PyTorch 2.4

[alecandido]

It seems that this release is breaking some Qibo tests, at least on Windows.

The following commit may be relevant: qiboteam/qibo@eb19847
This is what happened to the lock file when I added an upper bound in PyTorch version. And it's fixing the issue

The most significant differences are:

• nvidia-cudnn-cu12 major bump (8 -> 9)
• Intel APIs removal (tbb, mkl, intel-openmp)
• triton major bump (2 -> 3)