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
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393 | class ContextChatEngine(BaseChatEngine):
"""
Context Chat Engine.
Uses a retriever to retrieve a context, set the context in the system prompt,
and then uses an LLM to generate a response, for a fluid chat experience.
"""
def __init__(
self,
retriever: BaseRetriever,
llm: LLM,
memory: BaseMemory,
prefix_messages: List[ChatMessage],
node_postprocessors: Optional[List[BaseNodePostprocessor]] = None,
context_template: Optional[Union[str, PromptTemplate]] = None,
context_refine_template: Optional[Union[str, PromptTemplate]] = None,
callback_manager: Optional[CallbackManager] = None,
) -> None:
self._retriever = retriever
self._llm = llm
self._memory = memory
self._prefix_messages = prefix_messages
self._node_postprocessors = node_postprocessors or []
context_template = context_template or DEFAULT_CONTEXT_TEMPLATE
if isinstance(context_template, str):
context_template = PromptTemplate(context_template)
self._context_template = context_template
context_refine_template = context_refine_template or DEFAULT_REFINE_TEMPLATE
if isinstance(context_refine_template, str):
context_refine_template = PromptTemplate(context_refine_template)
self._context_refine_template = context_refine_template
self.callback_manager = callback_manager or CallbackManager([])
for node_postprocessor in self._node_postprocessors:
node_postprocessor.callback_manager = self.callback_manager
@classmethod
def from_defaults(
cls,
retriever: BaseRetriever,
chat_history: Optional[List[ChatMessage]] = None,
memory: Optional[BaseMemory] = None,
system_prompt: Optional[str] = None,
prefix_messages: Optional[List[ChatMessage]] = None,
node_postprocessors: Optional[List[BaseNodePostprocessor]] = None,
context_template: Optional[Union[str, PromptTemplate]] = None,
context_refine_template: Optional[Union[str, PromptTemplate]] = None,
llm: Optional[LLM] = None,
**kwargs: Any,
) -> "ContextChatEngine":
"""Initialize a ContextChatEngine from default parameters."""
llm = llm or Settings.llm
chat_history = chat_history or []
memory = memory or ChatMemoryBuffer.from_defaults(
chat_history=chat_history, token_limit=llm.metadata.context_window - 256
)
if system_prompt is not None:
if prefix_messages is not None:
raise ValueError(
"Cannot specify both system_prompt and prefix_messages"
)
prefix_messages = [
ChatMessage(content=system_prompt, role=llm.metadata.system_role)
]
prefix_messages = prefix_messages or []
node_postprocessors = node_postprocessors or []
return cls(
retriever,
llm=llm,
memory=memory,
prefix_messages=prefix_messages,
node_postprocessors=node_postprocessors,
callback_manager=Settings.callback_manager,
context_template=context_template,
context_refine_template=context_refine_template,
)
def _get_nodes(self, message: str) -> List[NodeWithScore]:
"""Generate context information from a message."""
nodes = self._retriever.retrieve(message)
for postprocessor in self._node_postprocessors:
nodes = postprocessor.postprocess_nodes(
nodes, query_bundle=QueryBundle(message)
)
return nodes
async def _aget_nodes(self, message: str) -> List[NodeWithScore]:
"""Generate context information from a message."""
nodes = await self._retriever.aretrieve(message)
for postprocessor in self._node_postprocessors:
nodes = postprocessor.postprocess_nodes(
nodes, query_bundle=QueryBundle(message)
)
return nodes
def _get_response_synthesizer(
self, chat_history: List[ChatMessage], streaming: bool = False
) -> CompactAndRefine:
# Pull the system prompt from the prefix messages
system_prompt = ""
prefix_messages = self._prefix_messages
if (
len(self._prefix_messages) != 0
and self._prefix_messages[0].role == MessageRole.SYSTEM
):
system_prompt = str(self._prefix_messages[0].content)
prefix_messages = self._prefix_messages[1:]
# Get the messages for the QA and refine prompts
qa_messages = get_prefix_messages_with_context(
self._context_template,
system_prompt,
prefix_messages,
chat_history,
self._llm.metadata.system_role,
)
refine_messages = get_prefix_messages_with_context(
self._context_refine_template,
system_prompt,
prefix_messages,
chat_history,
self._llm.metadata.system_role,
)
# Get the response synthesizer
return get_response_synthesizer(
self._llm,
self.callback_manager,
qa_messages,
refine_messages,
streaming,
qa_function_mappings=self._context_template.function_mappings,
refine_function_mappings=self._context_refine_template.function_mappings,
)
@trace_method("chat")
def chat(
self,
message: str,
chat_history: Optional[List[ChatMessage]] = None,
prev_chunks: Optional[List[NodeWithScore]] = None,
) -> AgentChatResponse:
if chat_history is not None:
self._memory.set(chat_history)
# get nodes and postprocess them
nodes = self._get_nodes(message)
if len(nodes) == 0 and prev_chunks is not None:
nodes = prev_chunks
# Get the response synthesizer with dynamic prompts
chat_history = self._memory.get(
input=message,
)
synthesizer = self._get_response_synthesizer(chat_history)
response = synthesizer.synthesize(message, nodes)
user_message = ChatMessage(content=message, role=MessageRole.USER)
ai_message = ChatMessage(content=str(response), role=MessageRole.ASSISTANT)
self._memory.put(user_message)
self._memory.put(ai_message)
return AgentChatResponse(
response=str(response),
sources=[
ToolOutput(
tool_name="retriever",
content=str(nodes),
raw_input={"message": message},
raw_output=nodes,
)
],
source_nodes=nodes,
)
@trace_method("chat")
def stream_chat(
self,
message: str,
chat_history: Optional[List[ChatMessage]] = None,
prev_chunks: Optional[List[NodeWithScore]] = None,
) -> StreamingAgentChatResponse:
if chat_history is not None:
self._memory.set(chat_history)
# get nodes and postprocess them
nodes = self._get_nodes(message)
if len(nodes) == 0 and prev_chunks is not None:
nodes = prev_chunks
# Get the response synthesizer with dynamic prompts
chat_history = self._memory.get(
input=message,
)
synthesizer = self._get_response_synthesizer(chat_history, streaming=True)
response = synthesizer.synthesize(message, nodes)
assert isinstance(response, StreamingResponse)
def wrapped_gen(response: StreamingResponse) -> ChatResponseGen:
full_response = ""
for token in response.response_gen:
full_response += token
yield ChatResponse(
message=ChatMessage(
content=full_response, role=MessageRole.ASSISTANT
),
delta=token,
)
user_message = ChatMessage(content=message, role=MessageRole.USER)
ai_message = ChatMessage(content=full_response, role=MessageRole.ASSISTANT)
self._memory.put(user_message)
self._memory.put(ai_message)
return StreamingAgentChatResponse(
chat_stream=wrapped_gen(response),
sources=[
ToolOutput(
tool_name="retriever",
content=str(nodes),
raw_input={"message": message},
raw_output=nodes,
)
],
source_nodes=nodes,
is_writing_to_memory=False,
)
@trace_method("chat")
async def achat(
self,
message: str,
chat_history: Optional[List[ChatMessage]] = None,
prev_chunks: Optional[List[NodeWithScore]] = None,
) -> AgentChatResponse:
if chat_history is not None:
self._memory.set(chat_history)
# get nodes and postprocess them
nodes = await self._aget_nodes(message)
if len(nodes) == 0 and prev_chunks is not None:
nodes = prev_chunks
# Get the response synthesizer with dynamic prompts
chat_history = self._memory.get(
input=message,
)
synthesizer = self._get_response_synthesizer(chat_history)
response = await synthesizer.asynthesize(message, nodes)
user_message = ChatMessage(content=message, role=MessageRole.USER)
ai_message = ChatMessage(content=str(response), role=MessageRole.ASSISTANT)
await self._memory.aput(user_message)
await self._memory.aput(ai_message)
return AgentChatResponse(
response=str(response),
sources=[
ToolOutput(
tool_name="retriever",
content=str(nodes),
raw_input={"message": message},
raw_output=nodes,
)
],
source_nodes=nodes,
)
@trace_method("chat")
async def astream_chat(
self,
message: str,
chat_history: Optional[List[ChatMessage]] = None,
prev_chunks: Optional[List[NodeWithScore]] = None,
) -> StreamingAgentChatResponse:
if chat_history is not None:
self._memory.set(chat_history)
# get nodes and postprocess them
nodes = await self._aget_nodes(message)
if len(nodes) == 0 and prev_chunks is not None:
nodes = prev_chunks
# Get the response synthesizer with dynamic prompts
chat_history = self._memory.get(
input=message,
)
synthesizer = self._get_response_synthesizer(chat_history, streaming=True)
response = await synthesizer.asynthesize(message, nodes)
assert isinstance(response, AsyncStreamingResponse)
async def wrapped_gen(response: AsyncStreamingResponse) -> ChatResponseAsyncGen:
full_response = ""
async for token in response.async_response_gen():
full_response += token
yield ChatResponse(
message=ChatMessage(
content=full_response, role=MessageRole.ASSISTANT
),
delta=token,
)
user_message = ChatMessage(content=message, role=MessageRole.USER)
ai_message = ChatMessage(content=full_response, role=MessageRole.ASSISTANT)
await self._memory.aput(user_message)
await self._memory.aput(ai_message)
return StreamingAgentChatResponse(
achat_stream=wrapped_gen(response),
sources=[
ToolOutput(
tool_name="retriever",
content=str(nodes),
raw_input={"message": message},
raw_output=nodes,
)
],
source_nodes=nodes,
is_writing_to_memory=False,
)
def reset(self) -> None:
self._memory.reset()
@property
def chat_history(self) -> List[ChatMessage]:
"""Get chat history."""
return self._memory.get_all()
|
