize
|
8 |
from langchain.vectorstores import Chroma
|
9 |
from langchain_core.output_parsers import StrOutputParser
|
10 |
from langchain_core.runnables import RunnablePassthrough
|
@@ -20,13 +16,9 @@ import pandas as pd
|
|
20 |
import json
|
21 |
|
22 |
# Enable logging for debugging
|
23 |
-
logging.basicConfig(level=logging.
|
24 |
logger = logging.getLogger(__name__)
|
25 |
|
26 |
-
# Set NLTK data path to the local 'nltk_data' directory
|
27 |
-
nltk.data.path.append(os.path.join(os.path.dirname(__file__), 'nltk_data'))
|
28 |
-
logger.debug("Configured NLTK data path to local 'nltk_data' directory.")
|
29 |
-
|
30 |
# Function to clean the API key
|
31 |
def clean_api_key(key):
|
32 |
return ''.join(c for c in key if ord(c) < 128)
|
@@ -77,83 +69,41 @@ def load_documents(file_paths):
|
|
77 |
logger.warning(f"Unsupported file format: {file_path}")
|
78 |
except Exception as e:
|
79 |
logger.error(f"Error processing file {file_path}: {e}")
|
80 |
-
logger.
|
81 |
return docs
|
82 |
|
83 |
-
# Function to ensure the response ends with complete sentences
|
84 |
def ensure_complete_sentences(text):
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
nltk.download('punkt', download_dir=os.path.join(os.path.dirname(__file__), 'nltk_data'))
|
95 |
-
nltk.data.path.append(os.path.join(os.path.dirname(__file__), 'nltk_data'))
|
96 |
-
sentences = sent_tokenize(text)
|
97 |
-
return ' '.join(sentences).strip()
|
98 |
-
except Exception as e_inner:
|
99 |
-
logger.error("Failed to download 'punkt' resource.")
|
100 |
-
logger.error(traceback.format_exc())
|
101 |
-
raise e_inner
|
102 |
-
except Exception as e:
|
103 |
-
logger.error("Unexpected error during sentence tokenization.")
|
104 |
-
logger.error(traceback.format_exc())
|
105 |
-
raise e
|
106 |
-
|
107 |
-
# Advanced input validation using spaCy (Section 8a)
|
108 |
-
def is_valid_input_nlp(text, threshold=0.5):
|
109 |
"""
|
110 |
-
|
111 |
-
|
112 |
-
Parameters:
|
113 |
-
- text (str): The input text to validate.
|
114 |
-
- threshold (float): The minimum ratio of meaningful tokens required.
|
115 |
-
|
116 |
-
Returns:
|
117 |
-
- bool: True if the input is valid, False otherwise.
|
118 |
"""
|
119 |
if not text or text.strip() == "":
|
120 |
-
logger.debug("Input text is empty or contains only whitespace.")
|
121 |
return False
|
122 |
-
|
123 |
-
|
124 |
-
if not meaningful_tokens:
|
125 |
-
logger.debug("No meaningful (alphabetic) tokens found in input.")
|
126 |
return False
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
# Function to estimate prompt tokens (simple word count approximation)
|
132 |
-
def estimate_prompt_tokens(prompt):
|
133 |
-
"""
|
134 |
-
Estimates the number of tokens in the prompt.
|
135 |
-
This is a placeholder function. Replace it with actual token estimation logic.
|
136 |
-
|
137 |
-
Parameters:
|
138 |
-
- prompt (str): The prompt text.
|
139 |
-
|
140 |
-
Returns:
|
141 |
-
- int: Estimated number of tokens.
|
142 |
-
"""
|
143 |
-
return len(prompt.split())
|
144 |
|
145 |
# Initialize the LLM using ChatGroq with GROQ's API
|
146 |
-
def initialize_llm(model, temperature, max_tokens
|
147 |
try:
|
148 |
-
#
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
# Allocate remaining tokens to response
|
153 |
-
response_max_tokens = max_tokens - estimated_prompt_tokens
|
154 |
-
logger.debug(f"Response max tokens: {response_max_tokens}")
|
155 |
-
|
156 |
-
if response_max_tokens <= 100:
|
157 |
raise ValueError("max_tokens is too small to allocate for the response.")
|
158 |
|
159 |
llm = ChatGroq(
|
@@ -162,53 +112,26 @@ def initialize_llm(model, temperature, max_tokens, prompt_template):
|
|
162 |
max_tokens=response_max_tokens, # Adjusted max_tokens
|
163 |
api_key=api_key # Ensure the API key is passed correctly
|
164 |
)
|
165 |
-
logger.
|
166 |
return llm
|
167 |
except Exception as e:
|
168 |
logger.error(f"Error initializing LLM: {e}")
|
169 |
-
|
170 |
-
raise e
|
171 |
|
172 |
# Create the RAG pipeline
|
173 |
def create_rag_pipeline(file_paths, model, temperature, max_tokens):
|
174 |
try:
|
175 |
-
|
176 |
-
custom_prompt_template = PromptTemplate(
|
177 |
-
input_variables=["context", "question"],
|
178 |
-
template="""
|
179 |
-
You are an AI assistant with expertise in daily wellness. Your aim is to provide detailed and comprehensive solutions regarding daily wellness topics without unnecessary verbosity.
|
180 |
-
|
181 |
-
Context:
|
182 |
-
{context}
|
183 |
-
|
184 |
-
Question:
|
185 |
-
{question}
|
186 |
-
|
187 |
-
Provide a thorough and complete answer, including relevant examples and a suggested schedule. Ensure that the response does not end abruptly.
|
188 |
-
"""
|
189 |
-
)
|
190 |
-
|
191 |
-
# Estimate prompt tokens
|
192 |
-
estimated_prompt_tokens = estimate_prompt_tokens(custom_prompt_template.template)
|
193 |
-
logger.debug(f"Estimated prompt tokens from template: {estimated_prompt_tokens}")
|
194 |
-
|
195 |
-
# Initialize the LLM with token allocation
|
196 |
-
llm = initialize_llm(model, temperature, max_tokens, custom_prompt_template.template)
|
197 |
-
|
198 |
-
# Load and process documents
|
199 |
docs = load_documents(file_paths)
|
200 |
if not docs:
|
201 |
logger.warning("No documents were loaded. Please check your file paths and formats.")
|
202 |
return None, "No documents were loaded. Please check your file paths and formats."
|
203 |
|
204 |
-
# Split documents into chunks
|
205 |
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
|
206 |
splits = text_splitter.split_documents(docs)
|
207 |
-
logger.debug(f"Documents split into {len(splits)} chunks.")
|
208 |
|
209 |
# Initialize the embedding model
|
210 |
embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
211 |
-
logger.debug("Embedding model initialized successfully.")
|
212 |
|
213 |
# Use a temporary directory for Chroma vectorstore to prevent caching issues on Hugging Face Spaces
|
214 |
vectorstore = Chroma.from_documents(
|
@@ -217,25 +140,39 @@ def create_rag_pipeline(file_paths, model, temperature, max_tokens):
|
|
217 |
persist_directory="/tmp/chroma_db" # Temporary storage directory
|
218 |
)
|
219 |
vectorstore.persist() # Save the database to disk
|
220 |
-
logger.
|
221 |
|
222 |
retriever = vectorstore.as_retriever()
|
223 |
|
224 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
225 |
rag_chain = RetrievalQA.from_chain_type(
|
226 |
llm=llm,
|
227 |
chain_type="stuff",
|
228 |
retriever=retriever,
|
229 |
chain_type_kwargs={"prompt": custom_prompt_template}
|
230 |
)
|
231 |
-
logger.
|
232 |
return rag_chain, "Pipeline created successfully."
|
233 |
except Exception as e:
|
234 |
logger.error(f"Error creating RAG pipeline: {e}")
|
235 |
-
logger.
|
236 |
return None, f"Error creating RAG pipeline: {e}"
|
237 |
|
238 |
-
# Function to handle feedback (
|
239 |
def handle_feedback(feedback_text):
|
240 |
"""
|
241 |
Handles user feedback by logging it.
|
@@ -254,43 +191,48 @@ def handle_feedback(feedback_text):
|
|
254 |
else:
|
255 |
return "No feedback provided."
|
256 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
257 |
# Function to answer questions with input validation and post-processing
|
258 |
-
def answer_question(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
259 |
try:
|
260 |
-
|
261 |
-
|
262 |
-
|
263 |
-
|
264 |
|
265 |
-
|
266 |
-
|
267 |
-
logger.debug("RAG pipeline creation failed.")
|
268 |
-
return message, ""
|
269 |
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
274 |
-
|
275 |
-
|
276 |
-
|
277 |
-
feedback_response = handle_feedback(feedback)
|
278 |
-
|
279 |
-
return complete_answer, feedback_response
|
280 |
-
except Exception as e_inner:
|
281 |
-
logger.error(f"Error during RAG pipeline execution: {e_inner}")
|
282 |
-
logger.error(traceback.format_exc())
|
283 |
-
return f"Error during RAG pipeline execution: {e_inner}", ""
|
284 |
-
|
285 |
-
except Exception as e_outer:
|
286 |
-
logger.error(f"Unexpected error in answer_question: {e_outer}")
|
287 |
-
logger.error(traceback.format_exc())
|
288 |
-
return f"Unexpected error: {e_outer}", ""
|
289 |
-
|
290 |
-
# Gradio Interface with Feedback Mechanism (Section 8d)
|
291 |
def gradio_interface(model, temperature, max_tokens, question, feedback):
|
292 |
-
|
293 |
-
|
|
|
294 |
|
295 |
# Define Gradio UI
|
296 |
interface = gr.Interface(
|
@@ -298,7 +240,7 @@ interface = gr.Interface(
|
|
298 |
inputs=[
|
299 |
gr.Textbox(
|
300 |
label="Model Name",
|
301 |
-
value=
|
302 |
placeholder="e.g., llama3-8b-8192"
|
303 |
),
|
304 |
gr.Slider(
|
@@ -306,7 +248,7 @@ interface = gr.Interface(
|
|
306 |
minimum=0,
|
307 |
maximum=1,
|
308 |
step=0.01,
|
309 |
-
value=
|
310 |
info="Controls the randomness of the response. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic."
|
311 |
),
|
312 |
gr.Slider(
|
@@ -314,7 +256,7 @@ interface = gr.Interface(
|
|
314 |
minimum=200,
|
315 |
maximum=2048,
|
316 |
step=1,
|
317 |
-
value=
|
318 |
info="Determines the maximum number of tokens in the response. Higher values allow for longer answers."
|
319 |
),
|
320 |
gr.Textbox(
|
|
|
1 |
import os
|
2 |
import logging
|
3 |
import re
|
|
|
|
|
|
|
|
|
4 |
from langchain.vectorstores import Chroma
|
5 |
from langchain_core.output_parsers import StrOutputParser
|
6 |
from langchain_core.runnables import RunnablePassthrough
|
|
|
16 |
import json
|
17 |
|
18 |
# Enable logging for debugging
|
19 |
+
logging.basicConfig(level=logging.INFO) # Changed to INFO to reduce verbosity
|
20 |
logger = logging.getLogger(__name__)
|
21 |
|
|
|
|
|
|
|
|
|
22 |
# Function to clean the API key
|
23 |
def clean_api_key(key):
|
24 |
return ''.join(c for c in key if ord(c) < 128)
|
|
|
69 |
logger.warning(f"Unsupported file format: {file_path}")
|
70 |
except Exception as e:
|
71 |
logger.error(f"Error processing file {file_path}: {e}")
|
72 |
+
logger.debug("Exception details:", exc_info=True)
|
73 |
return docs
|
74 |
|
75 |
+
# Function to ensure the response ends with complete sentences
|
76 |
def ensure_complete_sentences(text):
|
77 |
+
# Use regex to find all complete sentences
|
78 |
+
sentences = re.findall(r'[^.!?]*[.!?]', text)
|
79 |
+
if sentences:
|
80 |
+
# Join all complete sentences to form the complete answer
|
81 |
+
return ' '.join(sentences).strip()
|
82 |
+
return text # Return as is if no complete sentence is found
|
83 |
+
|
84 |
+
# Function to check if input is valid
|
85 |
+
def is_valid_input(text):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
86 |
"""
|
87 |
+
Checks if the input text is meaningful.
|
88 |
+
Returns True if the text contains alphabetic characters and is of sufficient length.
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
"""
|
90 |
if not text or text.strip() == "":
|
|
|
91 |
return False
|
92 |
+
# Regex to check for at least one alphabetic character
|
93 |
+
if not re.search('[A-Za-z]', text):
|
|
|
|
|
94 |
return False
|
95 |
+
# Additional check: minimum length
|
96 |
+
if len(text.strip()) < 5:
|
97 |
+
return False
|
98 |
+
return True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
|
100 |
# Initialize the LLM using ChatGroq with GROQ's API
|
101 |
+
def initialize_llm(model, temperature, max_tokens):
|
102 |
try:
|
103 |
+
# Allocate a portion of tokens for the prompt, e.g., 20%
|
104 |
+
prompt_allocation = int(max_tokens * 0.2)
|
105 |
+
response_max_tokens = max_tokens - prompt_allocation
|
106 |
+
if response_max_tokens <= 50:
|
|
|
|
|
|
|
|
|
|
|
107 |
raise ValueError("max_tokens is too small to allocate for the response.")
|
108 |
|
109 |
llm = ChatGroq(
|
|
|
112 |
max_tokens=response_max_tokens, # Adjusted max_tokens
|
113 |
api_key=api_key # Ensure the API key is passed correctly
|
114 |
)
|
115 |
+
logger.info("LLM initialized successfully.")
|
116 |
return llm
|
117 |
except Exception as e:
|
118 |
logger.error(f"Error initializing LLM: {e}")
|
119 |
+
raise
|
|
|
120 |
|
121 |
# Create the RAG pipeline
|
122 |
def create_rag_pipeline(file_paths, model, temperature, max_tokens):
|
123 |
try:
|
124 |
+
llm = initialize_llm(model, temperature, max_tokens)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
125 |
docs = load_documents(file_paths)
|
126 |
if not docs:
|
127 |
logger.warning("No documents were loaded. Please check your file paths and formats.")
|
128 |
return None, "No documents were loaded. Please check your file paths and formats."
|
129 |
|
|
|
130 |
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
|
131 |
splits = text_splitter.split_documents(docs)
|
|
|
132 |
|
133 |
# Initialize the embedding model
|
134 |
embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
|
|
135 |
|
136 |
# Use a temporary directory for Chroma vectorstore to prevent caching issues on Hugging Face Spaces
|
137 |
vectorstore = Chroma.from_documents(
|
|
|
140 |
persist_directory="/tmp/chroma_db" # Temporary storage directory
|
141 |
)
|
142 |
vectorstore.persist() # Save the database to disk
|
143 |
+
logger.info("Vectorstore initialized and persisted successfully.")
|
144 |
|
145 |
retriever = vectorstore.as_retriever()
|
146 |
|
147 |
+
custom_prompt_template = PromptTemplate(
|
148 |
+
input_variables=["context", "question"],
|
149 |
+
template="""
|
150 |
+
You are an AI assistant with expertise in daily wellness. Your aim is to provide detailed and comprehensive solutions regarding daily wellness topics without unnecessary verbosity.
|
151 |
+
|
152 |
+
Context:
|
153 |
+
{context}
|
154 |
+
|
155 |
+
Question:
|
156 |
+
{question}
|
157 |
+
|
158 |
+
Provide a thorough and complete answer, including relevant examples and a suggested schedule. Ensure that the response does not end abruptly.
|
159 |
+
"""
|
160 |
+
)
|
161 |
+
|
162 |
rag_chain = RetrievalQA.from_chain_type(
|
163 |
llm=llm,
|
164 |
chain_type="stuff",
|
165 |
retriever=retriever,
|
166 |
chain_type_kwargs={"prompt": custom_prompt_template}
|
167 |
)
|
168 |
+
logger.info("RAG pipeline created successfully.")
|
169 |
return rag_chain, "Pipeline created successfully."
|
170 |
except Exception as e:
|
171 |
logger.error(f"Error creating RAG pipeline: {e}")
|
172 |
+
logger.debug("Exception details:", exc_info=True)
|
173 |
return None, f"Error creating RAG pipeline: {e}"
|
174 |
|
175 |
+
# Function to handle feedback (Optional Enhancement)
|
176 |
def handle_feedback(feedback_text):
|
177 |
"""
|
178 |
Handles user feedback by logging it.
|
|
|
191 |
else:
|
192 |
return "No feedback provided."
|
193 |
|
194 |
+
# Initialize the RAG pipeline once at startup
|
195 |
+
# Define the file paths (ensure 'AIChatbot.csv' is in the root directory of your Space)
|
196 |
+
file_paths = ['AIChatbot.csv']
|
197 |
+
model = "llama3-8b-8192" # Default model name
|
198 |
+
temperature = 0.7 # Default temperature
|
199 |
+
max_tokens = 500 # Default max tokens
|
200 |
+
|
201 |
+
rag_chain, message = create_rag_pipeline(file_paths, model, temperature, max_tokens)
|
202 |
+
if rag_chain is None:
|
203 |
+
logger.error("Failed to initialize RAG pipeline at startup.")
|
204 |
+
# Depending on your preference, you might want to exit or continue. Here, we'll continue.
|
205 |
+
|
206 |
# Function to answer questions with input validation and post-processing
|
207 |
+
def answer_question(model, temperature, max_tokens, question, feedback):
|
208 |
+
# Validate input
|
209 |
+
if not is_valid_input(question):
|
210 |
+
logger.info("Received invalid input from user.")
|
211 |
+
return "Please provide a valid question or input containing meaningful text.", ""
|
212 |
+
|
213 |
+
# Check if the RAG pipeline needs to be re-initialized (e.g., if model or parameters have changed)
|
214 |
+
# For simplicity, we'll assume the pipeline remains the same. For dynamic models, implement re-initialization here.
|
215 |
+
|
216 |
try:
|
217 |
+
answer = rag_chain.run(question)
|
218 |
+
logger.info("Question answered successfully.")
|
219 |
+
# Post-process to ensure the answer ends with complete sentences
|
220 |
+
complete_answer = ensure_complete_sentences(answer)
|
221 |
|
222 |
+
# Handle feedback
|
223 |
+
feedback_response = handle_feedback(feedback)
|
|
|
|
|
224 |
|
225 |
+
return complete_answer, feedback_response
|
226 |
+
except Exception as e_inner:
|
227 |
+
logger.error(f"Error during RAG pipeline execution: {e_inner}")
|
228 |
+
logger.debug("Exception details:", exc_info=True)
|
229 |
+
return f"Error during RAG pipeline execution: {e_inner}", ""
|
230 |
+
|
231 |
+
# Gradio Interface with Feedback Mechanism
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
232 |
def gradio_interface(model, temperature, max_tokens, question, feedback):
|
233 |
+
# Optionally, you can add functionality to update the RAG pipeline if model or parameters change
|
234 |
+
# For now, we'll ignore changes to model parameters after initialization
|
235 |
+
return answer_question(model, temperature, max_tokens, question, feedback)
|
236 |
|
237 |
# Define Gradio UI
|
238 |
interface = gr.Interface(
|
|
|
240 |
inputs=[
|
241 |
gr.Textbox(
|
242 |
label="Model Name",
|
243 |
+
value=model,
|
244 |
placeholder="e.g., llama3-8b-8192"
|
245 |
),
|
246 |
gr.Slider(
|
|
|
248 |
minimum=0,
|
249 |
maximum=1,
|
250 |
step=0.01,
|
251 |
+
value=temperature,
|
252 |
info="Controls the randomness of the response. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic."
|
253 |
),
|
254 |
gr.Slider(
|
|
|
256 |
minimum=200,
|
257 |
maximum=2048,
|
258 |
step=1,
|
259 |
+
value=max_tokens,
|
260 |
info="Determines the maximum number of tokens in the response. Higher values allow for longer answers."
|
261 |
),
|
262 |
gr.Textbox(
|