You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
81 lines
2.9 KiB
81 lines
2.9 KiB
2 years ago
|
from typing import Any, Dict, List
|
||
|
|
|
||
|
|
from langchain.chains.llm import LLMChain
|
||
|
|
from langchain.memory.chat_memory import BaseChatMemory
|
||
|
|
from langchain.memory.prompt import SUMMARY_PROMPT
|
||
|
|
from langchain.prompts.base import BasePromptTemplate
|
||
|
|
from langchain.schema import BaseLanguageModel, BaseMessage, get_buffer_string
|
||
|
|
|
||
|
|
|
||
|
|
class BotConversationSummerBufferWindowMemory(BaseChatMemory):
|
||
|
|
"""Buffer for storing conversation memory."""
|
||
|
|
|
||
|
|
human_prefix: str = "Human"
|
||
|
|
ai_prefix: str = "AI"
|
||
|
|
# Define key to pass information about entities into prompt.
|
||
|
|
memory_key: str = "history" #: :meta private:
|
||
|
|
#k: int = 5
|
||
|
|
max_token_limit: int = 1200
|
||
|
|
min_token_limit: int = 200
|
||
|
|
moving_summary_buffer: str = ""
|
||
|
|
|
||
|
|
llm: BaseLanguageModel
|
||
|
|
summary_prompt: BasePromptTemplate = SUMMARY_PROMPT
|
||
|
|
|
||
|
|
@property
|
||
|
|
def buffer(self) -> List[BaseMessage]:
|
||
|
|
"""String buffer of memory."""
|
||
|
|
return self.chat_memory.messages
|
||
|
|
|
||
|
|
@property
|
||
|
|
def memory_variables(self) -> List[str]:
|
||
|
|
"""Will always return list of memory variables.
|
||
|
|
:meta private:
|
||
|
|
"""
|
||
|
|
return [self.memory_key]
|
||
|
|
|
||
|
|
def load_memory_variables(self, inputs: Dict[str, Any]) -> Dict[str, str]:
|
||
|
|
"""Return history buffer."""
|
||
|
|
buffer = self.buffer
|
||
|
|
#buffer: Any = self.buffer[-self.k * 2 :] if self.k > 0 else []
|
||
|
|
if not self.return_messages:
|
||
|
|
buffer = get_buffer_string(
|
||
|
|
buffer,
|
||
|
|
human_prefix=self.human_prefix,
|
||
|
|
ai_prefix=self.ai_prefix,
|
||
|
|
)
|
||
|
|
return {self.memory_key: buffer}
|
||
|
|
|
||
|
|
def save_context(self, inputs: Dict[str, Any], outputs: Dict[str, str]) -> None:
|
||
|
|
"""Save context from this conversation to buffer. Pruned."""
|
||
|
|
super().save_context(inputs, outputs)
|
||
|
|
# Prune buffer if it exceeds max token limit
|
||
|
|
buffer = self.chat_memory.messages
|
||
|
|
curr_buffer_length = self.llm.get_num_tokens_from_messages(buffer)
|
||
|
|
if curr_buffer_length > self.max_token_limit:
|
||
|
|
pruned_memory = []
|
||
|
|
while curr_buffer_length > self.min_token_limit:
|
||
|
|
pruned_memory.append(buffer.pop(0))
|
||
|
|
curr_buffer_length = self.llm.get_num_tokens_from_messages(buffer)
|
||
|
|
self.moving_summary_buffer = self.predict_new_summary(
|
||
|
|
pruned_memory, self.moving_summary_buffer
|
||
|
|
)
|
||
|
|
|
||
|
|
def clear(self) -> None:
|
||
|
|
"""Clear memory contents."""
|
||
|
|
super().clear()
|
||
|
|
self.moving_summary_buffer = ""
|
||
|
|
|
||
|
|
|
||
|
|
def predict_new_summary(
|
||
|
|
self, messages: List[BaseMessage], existing_summary: str
|
||
|
|
) -> str:
|
||
|
|
new_lines = get_buffer_string(
|
||
|
|
messages,
|
||
|
|
human_prefix=self.human_prefix,
|
||
|
|
ai_prefix=self.ai_prefix,
|
||
|
|
)
|
||
|
|
|
||
|
|
chain = LLMChain(llm=self.llm, prompt=self.summary_prompt)
|
||
|
|
return chain.predict(summary=existing_summary, new_lines=new_lines)
|