@ancient-wolf-19325 When I run a workflow with new paramaters I get error

Request rejected by the API, due to Invalid input. Details: train.setup task with different structure already exists.

I need to be able to run the same workflow with different input parameters

The error you're encountering is due to Flyte's versioning mechanism, which identifies tasks and workflows by their structure and version. When you change input parameters, Flyte sees it as a different structure and requires a new version. To resolve this, you can specify a new version when registering the workflow. This can be done using

pyflyte register

or

flytectl register

commands. Here's an example of how to do it: 1. Package the workflow:

pyflyte package

2. Register the workflow with a new version:

flytectl register files --project <project> --domain <domain> --version <new_version> --archive flyte-package.tgz

Alternatively, you can use

pyflyte register

directly:

pyflyte register --project <project> --domain <domain> --version <new_version>

Would you like more details on this process or any other aspect of Flyte? Sources: - https://github.com/flyteorg/flyte/issues/5426 - https://github.com/flyteorg/flyte/issues/5415 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.

@ancient-wolf-19325 how can I keep same structure with varying inputs

To keep the same workflow structure while varying inputs, you should avoid re-registering the workflow for each input change. Instead, you can use Flyte's dynamic workflows or launch plans to handle varying inputs without changing the workflow structure. 1. Dynamic Workflows: Use dynamic workflows to handle varying inputs at runtime.

from flytekit import task, dynamic, workflow

   @task
   def process_data(input: str) -> str:
       return f"Processed {input}"

   @dynamic
   def dynamic_wf(inputs: list[str]):
       results = []
       for input in inputs:
           results.append(process_data(input=input))
       return results

   @workflow
   def wf(inputs: list[str]):
       return dynamic_wf(inputs=inputs)

2. Launch Plans: Create a launch plan with default inputs and override them at execution time.

from flytekit import LaunchPlan, workflow

   @workflow
   def wf(input: str):
       process_data(input=input)

   launch_plan = LaunchPlan.create(
       "wf_launch_plan",
       wf,
       default_inputs={"input": "default_value"}
   )

You can then override the input when executing the launch plan:

flytectl create execution --project <project> --domain <domain> --name <name> --launchplan wf_launch_plan --inputs '{"input": "new_value"}'

Would you like more details on dynamic workflows or launch plans? Sources: - Flyte Dynamic Workflows - Flyte Launch Plans 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.

@salmon-flower-36598 see above