[functorch] fix batching rule for dropout

[kshitij12345]

Fixes #92283

The repro now works:

import torch
import torch.func
import torch.nn as nn

x = torch.randn(3, device='cuda')
y = torch.randn(1, 3, device='cuda')

def fn(x, y):
    # previously output of dropout used to be incorrect [B, 3] (B=1) and thus `mean(1)` used to fail
    # post the fix output of dropout is [B, 1, 3] and `mean(1)` works.
    return x + nn.functional.dropout(y, 0.3).mean(1)


o = torch.func.vmap(fn, in_dims=(0, None), randomness='different')(x, y)

NOTE:
native_dropout_batching_rule(const Tensor& tensor, double p, c10::optional<bool> train) was called only for CUDA tensor. Hence this issue only affected CUDA tensors and not CPU tensors

Ref:

pytorch/aten/src/ATen/functorch/PyTorchOperatorHacks.cpp

Lines 251 to 258 in a6ac922

	Tensor dropout(const Tensor& input, double p, bool train) {
	auto result = [&]() {
	NoNamesGuard guard;
	if (train && is_fused_kernel_acceptable(input, p)) {
	return std::get<0>(at::native_dropout(input, p, train));
	}
	return _dropout<false>(input, p, train);
	}();

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/92975

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 6acfa4c:
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[Skylion007]

Looks good to me now.

[kshitij12345]

@pytorchbot merge

[pytorchmergebot]

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here