search

conda-forge / coremltools-feedstock

A conda-smithy repository for coremltools.
BSD 3-Clause "New" or "Revised" License
3 stars 7 forks source link

Still maintained? #24

Open mittagessen opened 1 year ago

mittagessen commented 1 year ago

Comment:

Hi,

The current version of coremltools built by this feedstock is quite old and there are a couple of auto-generated pull requests from some conda-forge bot for newer versions without any interaction. So I'd like to inquire if this feedstock is still maintained or not.

TobyRoseman commented 1 year ago

It is not currently maintained. Is there a reason you prefer to use conda rather than pip?

mittagessen commented 1 year ago

I'm packaging another python software that depends on coremltools and it is easier for many users to install that one through conda. An older version isn't much of a problem right now but at some point I'd like to add native torchscript conversion support.

hmaarrfk commented 3 months ago

For a long time conda(-forge) was the preferred way to to install python + ML tools for the M1 ;)

@mittagessen if it is in your capabilities, you can certainly help maintain it.

Often new maintainers jump in over time to help out. simply make a PR and ping @conda-forge/core if you don't hear from the original maintainers.

TobyRoseman commented 3 months ago

For a long time conda(-forge) was the preferred way to to install python + ML tools for the M1 ;)

This is no longer true. I'm not sure if it was ever true, but it certainly hasn't been true for a long time. Please just use pip.

hmaarrfk commented 3 months ago

but it certainly hasn't been true for a long time

this may be the case, but I remember one day when https://developer.apple.com/metal/tensorflow-plugin/ pointed to conda, and I somewhat remember a website (maybe tensorflow's or apple's) even making a reference to conda-forge.

Please just use pip.

I think it is reasonable (and expected!) that when users report issues upstream (presumably you are involved in this project at Apple Toby) that they use the suggested methods. It does seem indeed that the website above has changed to suggest a pip based workflow.

Back to the original question:

The current version of coremltools built by this feedstock is quite old

As a community driven project, the version of the package we release is dependent on volunteers.

ML packages are particularly challenging. I notice that Toby had made an attempt a few years back, but it seems that the original maintainers did not have time to review it: https://github.com/conda-forge/coremltools-feedstock/pull/18

Other attempts include:

Your help mittagessen would be appreciated where you can.