Core: Added partial functionality to PipelinePromptTemplate

[Aarya2004]

Description: Added extra logic in the PipelinePromptTemplate to support partial assignments primarily by overriding the partial method from BasePromptTemplate
Issue: Fixes issue #27868
Dependencies: No dependencies required or new dependencies added

Add tests
Added test in test_pipeline_prompt.py to check for this case.

Lint and test: Done

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

1 Skipped Deployment

Name	Status	Preview	Comments	Updated (UTC)
langchain	⬜️ Ignored (Inspect)	Visit Preview		Nov 27, 2024 5:33pm

[Aarya2004]

Hello @efriis! Sorry for the ping, but you had reviewed a PR related to Prompt Templates recently (#28011) so I was hoping you would be able to review this too. Let me know if you have any questions, thank you!

[efriis]

I think this is ok. Quickly discussing with team to see what we want to do with PipelinePromptTemplate because it's not that intuitive of an abstraction, and it might make sense to just implement as a RunnableSequence of RunnableAssigns instead...

i.e. I think this does the same thing from my understanding:

keys_and_prompts: list[tuple[str, PromptTemplate]] = ...


chain = RunnablePassthrough()
for key, prompt in keys_and_prompts:
  chain = chain | RunnablePassthrough.assign(key, prompt)

chain = chain | lambda x: x[keys_and_prompts[-1][0]]

[Aarya2004]

Sounds good, I wasn't sure if this complied with how PipelinePromptTemplate is 'supposed' to be used but I thought it made sense as a functionality given that the PromptTemplate class implements this function (and PipelinePromptTemplate used a list of these PromptTemplates for its functionality) but I do see how it makes sense through your code. Thanks for taking a look at the code, let me know what you and the team decide!

[efriis]

Are you using it personally? If so could you check if the code snippet solves your use case?

[Aarya2004]

No I'm not, I'm just trying to fix the issue #27868

[efriis]

got it. I can respond there, and I'll close this because it introduces a new bug, where partial implements a different interface (stateful mutation instead of stateless wrap)

[Aarya2004]

Thank you!