Spaces:
Sleeping
Sleeping
Update pipeline.py
Browse files- pipeline.py +8 -85
pipeline.py
CHANGED
@@ -40,28 +40,11 @@ class QueryInput(BaseModel):
|
|
40 |
raise ValueError("Query cannot be empty or just whitespace")
|
41 |
return v.strip()
|
42 |
|
43 |
-
class ClassificationResult(BaseModel):
|
44 |
-
category: str = Field(..., description="The classification category")
|
45 |
-
confidence: float = Field(..., ge=0.0, le=1.0, description="Classification confidence score")
|
46 |
-
|
47 |
class ModerationResult(BaseModel):
|
48 |
is_safe: bool = Field(..., description="Whether the content is safe")
|
49 |
categories: Dict[str, bool] = Field(default_factory=dict, description="Detected content categories")
|
50 |
original_text: str = Field(..., description="The original input text")
|
51 |
|
52 |
-
class RAGResponse(BaseModel):
|
53 |
-
answer: str = Field(..., description="The generated answer")
|
54 |
-
sources: List[str] = Field(default_factory=list, description="Source documents used")
|
55 |
-
confidence: float = Field(..., ge=0.0, le=1.0, description="Confidence score of the answer")
|
56 |
-
|
57 |
-
def sanitize_message(message: Any) -> str:
|
58 |
-
"""Sanitize message input to ensure it's a valid string."""
|
59 |
-
if hasattr(message, 'content'):
|
60 |
-
return str(message.content)
|
61 |
-
if isinstance(message, (list, dict)):
|
62 |
-
return str(message)
|
63 |
-
return str(message)
|
64 |
-
|
65 |
# Load spaCy model for NER
|
66 |
def install_spacy_model():
|
67 |
try:
|
@@ -128,27 +111,18 @@ def moderate_text(query: str) -> ModerationResult:
|
|
128 |
except Exception as e:
|
129 |
raise RuntimeError(f"Moderation failed: {str(e)}")
|
130 |
|
131 |
-
def classify_query(query: str) ->
|
132 |
try:
|
133 |
query_input = QueryInput(query=query)
|
134 |
|
135 |
wellness_keywords = ["box breathing", "meditation", "yoga", "mindfulness", "breathing exercises"]
|
136 |
if any(keyword in query_input.query.lower() for keyword in wellness_keywords):
|
137 |
-
return
|
138 |
|
139 |
class_result = classification_chain.invoke({"query": query_input.query})
|
140 |
classification = class_result.get("text", "").strip()
|
141 |
|
142 |
-
|
143 |
-
"Wellness": 0.8,
|
144 |
-
"Brand": 0.8,
|
145 |
-
"OutOfScope": 0.6
|
146 |
-
}
|
147 |
-
|
148 |
-
return ClassificationResult(
|
149 |
-
category=classification if classification != "" else "OutOfScope",
|
150 |
-
confidence=confidence_map.get(classification, 0.5)
|
151 |
-
)
|
152 |
except ValidationError as e:
|
153 |
raise ValueError(f"Classification input validation failed: {str(e)}")
|
154 |
except Exception as e:
|
@@ -166,14 +140,6 @@ def build_or_load_vectorstore(csv_path: str, store_dir: str) -> FAISS:
|
|
166 |
df = df.loc[:, ~df.columns.str.contains('^Unnamed')]
|
167 |
df.columns = df.columns.str.strip()
|
168 |
|
169 |
-
if "Answer" in df.columns:
|
170 |
-
df.rename(columns={"Answer": "Answers"}, inplace=True)
|
171 |
-
if "Question" not in df.columns and "Question " in df.columns:
|
172 |
-
df.rename(columns={"Question ": "Question"}, inplace=True)
|
173 |
-
|
174 |
-
if "Question" not in df.columns or "Answers" not in df.columns:
|
175 |
-
raise ValueError("CSV must have 'Question' and 'Answers' columns")
|
176 |
-
|
177 |
docs = [
|
178 |
Document(page_content=str(row["Answers"]), metadata={"question": str(row["Question"])})
|
179 |
for _, row in df.iterrows()
|
@@ -209,98 +175,55 @@ def build_rag_chain(llm_model: LiteLLMModel, vectorstore: FAISS) -> RetrievalQA:
|
|
209 |
except Exception as e:
|
210 |
raise RuntimeError(f"Error building RAG chain: {str(e)}")
|
211 |
|
212 |
-
def do_web_search(query: str) -> str:
|
213 |
-
try:
|
214 |
-
query_input = QueryInput(query=query)
|
215 |
-
search_tool = DuckDuckGoSearchTool()
|
216 |
-
web_agent = CodeAgent(tools=[search_tool], model=pydantic_agent)
|
217 |
-
managed_web_agent = ManagedAgent(agent=web_agent, name="web_search", description="Performs web searches")
|
218 |
-
manager_agent = CodeAgent(tools=[], model=pydantic_agent, managed_agents=[managed_web_agent])
|
219 |
-
|
220 |
-
search_query = f"Give me relevant info: {query_input.query}"
|
221 |
-
return manager_agent.run(search_query)
|
222 |
-
except Exception as e:
|
223 |
-
return f"Web search failed: {str(e)}"
|
224 |
-
|
225 |
-
def merge_responses(kb_answer: str, web_answer: str) -> str:
|
226 |
-
try:
|
227 |
-
if not kb_answer and not web_answer:
|
228 |
-
return "No relevant information found."
|
229 |
-
|
230 |
-
if not web_answer:
|
231 |
-
return kb_answer.strip()
|
232 |
-
|
233 |
-
if not kb_answer:
|
234 |
-
return web_answer.strip()
|
235 |
-
|
236 |
-
return f"Knowledge Base Answer: {kb_answer.strip()}\n\nWeb Search Result: {web_answer.strip()}"
|
237 |
-
except Exception as e:
|
238 |
-
return f"Error merging responses: {str(e)}"
|
239 |
-
|
240 |
def run_pipeline(query: str) -> str:
|
241 |
try:
|
242 |
-
# Sanitize and validate input
|
243 |
query = sanitize_message(query)
|
244 |
|
245 |
-
# Moderate content
|
246 |
moderation_result = moderate_text(query)
|
247 |
if not moderation_result.is_safe:
|
248 |
return "Sorry, this query contains harmful or inappropriate content."
|
249 |
|
250 |
-
|
251 |
-
classification_result = classify_query(moderation_result.original_text)
|
252 |
|
253 |
-
|
254 |
-
if classification_result.category == "OutOfScope":
|
255 |
refusal_text = refusal_chain.run({"topic": "this topic"})
|
256 |
return tailor_chain.run({"response": refusal_text}).strip()
|
257 |
|
258 |
-
if
|
259 |
rag_result = wellness_rag_chain({"query": moderation_result.original_text})
|
260 |
csv_answer = rag_result["result"].strip()
|
261 |
web_answer = "" if csv_answer else do_web_search(moderation_result.original_text)
|
262 |
final_merged = merge_responses(csv_answer, web_answer)
|
263 |
return tailor_chain.run({"response": final_merged}).strip()
|
264 |
|
265 |
-
if
|
266 |
rag_result = brand_rag_chain({"query": moderation_result.original_text})
|
267 |
csv_answer = rag_result["result"].strip()
|
268 |
final_merged = merge_responses(csv_answer, "")
|
269 |
return tailor_chain.run({"response": final_merged}).strip()
|
270 |
|
271 |
-
# Default fallback
|
272 |
refusal_text = refusal_chain.run({"topic": "this topic"})
|
273 |
return tailor_chain.run({"response": refusal_text}).strip()
|
274 |
|
275 |
-
except Exception as e:
|
276 |
-
return f"An error occurred while processing your request: {str(e)}"
|
277 |
-
|
278 |
# Initialize chains and vectorstores
|
279 |
try:
|
280 |
-
# Initialize chain components
|
281 |
classification_chain = get_classification_chain()
|
282 |
refusal_chain = get_refusal_chain()
|
283 |
tailor_chain = get_tailor_chain()
|
284 |
cleaner_chain = get_cleaner_chain()
|
285 |
|
286 |
-
# Set up paths
|
287 |
wellness_csv = "AIChatbot.csv"
|
288 |
brand_csv = "BrandAI.csv"
|
289 |
wellness_store_dir = "faiss_wellness_store"
|
290 |
brand_store_dir = "faiss_brand_store"
|
291 |
|
292 |
-
# Build or load vectorstores
|
293 |
wellness_vectorstore = build_or_load_vectorstore(wellness_csv, wellness_store_dir)
|
294 |
brand_vectorstore = build_or_load_vectorstore(brand_csv, brand_store_dir)
|
295 |
|
296 |
-
# Initialize LLM and RAG chains
|
297 |
gemini_llm = LiteLLMModel(model_id="gemini/gemini-pro", api_key=os.environ.get("GEMINI_API_KEY"))
|
298 |
wellness_rag_chain = build_rag_chain(gemini_llm, wellness_vectorstore)
|
299 |
brand_rag_chain = build_rag_chain(gemini_llm, brand_vectorstore)
|
300 |
-
|
301 |
print("Pipeline initialized successfully!")
|
302 |
except Exception as e:
|
303 |
print(f"Error initializing pipeline: {str(e)}")
|
304 |
-
|
305 |
-
def run_with_chain(query: str) -> str:
|
306 |
-
return run_pipeline(query)
|
|
|
40 |
raise ValueError("Query cannot be empty or just whitespace")
|
41 |
return v.strip()
|
42 |
|
|
|
|
|
|
|
|
|
43 |
class ModerationResult(BaseModel):
|
44 |
is_safe: bool = Field(..., description="Whether the content is safe")
|
45 |
categories: Dict[str, bool] = Field(default_factory=dict, description="Detected content categories")
|
46 |
original_text: str = Field(..., description="The original input text")
|
47 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
48 |
# Load spaCy model for NER
|
49 |
def install_spacy_model():
|
50 |
try:
|
|
|
111 |
except Exception as e:
|
112 |
raise RuntimeError(f"Moderation failed: {str(e)}")
|
113 |
|
114 |
+
def classify_query(query: str) -> str:
|
115 |
try:
|
116 |
query_input = QueryInput(query=query)
|
117 |
|
118 |
wellness_keywords = ["box breathing", "meditation", "yoga", "mindfulness", "breathing exercises"]
|
119 |
if any(keyword in query_input.query.lower() for keyword in wellness_keywords):
|
120 |
+
return "Wellness"
|
121 |
|
122 |
class_result = classification_chain.invoke({"query": query_input.query})
|
123 |
classification = class_result.get("text", "").strip()
|
124 |
|
125 |
+
return classification if classification != "" else "OutOfScope"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
except ValidationError as e:
|
127 |
raise ValueError(f"Classification input validation failed: {str(e)}")
|
128 |
except Exception as e:
|
|
|
140 |
df = df.loc[:, ~df.columns.str.contains('^Unnamed')]
|
141 |
df.columns = df.columns.str.strip()
|
142 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
143 |
docs = [
|
144 |
Document(page_content=str(row["Answers"]), metadata={"question": str(row["Question"])})
|
145 |
for _, row in df.iterrows()
|
|
|
175 |
except Exception as e:
|
176 |
raise RuntimeError(f"Error building RAG chain: {str(e)}")
|
177 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
178 |
def run_pipeline(query: str) -> str:
|
179 |
try:
|
|
|
180 |
query = sanitize_message(query)
|
181 |
|
|
|
182 |
moderation_result = moderate_text(query)
|
183 |
if not moderation_result.is_safe:
|
184 |
return "Sorry, this query contains harmful or inappropriate content."
|
185 |
|
186 |
+
classification = classify_query(moderation_result.original_text)
|
|
|
187 |
|
188 |
+
if classification == "OutOfScope":
|
|
|
189 |
refusal_text = refusal_chain.run({"topic": "this topic"})
|
190 |
return tailor_chain.run({"response": refusal_text}).strip()
|
191 |
|
192 |
+
if classification == "Wellness":
|
193 |
rag_result = wellness_rag_chain({"query": moderation_result.original_text})
|
194 |
csv_answer = rag_result["result"].strip()
|
195 |
web_answer = "" if csv_answer else do_web_search(moderation_result.original_text)
|
196 |
final_merged = merge_responses(csv_answer, web_answer)
|
197 |
return tailor_chain.run({"response": final_merged}).strip()
|
198 |
|
199 |
+
if classification == "Brand":
|
200 |
rag_result = brand_rag_chain({"query": moderation_result.original_text})
|
201 |
csv_answer = rag_result["result"].strip()
|
202 |
final_merged = merge_responses(csv_answer, "")
|
203 |
return tailor_chain.run({"response": final_merged}).strip()
|
204 |
|
|
|
205 |
refusal_text = refusal_chain.run({"topic": "this topic"})
|
206 |
return tailor_chain.run({"response": refusal_text}).strip()
|
207 |
|
|
|
|
|
|
|
208 |
# Initialize chains and vectorstores
|
209 |
try:
|
|
|
210 |
classification_chain = get_classification_chain()
|
211 |
refusal_chain = get_refusal_chain()
|
212 |
tailor_chain = get_tailor_chain()
|
213 |
cleaner_chain = get_cleaner_chain()
|
214 |
|
|
|
215 |
wellness_csv = "AIChatbot.csv"
|
216 |
brand_csv = "BrandAI.csv"
|
217 |
wellness_store_dir = "faiss_wellness_store"
|
218 |
brand_store_dir = "faiss_brand_store"
|
219 |
|
|
|
220 |
wellness_vectorstore = build_or_load_vectorstore(wellness_csv, wellness_store_dir)
|
221 |
brand_vectorstore = build_or_load_vectorstore(brand_csv, brand_store_dir)
|
222 |
|
|
|
223 |
gemini_llm = LiteLLMModel(model_id="gemini/gemini-pro", api_key=os.environ.get("GEMINI_API_KEY"))
|
224 |
wellness_rag_chain = build_rag_chain(gemini_llm, wellness_vectorstore)
|
225 |
brand_rag_chain = build_rag_chain(gemini_llm, brand_vectorstore)
|
226 |
+
|
227 |
print("Pipeline initialized successfully!")
|
228 |
except Exception as e:
|
229 |
print(f"Error initializing pipeline: {str(e)}")
|
|
|
|
|
|