warp env pulls PyTorch 2.0.1 (cu118) (built against NumPy 1.x); NumPy 2.x (current default on conda-forge)

[Chris]

Hi All,

I just ran a fresh warp conda env install via:

conda create -n warp warp -c warpem -c nvidia/label/cuda-11.8.0 -c pytorch -c conda-forge

I noticed that the conda warp env pulls PyTorch 2.0.1 (cu118) (which was built against NumPy 1.x); and the solver also installed NumPy 2.x (the current default on conda-forge).

And I was just wondering if there will be any issues running warp 2.0.0dev36 from this environment, or if it truly requires a NumPy downgrade?

Thanks all!

Best,

Chris

Example:

(warp) [chris ~]$ python -c "import numpy, torch; print('NumPy', numpy.__version__, 'Torch', torch.__version__)"
NumPy 2.3.4 Torch 2.0.1

(warp) [chris ~]$ python - <<'PY'
import torch
print("Torch:", torch.__version__)
print("CUDA available:", torch.cuda.is_available())
print("CUDA runtime reported by torch:", torch.version.cuda)
PY
Torch: 2.0.1
CUDA available: True
CUDA runtime reported by torch: 11.8

I ask because another check threw a warning: "A module that was compiled using NumPy 1.x cannot be run in NumPy 2.3.4 as it may crash. To support both 1.x and 2.x

versions of NumPy, modules must be compiled with NumPy 2.0.
Some module may need to rebuild instead e.g. with 'pybind11>=2.12'.

If you are a user of the module, the easiest solution will be to
downgrade to 'numpy<2' or try to upgrade the affected module.
We expect that some modules will need time to support NumPy 2.

From:

(warp) [chris ~]$ python - <<'PY'

import torch
x = torch.rand(3,3, device="cuda")
print("Device:", x.device)
print("GPU:", torch.cuda.get_device_name(0))
print("OK, did a CUDA op:", (x @ x).sum().item())
PY

A module that was compiled using NumPy 1.x cannot be run in
NumPy 2.3.4 as it may crash. To support both 1.x and 2.x
versions of NumPy, modules must be compiled with NumPy 2.0.
Some module may need to rebuild instead e.g. with 'pybind11>=2.12'.

If you are a user of the module, the easiest solution will be to
downgrade to 'numpy<2' or try to upgrade the affected module.
We expect that some modules will need time to support NumPy 2.

Traceback (most recent call last):  File "<stdin>", line 2, in <module>
<stdin>:2: UserWarning: Failed to initialize NumPy: _ARRAY_API not found (Triggered internally at /opt/conda/conda-bld/pytorch_1682343995622/work/torch/csrc/utils/

[Alister Burt]

hey Chris,

Thanks for the note, I think we only link against the C++ library behind torch so don’t expect this discrepancy would cause issues… do let us know if you spot any issues though!

Cheers,

Alister

Sent from mobile - apologies for brevity

On Nov 7, 2025, at 09:27, Chris <ccaff...@gmail.com> wrote:

Hi All,

--
You received this message because you are subscribed to the Google Groups "Warp" group.
To unsubscribe from this group and stop receiving emails from it, send an email to warp-em+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/warp-em/32b037d5-04c6-44a3-924c-a8771dadbdfcn%40googlegroups.com.