Feat/memory impl#157
Conversation
|
|
||
| async def chat( | ||
| self, | ||
| messages: list[ArkMessage], |
There was a problem hiding this comment.
我觉得ArkMessage其实叫Message就挺好的，加个Ark有点怪。
对等的，比如OpenaiMessage or VetexAIMessage听起来是不是都有点怪..
不过这个还好改吗
There was a problem hiding this comment.
哎这个结构由来已久了 感觉不太好改了
There was a problem hiding this comment.
这样呢？只留一个 def ArkMessage(Message) 的定义，其他用户代码如果引用了 ArkMessage 可以继续用，但我们自己的代码里，就全都改成直接用 Message
| ] | ||
|
|
||
|
|
||
| class ResponsesClientWithLongTermMemory: |
There was a problem hiding this comment.
感觉 ResponsesClientWithLongTermMemory 可以作为一个基础组件，可以从 examples 里摘出来
There was a problem hiding this comment.
+1 先做为了一个参考实现，consolidate下放进去
| memories = self.memory[user_id] | ||
| results = "用户过去的交互记录\n\n" | ||
| for memory in memories: | ||
| content = format_ark_message_as_string(memory) |
There was a problem hiding this comment.
同理，这里没有必要把 ark 也带上；
另外，用户的原始 query 是不是最好也能带上呢
There was a problem hiding this comment.
嗯嗯在update memory的时候 这个list里面默认是带用户原始query的
| new_messages: list[ArkMessage | dict | Response | ChatCompletionMessage], | ||
| **kwargs: Any, | ||
| ) -> None: | ||
| pass |
There was a problem hiding this comment.
add_or_update有点冗长，叫upsert_memory 呢？
然后接口感觉可以扩展成
@abstractmethod async def upsert_memory( self, user_id: str, data: list[Any], source: DataSource = DataSource.CHAT_MESSAGE, # 将来可以扩展更多的入库的数据，不单单局限在对话的message **kwargs: Any, ) -> None: ...
从而将来可以接入更多数据源，而不单单是对话消息
There was a problem hiding this comment.
嗯嗯 我觉得要不先把new_messages 改为data: list[Any] 吧，其他source 之后需要加了再加上？暂时不放在base 的interface里
不一定强求所有实现都可以支持各种datasource，但来自message的应该是都会支持的
There was a problem hiding this comment.
好，我其实想留一个 datasource 是想让接口看起来不是专门针对对话消息的，将来可能可以继续扩展出来各种生产资料，比如doc啥的
wgzesg-bd
force-pushed
the
feat/memory_impl
branch
from
5b04309 to
ab6e8f2
Compare
2 participants
No description provided.