LLM Finetuning template updates in examples/llm_finetuning have been pushed.

I have more of a general question:
Your current implementation solves how an artifact was saved, but how do we expose the way it was loaded? E.g. even if an artifact is a regular step output, I can load it either the "normal" way by defining it as a step input, or manually by calling load_artifact(...).

I have more of a general question: Your current implementation solves how an artifact was saved, but how do we expose the way it was loaded? E.g. even if an artifact is a regular step output, I can load it either the "normal" way by defining it as a step input, or manually by calling load_artifact(...).

True, I didn't touch inputs here at all, since it was not in the scope of the ticket. I would prefer to handle it separately.
Do you refer to the DAG and how we define which input type is it? It would be hell complex design/FE wise, honestly speaking.

I can do this crazy thing and I have no clue how it would be shown:

@step
def step_1()->Annotated[int,"my_int"]:
    return 42

@pipeline
def pipe_1():
    step_1()
    step_2(Client().get_artifact_version("my_int"), after=["step_1"])

So it is a Schrödinger-Artifact now: it is DEFAULT and LAZY_LOADED at the same time...

Classification template updates in examples/mlops_starter have been pushed.

E2E template updates in examples/e2e have been pushed.

I see, makes sense let's handle that separately.
I think the problem is mostly on how to visualize what type of input it is right? Could just be a different colored line depending on the type type of input, but I'm sure Zuri will come up with something much better 😄
In general I think it's both, it's a regular output for the first step, but for the second step it's a coincidence that it was generated in the same run, and the general intention of the user was to load an artifact using the lazy loading, not using the regular way to pass artifacts, otherwise they would have done that ;)

That's also why the input type can't be stored in the artifact table, one artifact might be different types of inputs for different steps/runs or even the same exact step

That's also why the input type can't be stored in the artifact table, one artifact might be different types of inputs for different steps/runs or even the same exact step

Yep, I also realized that during development - this will stay separately as is it is now, but we would need to enrich it for more types beyond default and manual, IMO.

Thank you for using CodeRabbit. We offer it for free to the OSS community and would appreciate your support in helping us grow. If you find it useful, would you consider giving us a shout-out on your favorite social media?

There is still some unrelated error on mac 3.9, but I will merge this in as is.

Using Python 3.9.20 environment at /Users/runner/hostedtoolcache/Python/3.9.20/x64
  × No solution found when resolving dependencies:
  ╰─▶ Because torch==2.4.0 has no wheels with a matching Python implementation
      tag and vllm>=0.6.0 depends on torch==2.4.0, we can conclude that
      vllm>=0.6.0 cannot be used.
      And because only the following versions of vllm are available:
          vllm<=0.6.0
          vllm==0.6.1
          vllm==0.6.1.post1
          vllm==0.6.1.post2
          vllm==0.6.2
          vllm==0.6.3
          vllm==0.6.3.post1
      and you require vllm>=0.6.0, we can conclude that your requirements
      are unsatisfiable.