[BUG] FLOPS of `interpolate` op is WRONG

[xmfbit]

Describe the bug

The flops compute of F.interpolate is WRONG when using scale_factor, see https://github.com/microsoft/DeepSpeed/blob/master/deepspeed/profiling/flops_profiler/profiler.py#L725

it should be:

flops *= int(_prod(scale_factor))

To Reproduce

It is so obvious that I think code to reproduce is unnecessary.

Expected behavior
A clear and concise description of what you expected to happen.

ds_report output

This bug remains in the latest main branch of deepspeed. And I list my config below:

[2023-10-11 11:00:00,952] [INFO] [real_accelerator.py:158:get_accelerator] Setting ds_accelerator to cuda (auto detect)
--------------------------------------------------
DeepSpeed C++/CUDA extension op report
--------------------------------------------------
NOTE: Ops not installed will be just-in-time (JIT) compiled at
      runtime if needed. Op compatibility means that your system
      meet the required dependencies to JIT install the op.
--------------------------------------------------
JIT compiled ops requires ninja
ninja .................. [OKAY]
--------------------------------------------------
op name ................ installed .. compatible
--------------------------------------------------
 [WARNING]  async_io requires the dev libaio .so object and headers but these were not found.
 [WARNING]  async_io: please install the libaio-dev package with apt
 [WARNING]  If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.
async_io ............... [NO] ....... [NO]
fused_adam ............. [NO] ....... [OKAY]
cpu_adam ............... [NO] ....... [OKAY]
cpu_adagrad ............ [NO] ....... [OKAY]
fused_lamb ............. [NO] ....... [OKAY]
quantizer .............. [NO] ....... [OKAY]
random_ltd ............. [NO] ....... [OKAY]
 [WARNING]  sparse_attn requires a torch version >= 1.5 and < 2.0 but detected 2.0
 [WARNING]  using untested triton version (2.0.0), only 1.0.0 is known to be compatible
sparse_attn ............ [NO] ....... [NO]
spatial_inference ...... [NO] ....... [OKAY]
transformer ............ [NO] ....... [OKAY]
stochastic_transformer . [NO] ....... [OKAY]
transformer_inference .. [NO] ....... [OKAY]
--------------------------------------------------
DeepSpeed general environment info:
deepspeed info ................... 0.10.3, unknown, unknown
torch cuda version ............... 11.7
torch hip version ................ None
nvcc version ..................... 11.8
deepspeed wheel compiled w. ...... torch 2.0, cuda 11.7
shared memory (/dev/shm) size .... 1007.58 GB

[loadams]

@xmfbit - thanks for pointing this out, not sure how this wasn't caught for so long. Would you review the linked PR?

[xmfbit]

@loadams Hi, it's glad to see the bug has fixed! Thanks for your rapid response!

But unfortunately, I find that there is another bug in the implementation of upsample_flops_compute function. I have submit a new issue, please refer to #4504 for details. I have posted the bug info and how to fix it in the issue. Please check it! Thanks!