taichi icon indicating copy to clipboard operation
taichi copied to clipboard

PYPI project is not updated

Open SohamTilekar opened this issue 1 year ago • 8 comments

How do I install Latest taichi using pypi insted of using pip install git+... taichi project is not updated on pypi. It is Hard to install it using git+ way for any module.

SohamTilekar avatar Feb 20 '24 11:02 SohamTilekar

Taichi 1.7 is the latest release on both.

https://github.com/taichi-dev/taichi/releases https://pypi.org/project/taichi/#history

pip install git+https://github... builds a development version from source. However other code hosting services are available too. Even non-git ones.

Pypi doesn't check every possible location a project could host its source code.

JamesParrott avatar Feb 20 '24 20:02 JamesParrott

But It Does not Support pyhton version 3.12. but git one dose.

pip install taichi

> ERROR: Could not find a version that satisfies the requirement taichi (from versions: none)
> ERROR: No matching distribution found for taichi

SohamTilekar avatar Feb 21 '24 05:02 SohamTilekar

But It Does not Support pyhton version 3.12. but git one dose.

pip install taichi

> ERROR: Could not find a version that satisfies the requirement taichi (from versions: none)
> ERROR: No matching distribution found for taichi

We won't support 3.12 until PyTorch reached it... Otherwise our CI tests simply won't pass

bobcao3 avatar Feb 21 '24 10:02 bobcao3

git+... version supports the python3.12 but the pypi hosted version do not.

SohamTilekar avatar Feb 21 '24 10:02 SohamTilekar

Presumably the git+ version is not a supported release, but the in-development version, that can't pass the CI tests until PyTorch is updated.

JamesParrott avatar Feb 21 '24 11:02 JamesParrott

@JamesParrott pytorch already supports python 3.12, so I guess the next version of taichi would support it as well?

Any idea when it is going to be available?

feature-engineer avatar Jun 19 '24 11:06 feature-engineer

I haven't the faintest idea. But to make it happen, you could run the Python 3.12 tests the devs mentioned, that were waiting on PyTorch. https://github.com/taichi-dev/taichi/issues/8365

It looks like they're very nearly there, there're just a couple of little issues to sort out on Apple M1 (int 32 overflows).

 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
tests/python/test_reduction.py:85: in _test_reduction_single
    assert criterion(tot[None], ground_truth)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

x = 1, y = np.int32(1)

>   _test_reduction_single(ti.i32, lambda x, y: x % 2**32 == y % 2**32, op)
E   OverflowError: Python integer 4294967296 out of bounds for int32

https://github.com/taichi-dev/taichi/actions/runs/9571475701/job/26388688466

JamesParrott avatar Jun 19 '24 16:06 JamesParrott

We are fixing some of the last bugs with Python 3.12 and numpy 2.0, I think for most of the x86 platforms we are building 3.12 wheels for nightly, you can install the nightly for now. The mac os arm64 nightly should be unblocked by #8522

bobcao3 avatar Jun 23 '24 22:06 bobcao3