MLServer
MLServer copied to clipboard
tritonclient dependency change - cuda-python cause poetry installation failure in MacOS
Hi there
i got issue with poetry install mlserver, and it is due to a tritonclient 2.37+ now depends on cuda-python, it will block mlserver installation if the machine does not have NVIDIA GPU, wondering if it's possible to constrain the tritonclient version bit <2.37?
Hey @liusha-H,
Sure thing - sounds reasonable. Would you be willing to contribute a PR with the change?
@adriangonz I am having the same issue; here is the tl;dr
I am running a ZenML pipeline with the following ZenML integration dependencies
mlserver = ">=0.5.3"
mlserver-mlflow = ">=0.5.3"
I get the following error running via python v3.9 on Mac M1 via poetry 👇
• Installing cuda-python (12.2.0): Failed
RuntimeError
Unable to find installation candidates for cuda-python (12.2.0)
at ~/.local/pipx/venvs/poetry/lib/python3.9/site-packages/poetry/installation/chooser.py:76 in choose_for
72│
73│ links.append(link)
74│
75│ if not links:
→ 76│ raise RuntimeError(f"Unable to find installation candidates for {package}")
77│
78│ # Get the best link
79│ chosen = max(links, key=lambda link: self._sort_key(package, link))
80│
Following @liusha-H 's suggestion fixed it 👇
mlserver = ">=0.5.3"
mlserver-mlflow = ">=0.5.3"
tritonclient = "^2.24,<2.27"
CC @RoyerRamirez
@adriangonz , I can't help with this one. One doubt from my end. Why does MLServer need tritonclient in the first place?
Hey @privatedumbo, I'm no longer involved with this project - best would be to check with @ramonpzg on this one.
Hi @privatedumbo -- Did you mean you can help with this one?
I believe the reason to have it there is improve communication with the triton server running alongside MLServer inside Seldon Core.
@ramonpzg , I could, yes. For what I see, it's only used for batch-processing, so maybe it could be an extra dependency?