SequentialChain that runs next chain on each output of the prior chain?

[jmontroy90]

Given:

• Chain A generates N outputs, e.g. we get back the text and immediately split it into a list based on post-processing we control and expect.
• Chain B should then run over each of those outputs from Chain A.

Is there any way to do this elegantly in langchain? Perhaps some way to provide an output formatter to a chain, or some for_each pre- / post-processor? Or does this seem like just two independent chains with processing in between?

[hwchase17]

@jmontroy90 im super glad you asked this question - i've been noodling around on some stuff but looking for some user feedback before landing.

im thinking that it would make sense to have a postprocessor for a prompt - since the return format of the LLM is dictated by the prompt. you could have many types of postprocessors, but one of the most common ones is probably a list. i have a branch that implements something like this, and here's a notebook using it: https://github.com/hwchase17/langchain/blob/harrison/use_output_parser/docs/examples/chains/llm_for_loop.ipynb

what do you think? would that cover your use case? happy to brainstorm/work on this together more - its def a common use case so id love to get it right

[jmontroy90]

Ah, lovely, just like that. I can try to pull down your branch and play with it, see how it works for us. Thank you!

[hwchase17]

Ah, lovely, just like that. I can try to pull down your branch and play with it, see how it works for us. Thank you!

perfect - let me know what you think!