core: Add aformat_messages to FewShotChatMessagePromptTemplate and ChatPromptTemplate #19648
Conversation
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎ 1 Ignored Deployment
|
dosubot
bot
added
size:L
| @@ -901,19 +929,31 @@ def from_messages( | |||
| partial_variables=partial_vars, | |||
| ) | |||
|
|
|||
| def format(self, **kwargs: Any) -> str: | |||
There was a problem hiding this comment.
Removed format from ChatPromptTemplate as it is identical to the one in the parent class BaseChatPromptTemplate
cbornet
force-pushed
the
async-FewShotChatMessagePromptTemplate
branch
2 times, most recently
from
1073750 to
8ae8e21
Compare
cbornet
force-pushed
the
async-FewShotChatMessagePromptTemplate
branch
from
8ae8e21 to
c9ee89f
Compare
| Returns: | ||
| List of BaseMessages. | ||
| """ | ||
| return await run_in_executor(None, self.format_messages, **kwargs) |
There was a problem hiding this comment.
This is another place where I feel like we need to delegate to the sync implementation.
I'm a bit worried about having this dichotomy / how to make it obvious that this is is what's going on
There was a problem hiding this comment.
cc @baskaryan / @nfcampos
Do we keep run_in_executor for consistency or use a blocking call to the sync code?
There was a problem hiding this comment.
Another PR where this question comes up: https://github.com/langchain-ai/langchain/pull/19652/files
There was a problem hiding this comment.
This is the part that makes it all weird:
Chat prompt templates are constructed of multiple messages each of which can be formatted independently -- this
would mean that interpolating a few variables into essentially a list of dicts we're launching several threads
There was a problem hiding this comment.
Another possibility if we don't want to choose now is to raise NotImplementedError in the default interface.
There was a problem hiding this comment.
I intend to implement all subclasses in the end so the default interface won't be called anyway.
There was a problem hiding this comment.
a list of dicts we're launching several threads
Actually, it's not launching several threads, it's submitting tasks to the default thread-pool (which has 1 thread per core).
There was a problem hiding this comment.
Anyway, I updated the PR to call the sync method directly. PTAL
dosubot
bot
added
the
lgtm
… and ChatPromptTemplate (langchain-ai#19648) Needed since the example selector may use a vector store.
… and ChatPromptTemplate (#19648) Needed since the example selector may use a vector store.
Needed since the example selector may use a vector store.