Llama index workflows: modularization support and manua...

At a glance

The community members are discussing whether Llama Index Workflows supports modularization out of the box, or if it requires a more manual approach. One community member suggests that you can define a function with the logic and call it with *args and **kwargs as parameters. Another community member confirms this is possible, and provides an example of using the @step decorator to define a free function as a workflow step. However, they note that the Workflow class is required for this approach, which raises a validation error in some cases. The community members are open to feedback on the user experience of this feature.

Useful resources

ccoder.ve

Hello, does Llama Index Workflows support modularization out of the box (splitting workflow steps into multiple files), or does it currently require a more manual approach? For example, defining a function with all the logic and then calling it with *args and **kwargs as parameters?

6 comments

ccoder.ve

Yes, that makes sense! I was curious if there’s a way to send a list of functions for the workflow to interpret as steps, rather than handling each step individually. But it’s not a big issue—just something I was curious about. Thank you! 🙂

LLogan M

Oh! You can do that as well (but your step definition will will have to follow the usual syntax of async def some_fn(ev: SomeEvent) -> SomeOtherEvent or async def some_fn(ctx: Context, ev: SomeEvent) -> SomeOtherEvent

For example

Plain Text

from llama_index.core.workflow import Context, StartEvent, StopEvent, Workflow, step

@step
def some_fn(ctx: Context, ev: StartEvent) -> StopEvent:
  return StopEvent(result="This came from a free function!")

w = Workflow()

w.add_step(some_fn)

LLogan M

I haven't seen this used in the wild tbh, so super open to feedback on the UX here

ccoder.ve

Thank you so much 🤘

ccoder.ve

qq: the step decorator needs a Workflow class to free functions, so if I pass llama_index.core.workflow.Workflow and then use the add_step instance method in a workflow instance, that is raising a validation error here -> https://github.com/run-llama/llama_index/blob/ef7207247e0884d48b05c85a710634b43dc04b05/llama-index-core/llama_index/core/workflow/workflow.py#L142

Is that the expected behaviour?

LLogan M

Ah ok, that explains why the example in the docs passes in a workflow instance

Dang, I thought I could skip that because it felt janky lol
https://docs.llamaindex.ai/en/stable/module_guides/workflow/#decorating-non-class-functions

Add a reply

Find answers from the community

Llama index workflows: modularization support and manual approaches