pytorch

8876c0b2 - [transformer benchmark] fix in sdp_bwd for scaled_dot_product_attention return type (#101341)

[transformer benchmark] fix in sdp_bwd for scaled_dot_product_attention return type (#101341) Summary: Otherwise we get ``` Traceback (most recent call last): File "<string>", line 49, in <module> File "<string>", line 47, in __run File "/usr/local/fbcode/platform010/lib/python3.8/runpy.py", line 194, in _run_module_as_main return _run_code(code, main_globals, None, File "/usr/local/fbcode/platform010/lib/python3.8/runpy.py", line 87, in _run_code exec(code, run_globals) File "/data/users/jongsoo/fbsource/buck-out/v2/gen/fbcode/ef4169ac7f95fb74/caffe2/benchmarks/transformer/__sdp_backwards__/sdp_backwards#link-tree/caffe2/benchmarks/transformer/sdp_backwards.py", line 188, in <module> main() File "/data/users/jongsoo/fbsource/buck-out/v2/gen/fbcode/ef4169ac7f95fb74/caffe2/benchmarks/transformer/__sdp_backwards__/sdp_backwards#link-tree/caffe2/benchmarks/transformer/sdp_backwards.py", line 184, in main run_timing(min_run_time, batch_size, embed_dim, num_heads, max_seq_len, dtype) File "/data/users/jongsoo/fbsource/buck-out/v2/gen/fbcode/ef4169ac7f95fb74/caffe2/benchmarks/transformer/__sdp_backwards__/sdp_backwards#link-tree/caffe2/benchmarks/transformer/sdp_backwards.py", line 105, in run_timing rand_fused_upward = cpt(x, x, x, mask).clone().detach() File "/data/users/jongsoo/fbsource/buck-out/v2/gen/fbcode/ef4169ac7f95fb74/caffe2/benchmarks/transformer/__sdp_backwards__/sdp_backwards#link-tree/torch/nn/modules/module.py", line 1502, in _wrapped_call_impl return self._call_impl(*args, **kwargs) File "/data/users/jongsoo/fbsource/buck-out/v2/gen/fbcode/ef4169ac7f95fb74/caffe2/benchmarks/transformer/__sdp_backwards__/sdp_backwards#link-tree/torch/nn/modules/module.py", line 1511, in _call_impl return forward_call(*args, **kwargs) File "/data/users/jongsoo/fbsource/buck-out/v2/gen/fbcode/ef4169ac7f95fb74/caffe2/benchmarks/transformer/__sdp_backwards__/sdp_backwards#link-tree/caffe2/benchmarks/transformer/sdp_backwards.py", line 39, in forward attn, _ = torch.nn.functional.scaled_dot_product_attention( ValueError: too many values to unpack (expected 2) ``` Test Plan: buck run mode/dev-nosan //caffe2/benchmarks/transformer:sdp_backwards Differential Revision: D45843838 Pull Request resolved: https://github.com/pytorch/pytorch/pull/101341 Approved by: https://github.com/drisspg