Remove confusing warning message from SDPA about mask alignment (pyto…

…rch#114909)

# Summary
Users have reported that this warning message leads to confusion about the correctness of the mask even though it is only concerned with performance.

Pull Request resolved: pytorch#114909
Approved by: https://github.com/Chillee

aten/src/ATen/native/transformers/attention.cpp

@@ -548,13 +548,6 @@ at::Tensor preprocess_mask(
   constexpr int mem_eff_alignment = 8;
   at::Tensor result_mask = mask;
   if (!aligned_tensor<mem_eff_alignment>(mask)) {
-    TORCH_WARN_ONCE(
-        "Memory Efficient Attention requires the attn_mask to be aligned to, ",
-        mem_eff_alignment,
-        " elements. "
-        "Prior to calling SDPA, pad the last dimension of the attn_mask "
-        "to be at least a multiple of ", mem_eff_alignment,
-        " and then slice the attn_mask to the original size.");
     result_mask = pad_bias<mem_eff_alignment>(mask);
   }
   return result_mask.expand_symint(