search

rapidsai / docs

RAPIDS Documentation Site
https://docs.rapids.ai
37 stars 52 forks source link

24.04 Selector updates #506

Closed jarmak-nv closed 7 months ago

jarmak-nv commented 8 months ago

This PR does a few things:

  1. Removes the nightly restriction for CUDA 12.2 and Python 3.11
  2. Makes CUDA 12.2 and Python 3.11 the default versions
  3. Fixes a long-standing bug where when pip is selected, a CUDA version isn't highlighted
  4. ~[PREFERENCE CHANGE] Changes versioned-items (CUDA, Python) to be descend sorted so left-most items are the default.~
    • Not married to this, but I think it's visually simpler/more intuitive
    • Removed!

Adding https://github.com/rapidsai/docs/labels/DO%20NOT%20MERGE until 24.04 is released.

netlify[bot] commented 8 months ago

Deploy Preview for docs-rapids-ai ready!

Built without sensitive environment variables

Name Link
Latest commit 35528dd903c10ffd7003a0ed93328399dc121c81
Latest deploy log https://app.netlify.com/sites/docs-rapids-ai/deploys/661982aca441440008e1ff4b
Deploy Preview https://deploy-preview-506--docs-rapids-ai.netlify.app
Preview on mobile
Toggle QR Code...

QR Code

Use your smartphone camera to open QR code link.

To edit notification comments on pull requests, go to your Netlify site configuration.

jarmak-nv commented 7 months ago

I've added cuVS to the selector as an option but this made me realize a few things I'm uncertain on. Perhaps you know this @bdice or @raydouglass or @cjnolet

  1. Is cuVS part of the rapids metapackage on conda?
  2. Right now for pip we don't have a metapackage so we just install every RAPIDS library - should we include cuVS? This is really just a followup for part 1 above
  3. Do we know if cuVS depends on RAFT/RAFT-dask? If yes I need to go in and update the logic to pin those in the nightly pip since we need to constrain the version of every RAPIDS lib there
bdice commented 7 months ago

I attempted to answer the cuVS questions below. I think we can safely defer on these and make no immediate changes. cuVS is under heavy development. There are lots of dependency relationships being established as part of the movement of compiled code from RAFT to cuVS in the 24.06 cycle. For 24.06, cuVS should be treated like cuDF/cuML/etc, but I don't think we need to change anything in the install guide for 24.04.

  1. Is cuVS part of the rapids metapackage on conda?

No, it's not in 24.04. We should change this for 24.06 but it's too late for 24.04. https://github.com/rapidsai/integration/blob/branch-24.04/conda/recipes/rapids/meta.yaml

  1. Right now for pip we don't have a metapackage so we just install every RAPIDS library - should we include cuVS? This is really just a followup for part 1 above

I'm leaning towards no. Let's match the contents of the rapids metapackage.

  1. Do we know if cuVS depends on RAFT/RAFT-dask? If yes I need to go in and update the logic to pin those in the nightly pip since we need to constrain the version of every RAPIDS lib there

Yes, there will be a dependency between these.

cjnolet commented 7 months ago

Thanks @jarmak-nv, sorry for being late to this discussion.

Is cuVS part of the rapids metapackage on conda?

If it's not then it should be, just like all other RAPIDS top-level libraries (cuML, cuGraph, cuDF, etc..)

Right now for pip we don't have a metapackage so we just install every RAPIDS library - should we include cuVS? This is really just a followup for part 1 above

Again, whatever we do for the other top-level RAPIDS libraries, we should do for cuVS also.

Do we know if cuVS depends on RAFT/RAFT-dask? If yes I need to go in and update the logic to pin those in the nightly pip since we need to constrain the version of every RAPIDS lib there

cuVS will depend on RAFT but not raft-dask. We can pin that for now. libraft will be going away, hopefully shortly before the 24.06 release and raft will revert to a header-only c++ library. cuvs python package will still have a dependency on pylibraft, though and it's likely the raft header-only library will remain a conda package for the foreseeable future.

As @bdice points out, we don't need to rush this into 24.04 but I would like it if we could update these for 24.06.

bdice commented 7 months ago

@cjnolet I started the process for getting cuvs into the rapids 24.06 metapackage here: https://github.com/rapidsai/integration/pull/708

cjnolet commented 7 months ago

Thanks @bdice!