48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369 | class CondenseQuestionChatEngine(BaseChatEngine):
"""Condense Question Chat Engine.
First generate a standalone question from conversation context and last message,
then query the query engine for a response.
"""
def __init__(
self,
query_engine: BaseQueryEngine,
condense_question_prompt: BasePromptTemplate,
memory: BaseMemory,
llm: LLMPredictorType,
verbose: bool = False,
callback_manager: Optional[CallbackManager] = None,
) -> None:
self._query_engine = query_engine
self._condense_question_prompt = condense_question_prompt
self._memory = memory
self._llm = llm
self._verbose = verbose
self.callback_manager = callback_manager or CallbackManager([])
@classmethod
def from_defaults(
cls,
query_engine: BaseQueryEngine,
condense_question_prompt: Optional[BasePromptTemplate] = None,
chat_history: Optional[List[ChatMessage]] = None,
memory: Optional[BaseMemory] = None,
memory_cls: Type[BaseMemory] = ChatMemoryBuffer,
service_context: Optional[ServiceContext] = None,
verbose: bool = False,
system_prompt: Optional[str] = None,
prefix_messages: Optional[List[ChatMessage]] = None,
llm: Optional[LLM] = None,
**kwargs: Any,
) -> "CondenseQuestionChatEngine":
"""Initialize a CondenseQuestionChatEngine from default parameters."""
condense_question_prompt = condense_question_prompt or DEFAULT_PROMPT
llm = llm or llm_from_settings_or_context(Settings, service_context)
chat_history = chat_history or []
memory = memory or memory_cls.from_defaults(chat_history=chat_history, llm=llm)
if system_prompt is not None:
raise NotImplementedError(
"system_prompt is not supported for CondenseQuestionChatEngine."
)
if prefix_messages is not None:
raise NotImplementedError(
"prefix_messages is not supported for CondenseQuestionChatEngine."
)
return cls(
query_engine,
condense_question_prompt,
memory,
llm,
verbose=verbose,
callback_manager=callback_manager_from_settings_or_context(
Settings, service_context
),
)
def _condense_question(
self, chat_history: List[ChatMessage], last_message: str
) -> str:
"""
Generate standalone question from conversation context and last message.
"""
chat_history_str = messages_to_history_str(chat_history)
logger.debug(chat_history_str)
return self._llm.predict(
self._condense_question_prompt,
question=last_message,
chat_history=chat_history_str,
)
async def _acondense_question(
self, chat_history: List[ChatMessage], last_message: str
) -> str:
"""
Generate standalone question from conversation context and last message.
"""
chat_history_str = messages_to_history_str(chat_history)
logger.debug(chat_history_str)
return await self._llm.apredict(
self._condense_question_prompt,
question=last_message,
chat_history=chat_history_str,
)
def _get_tool_output_from_response(
self, query: str, response: RESPONSE_TYPE
) -> ToolOutput:
if isinstance(response, StreamingResponse):
return ToolOutput(
content="",
tool_name="query_engine",
raw_input={"query": query},
raw_output=response,
)
else:
return ToolOutput(
content=str(response),
tool_name="query_engine",
raw_input={"query": query},
raw_output=response,
)
@trace_method("chat")
def chat(
self, message: str, chat_history: Optional[List[ChatMessage]] = None
) -> AgentChatResponse:
chat_history = chat_history or self._memory.get()
# Generate standalone question from conversation context and last message
condensed_question = self._condense_question(chat_history, message)
log_str = f"Querying with: {condensed_question}"
logger.info(log_str)
if self._verbose:
print(log_str)
# TODO: right now, query engine uses class attribute to configure streaming,
# we are moving towards separate streaming and non-streaming methods.
# In the meanwhile, use this hack to toggle streaming.
from llama_index.core.query_engine.retriever_query_engine import (
RetrieverQueryEngine,
)
if isinstance(self._query_engine, RetrieverQueryEngine):
is_streaming = self._query_engine._response_synthesizer._streaming
self._query_engine._response_synthesizer._streaming = False
# Query with standalone question
query_response = self._query_engine.query(condensed_question)
# NOTE: reset streaming flag
if isinstance(self._query_engine, RetrieverQueryEngine):
self._query_engine._response_synthesizer._streaming = is_streaming
tool_output = self._get_tool_output_from_response(
condensed_question, query_response
)
# Record response
self._memory.put(ChatMessage(role=MessageRole.USER, content=message))
self._memory.put(
ChatMessage(role=MessageRole.ASSISTANT, content=str(query_response))
)
return AgentChatResponse(response=str(query_response), sources=[tool_output])
@trace_method("chat")
def stream_chat(
self, message: str, chat_history: Optional[List[ChatMessage]] = None
) -> StreamingAgentChatResponse:
chat_history = chat_history or self._memory.get()
# Generate standalone question from conversation context and last message
condensed_question = self._condense_question(chat_history, message)
log_str = f"Querying with: {condensed_question}"
logger.info(log_str)
if self._verbose:
print(log_str)
# TODO: right now, query engine uses class attribute to configure streaming,
# we are moving towards separate streaming and non-streaming methods.
# In the meanwhile, use this hack to toggle streaming.
from llama_index.core.query_engine.retriever_query_engine import (
RetrieverQueryEngine,
)
if isinstance(self._query_engine, RetrieverQueryEngine):
is_streaming = self._query_engine._response_synthesizer._streaming
self._query_engine._response_synthesizer._streaming = True
# Query with standalone question
query_response = self._query_engine.query(condensed_question)
# NOTE: reset streaming flag
if isinstance(self._query_engine, RetrieverQueryEngine):
self._query_engine._response_synthesizer._streaming = is_streaming
tool_output = self._get_tool_output_from_response(
condensed_question, query_response
)
# Record response
if (
isinstance(query_response, StreamingResponse)
and query_response.response_gen is not None
):
# override the generator to include writing to chat history
self._memory.put(ChatMessage(role=MessageRole.USER, content=message))
response = StreamingAgentChatResponse(
chat_stream=response_gen_from_query_engine(query_response.response_gen),
sources=[tool_output],
)
thread = Thread(
target=response.write_response_to_history,
args=(self._memory,),
)
thread.start()
else:
raise ValueError("Streaming is not enabled. Please use chat() instead.")
return response
@trace_method("chat")
async def achat(
self, message: str, chat_history: Optional[List[ChatMessage]] = None
) -> AgentChatResponse:
chat_history = chat_history or self._memory.get()
# Generate standalone question from conversation context and last message
condensed_question = await self._acondense_question(chat_history, message)
log_str = f"Querying with: {condensed_question}"
logger.info(log_str)
if self._verbose:
print(log_str)
# TODO: right now, query engine uses class attribute to configure streaming,
# we are moving towards separate streaming and non-streaming methods.
# In the meanwhile, use this hack to toggle streaming.
from llama_index.core.query_engine.retriever_query_engine import (
RetrieverQueryEngine,
)
if isinstance(self._query_engine, RetrieverQueryEngine):
is_streaming = self._query_engine._response_synthesizer._streaming
self._query_engine._response_synthesizer._streaming = False
# Query with standalone question
query_response = await self._query_engine.aquery(condensed_question)
# NOTE: reset streaming flag
if isinstance(self._query_engine, RetrieverQueryEngine):
self._query_engine._response_synthesizer._streaming = is_streaming
tool_output = self._get_tool_output_from_response(
condensed_question, query_response
)
# Record response
self._memory.put(ChatMessage(role=MessageRole.USER, content=message))
self._memory.put(
ChatMessage(role=MessageRole.ASSISTANT, content=str(query_response))
)
return AgentChatResponse(response=str(query_response), sources=[tool_output])
@trace_method("chat")
async def astream_chat(
self, message: str, chat_history: Optional[List[ChatMessage]] = None
) -> StreamingAgentChatResponse:
chat_history = chat_history or self._memory.get()
# Generate standalone question from conversation context and last message
condensed_question = await self._acondense_question(chat_history, message)
log_str = f"Querying with: {condensed_question}"
logger.info(log_str)
if self._verbose:
print(log_str)
# TODO: right now, query engine uses class attribute to configure streaming,
# we are moving towards separate streaming and non-streaming methods.
# In the meanwhile, use this hack to toggle streaming.
from llama_index.core.query_engine.retriever_query_engine import (
RetrieverQueryEngine,
)
if isinstance(self._query_engine, RetrieverQueryEngine):
is_streaming = self._query_engine._response_synthesizer._streaming
self._query_engine._response_synthesizer._streaming = True
# Query with standalone question
query_response = await self._query_engine.aquery(condensed_question)
# NOTE: reset streaming flag
if isinstance(self._query_engine, RetrieverQueryEngine):
self._query_engine._response_synthesizer._streaming = is_streaming
tool_output = self._get_tool_output_from_response(
condensed_question, query_response
)
# Record response
if (
isinstance(query_response, StreamingResponse)
and query_response.response_gen is not None
):
# override the generator to include writing to chat history
# TODO: query engine does not support async generator yet
self._memory.put(ChatMessage(role=MessageRole.USER, content=message))
response = StreamingAgentChatResponse(
chat_stream=response_gen_from_query_engine(query_response.response_gen),
sources=[tool_output],
)
thread = Thread(
target=response.write_response_to_history, args=(self._memory,)
)
thread.start()
else:
raise ValueError("Streaming is not enabled. Please use achat() instead.")
return response
def reset(self) -> None:
# Clear chat history
self._memory.reset()
@property
def chat_history(self) -> List[ChatMessage]:
"""Get chat history."""
return self._memory.get_all()
|