Update CuPy and NumPy dependency specs by bdice · Pull Request #3058 · NVIDIA/raft

bdice · 2026-06-24T05:08:41Z

Contributes to rapidsai/build-planning#279

RAPIDS is taking on requirements cupy>=14.0.1,!=14.1.0 and numpy>=2.0. Wheel dependencies on cupy-cuda12x and cupy-cuda13x now use the [ctk] extra.

See the linked issue for details.

copy-pr-bot · 2026-06-24T05:08:44Z

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

coderabbitai · 2026-06-24T05:59:03Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 8f1fe369-cc35-44c6-97be-96232b1d9e9e

📥 Commits

Reviewing files that changed from the base of the PR and between 0337940 and beb60f2.

📒 Files selected for processing (1)

dependencies.yaml

🚧 Files skipped from review as they are similar to previous changes (1)

dependencies.yaml

📝 Walkthrough

Summary by CodeRabbit

Chores
- Updated CUDA-related package requirements across supported environments and packaging metadata.
- Raised the minimum supported NumPy version to 2.0 (kept upper bound <3.0).
- Updated CuPy compatibility to require newer 14.0.1+ builds while continuing to exclude 14.1.0 (and relaxing the previous exclusion of 14.0.0 where applicable).
- Aligned these constraints across x86_64 and aarch64, including recipe and test extra requirements (with the CUDA selector variant).

Walkthrough

Raises the minimum CuPy version from >=13.6.0 to >=14.0.1 and the minimum NumPy version from >=1.23 to >=2.0, with matching constraint updates in shared dependency files, conda recipes, and generated conda environment files. The pyproject test extra also switches to cupy-cuda13x[ctk].

Changes

CuPy and NumPy Version Bump

Layer / File(s)	Summary
Canonical dependency definitions `dependencies.yaml`, `python/pylibraft/pyproject.toml`	`depends_on_cupy` updates CuPy selector pins to `>=14.0.1` with `!=14.1.0`; `run_pylibraft` raises NumPy to `>=2.0,<3.0`; `pyproject.toml` raises NumPy to `>=2.0,<3.0` and changes the test extra CuPy spec to `cupy-cuda13x[ctk]>=14.0.1,!=14.1.0`.
Conda recipe constraint updates `conda/recipes/pylibraft/recipe.yaml`, `conda/recipes/raft-dask/recipe.yaml`	`pylibraft` raises the run NumPy requirement to `>=2.0,<3.0` and updates cupy run_constraints to `>=14.0.1,!=14.1.0`; `raft-dask` updates its cupy run_constraints pin to `>=14.0.1,!=14.1.0`.
Conda environment file propagation `conda/environments/all_cuda-129_arch-.yaml`, `conda/environments/all_cuda-133_arch-.yaml`	All four conda environment files update cupy to `>=14.0.1,!=14.1.0` and numpy to `>=2.0,<3.0`.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: updating CuPy and NumPy dependency specifications.
Description check	✅ Passed	The description matches the changeset by describing the CuPy, NumPy, and [ctk] dependency updates.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests

_{Comment @coderabbitai help to get the list of available commands.}

gforsyth · 2026-06-24T18:18:22Z

+              - cupy-cuda12x[ctk]>=14.0.1,!=14.1.0
          # fallback to CUDA 13 versions if 'cuda' is '13.*' or not provided
          - matrix:
            packages:
-              - cupy-cuda13x>=13.6.0,!=14.0.0,!=14.1.0
+              - cupy-cuda13x[ctk]>=14.0.1,!=14.1.0
  depends_on_rapids_logger:


The depends_on_cupy block is also used to define test_python -- and the requirements generated by dependency-file-generator get piped into $PIP_CONSTRAINTS in test_wheel_pylibraft.sh (and others).

But you can't use extras in the $PIP_CONSTRAINTS -- maybe we break it out into depends_on_cupy and depends_on_cupy_test and then explicitly list the ctk packages there?

I'm happy with the resolution pushed in recent commits, where the [ctk] extra is only present in output_type: pyproject lists.

jameslamb · 2026-06-25T14:17:35Z

              cuda: "12.*"
            packages:
-              - cupy-cuda12x>=13.6.0,!=14.0.0,!=14.1.0
+              - cupy-cuda12x[ctk]>=14.0.1,!=14.1.0


Sorry to complicate this, but I think this needs the use_cuda_wheels: true qualifier, like this:

raft/dependencies.yaml

Lines 271 to 283 in c1babf1

# if use_cuda_wheels=false is provided, do not add dependencies on any CUDA wheels

# (e.g. for DLFW and pip devcontainers)

- matrix:

use_cuda_wheels: "false"

packages:

- matrix:

arch: aarch64

cuda: "12.2"

use_cuda_wheels: "true"

packages:

# some components (like nvidia-cublas-cu12 and nvidia-cuda-nvcc-cu12) didn't have

# aarch64 wheels until CTK 12.3, so allow a slightly looser bound here

- cuda-toolkit>=12.2,<12.4

Otherwise, this will pull in cuda-toolkit[...]:

https://github.com/cupy/cupy/blob/9d193b65ea0a393b579cb50d5db8dbf23a667ab1/setup.py#L85-L87

And we'll end up having to patch it in builds like DLFW where we prefer a system-installed CTK.

This applies to all of the PRs.

Thanks for the updates across all the repos to address this, I'm happy with the new state of these PRs: rapidsai/build-planning#279 (comment)

jameslamb

Approved with one tiny suggestion

jameslamb · 2026-06-26T15:35:28Z

+          - matrix:
+              cuda: "12.*"
+            packages:
+              - cupy-cuda12x>=14.0.1,!=14.1.0
+          # fallback to CUDA 13 versions if 'cuda' is '13.*' or not provided
+          - matrix:
+            packages:
+              - cupy-cuda13x>=14.0.1,!=14.1.0
+      - output_types: pyproject
        matrices:
+          - matrix:
+              cuda: "12.*"
+              use_cuda_wheels: "false"
+            packages:
+              - cupy-cuda12x>=14.0.1,!=14.1.0
+          - matrix:
+              cuda: "13.*"
+              use_cuda_wheels: "false"
+            packages:
+              - cupy-cuda13x>=14.0.1,!=14.1.0


Suggested change

- matrix:

cuda: "12.*"

packages:

- cupy-cuda12x>=14.0.1,!=14.1.0

# fallback to CUDA 13 versions if 'cuda' is '13.*' or not provided

- matrix:

packages:

- cupy-cuda13x>=14.0.1,!=14.1.0

- output_types: pyproject

matrices:

- matrix:

cuda: "12.*"

use_cuda_wheels: "false"

packages:

- cupy-cuda12x>=14.0.1,!=14.1.0

- matrix:

cuda: "13.*"

use_cuda_wheels: "false"

packages:

- cupy-cuda13x>=14.0.1,!=14.1.0

- matrix:

cuda: "12.*"

packages:

- &cupy_cu12 cupy-cuda12x>=14.0.1,!=14.1.0

# fallback to CUDA 13 versions if 'cuda' is '13.*' or not provided

- matrix:

packages:

- &cupy_cu13 cupy-cuda13x>=14.0.1,!=14.1.0

- output_types: pyproject

matrices:

- matrix:

cuda: "12.*"

use_cuda_wheels: "false"

packages:

- *cupy_cu12

- matrix:

cuda: "13.*"

use_cuda_wheels: "false"

packages:

- *cupy_cu13

Minor comment... realizing that in many of these PRs, the spelling of this dependency is indentical in both of these cases:

requirements

pyproject.toml + use_cuda_wheels: "false"

Using YAML anchors could help keep them consistent (I think we always want them to be?) and reduce duplication.

jameslamb · 2026-06-26T15:35:56Z

              cuda: "12.*"
            packages:
-              - cupy-cuda12x>=13.6.0,!=14.0.0,!=14.1.0
+              - cupy-cuda12x[ctk]>=14.0.1,!=14.1.0


Thanks for the updates across all the repos to address this, I'm happy with the new state of these PRs: rapidsai/build-planning#279 (comment)

bdice added non-breaking Non-breaking change improvement Improvement / enhancement to an existing function labels Jun 24, 2026

github-project-automation Bot added this to Unstructured Data Processing Jun 24, 2026

bdice mentioned this pull request Jun 24, 2026

feat(deps): raise cupy lower bound to >=14 rapidsai/build-planning#279

Open

bdice force-pushed the cupy-pin branch from ad1a710 to 7fc8b01 Compare June 24, 2026 05:27

bdice changed the title ~~Update CuPy lower bound to 14.0.1~~ Update CuPy and NumPy lower bounds Jun 24, 2026

Update CuPy and NumPy dependency specs.

b50fd22

bdice force-pushed the cupy-pin branch from 7fc8b01 to b50fd22 Compare June 24, 2026 05:44

bdice changed the title ~~Update CuPy and NumPy lower bounds~~ Update CuPy and NumPy dependency specs Jun 24, 2026

bdice marked this pull request as ready for review June 24, 2026 05:54

bdice requested a review from a team as a code owner June 24, 2026 05:54

bdice requested a review from KyleFromNVIDIA June 24, 2026 05:54

gforsyth reviewed Jun 24, 2026

View reviewed changes

bdice added 2 commits June 25, 2026 00:01

Fix CuPy constraints for wheel tests

0337940

Merge branch 'main' into cupy-pin

4900437

jameslamb reviewed Jun 25, 2026

View reviewed changes

KyleFromNVIDIA approved these changes Jun 25, 2026

View reviewed changes

Guard CuPy CTK extra for wheel metadata

beb60f2

jameslamb approved these changes Jun 26, 2026

View reviewed changes

jameslamb mentioned this pull request Jun 26, 2026

feature request: enforce RAPIDS dependencies.yaml conventions rapidsai/pre-commit-hooks#132

Open

Reuse CuPy dependency anchors

d6a63e1

bdice requested a review from a team as a code owner June 26, 2026 22:35

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Update CuPy and NumPy dependency specs#3058

Update CuPy and NumPy dependency specs#3058
bdice wants to merge 5 commits into
NVIDIA:mainfrom
bdice:cupy-pin

bdice commented Jun 24, 2026 •

edited

Loading

Uh oh!

copy-pr-bot Bot commented Jun 24, 2026

Uh oh!

coderabbitai Bot commented Jun 24, 2026 •

edited

Loading

Summary by CodeRabbit

Walkthrough

Changes

Estimated code review effort

Uh oh!

gforsyth Jun 24, 2026

Uh oh!

jameslamb Jun 26, 2026 •

edited

Loading

Uh oh!

jameslamb Jun 25, 2026

Uh oh!

jameslamb Jun 26, 2026

Uh oh!

jameslamb left a comment

Uh oh!

jameslamb Jun 26, 2026

Uh oh!

jameslamb Jun 26, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

	# if use_cuda_wheels=false is provided, do not add dependencies on any CUDA wheels
	# (e.g. for DLFW and pip devcontainers)
	- matrix:
	use_cuda_wheels: "false"
	packages:
	- matrix:
	arch: aarch64
	cuda: "12.2"
	use_cuda_wheels: "true"
	packages:
	# some components (like nvidia-cublas-cu12 and nvidia-cuda-nvcc-cu12) didn't have
	# aarch64 wheels until CTK 12.3, so allow a slightly looser bound here
	- cuda-toolkit>=12.2,<12.4

Uh oh!

Conversation

bdice commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

copy-pr-bot Bot commented Jun 24, 2026

Uh oh!

coderabbitai Bot commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Walkthrough

Changes

Estimated code review effort

Uh oh!

gforsyth Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

jameslamb Jun 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jameslamb Jun 25, 2026

Choose a reason for hiding this comment

Uh oh!

jameslamb Jun 26, 2026

Choose a reason for hiding this comment

Uh oh!

jameslamb left a comment

Choose a reason for hiding this comment

Uh oh!

jameslamb Jun 26, 2026

Choose a reason for hiding this comment

Uh oh!

jameslamb Jun 26, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

bdice commented Jun 24, 2026 •

edited

Loading

coderabbitai Bot commented Jun 24, 2026 •

edited

Loading

jameslamb Jun 26, 2026 •

edited

Loading