Shangdi Yu comments

Results 32 comments of


                                            Shangdi Yu

`"bucket"` link missing for draft depositions

encountering the same issue

[export] Failed to trace HF Llama2 model

With the changes in #131914, the code below can be exported (we convert autocast to HOO as discussed above). `with torch.device("meta")` still doesn't work (see #130067). ```python import torch from...

check if the input fx graph of aot_compile return tuple

> @angelayi Hi, > > I have implemented the PR, and I am wondering if I need to add a unit test for it. If so, could you please advise...

check if the input fx graph of aot_compile return tuple

lgtm. Let's wait for the CI checks to finish.

check if the input fx graph of aot_compile return tuple

@pytorchbot merge

check if the input fx graph of aot_compile return tuple

@pytorchbot merge

check if the input fx graph of aot_compile return tuple

@pytorchbot merge -i

[Torch.Export] [Regression] Index Put Outputs Fake Tensor in 2.4 (Was Good in 2.3)

Got normal output like below in main: ``` tensor([[0.2307, 0.5860, 1.0000, 0.8029], [0.4501, 0.3431, 0.3892, 0.0574], [0.9867, 5.0000, 0.0976, 0.8971]]) ``` Feel free to re-open if issue persists.

Aoti minifier flatten 2

@pytorchbot merge -i

Remove capture_pre_autograd_graph references in quantization

There are some test failures because a call to `capture_pre_autograd_graph` is recently added in https://github.com/pytorch/pytorch/pull/139054. I'd have to wait until it's imported internally to overwrite it.