Support evaluating runnables and arbitrary functions

[hinthornw]

Added a couple of "integration tests" for these that I ran.

Main design point of feedback: at this point, would it just be better to have separate arguments for each type? Little confusing what is or isn't supported and what is the intended usage at this point since I try to wrap the function as runnable or pack or unpack chains/llms.

run_on_dataset(
...
llm_or_chain_factory = None,
llm = None,
chain = NOne,
runnable=None,
function=None
):
# raise error if none set

Downside with runnables and arbitrary function support is that you get much less helpful validation and error messages, but I don't think we should block you from this, at least.

[vercel]

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

1 Ignored Deployment

Name	Status	Preview	Comments	Updated (UTC)
langchain	⬜️ Ignored (Inspect)	Visit Preview		Aug 3, 2023 6:35pm

[hwchase17]

im reviewing super fast but is it possible that it may be cleaner to branch at a higher level rather than pass around the duck type everywhere?

[hinthornw]

Probably - i'll think about it. Would want to still retain the key checking where available