Enable fastsafetensors to support the Paddle

[zeroRains]

Environment：

• Device: 2 * Tesla V100
• Cuda: 11.8
• paddlepaddle-gpu == 3.0.0
• pytorch == 2.5.1

I support fastsaftensors run with paddle.

Write 2 example cases in examples/padle_case (single and parallel).

Add all unitest for paddle.

[takeshi-yoshimura]

@zeroRains
Thank you very much for contributing your code! I am very happy to see code for supporting other frameworks than pytorch.

I would like to merge this immediately after we confirm no regressions on both pytorch and paddle, and then, we can clean up the change as separate patches later as v0.1.14.

I would like to ask you two things:

1. Could you please share make unittest result, directory htmlcov?
2. According to the new code-of-conduct, all of your commits needs to contain the signed-off-by line. See https://github.com/foundation-model-stack/fastsafetensors/pull/16/checks?check_run_id=42825628638 for more details. Could you please edit your commits?

[zeroRains]

Thank you for your recognition.

The current version can read tensors successfully by executing each unittest command . However, it gets stuck at the end of a unittest command (possibly related to my failure to install nvidia-fs successfully. If it's convenient for you, please try to see if there is the same problem). I still need to check again then I will share the htmlcov.

When all the modifications are completed, all commits will be rebase into one commit.

[takeshi-yoshimura]

@zeroRains
can you please rebase your change with the current main branch? it should trigger tests on actions and show htmlcov if tests successfully finish.
lint should fail, but please ignore. I will fix them later.

[zeroRains]

@takeshi-yoshimura

I have rebased all commit to one commit.

[zeroRains]

I forget to add paddlepaddle-gpu as dependency to pyproject.toml. Maybe we should retrigger the test.

[takeshi-yoshimura]

thanks. maybe there are issues around package installations. Please note that we do not have GPUs on test machines.
Let me add a fixing PR onto your branch soon later.

[takeshi-yoshimura]

@zeroRains
Could you please check my PR zeroRains#2 on your branch?
If you do not have time to work on that, please just modify and re-push your commit with the Signed-off-by line with your name and email so that I can merge this PR. then I can update your change as separate commits.

[zeroRains]

@zeroRains Could you please check my PR zeroRains#2 on your branch? If you do not have time to work on that, please just modify and re-push your commit with the Signed-off-by line with your name and email so that I can merge this PR. then I can update your change as separate commits.

@takeshi-yoshimura Sorry, I misunderstood what you meant before, I will check the pr soon then merge it.

[zeroRains]

@takeshi-yoshimura Hello, I have merged the pr and rebase all commit to one.

[takeshi-yoshimura]

@zeroRains
thanks! please run
git commit --amend -s
to add your signed-off-by line and re-push the commit. This is required to show your agreement on DCO.

[zeroRains]

@takeshi-yoshimura I have run it.

[takeshi-yoshimura]

finally I have merged your commit to main. I will keep updating code, but anyway, thank you very much for your contribution!

[zeroRains]

@takeshi-yoshimura Hello, Are there any plans to release the fastsafetensors version supporting paddle on PyPI in the near future? If so, could you please inform us of the time?

[takeshi-yoshimura]

@zeroRains
I am working on it. so, hopefully it will be there this week or next week.

[zeroRains]

@takeshi-yoshimura Sorry, I find that distributed loading has not been fully verified yet. I need to test it. Later, I will resubmit a PR regarding distributed fixes.