[Bugfix] Fix quant RMS norm fusion for quantization with TMA-aligned scales (#33255)

Signed-off-by: ElizaWszola <ewszola@redhat.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>

csrc/torch_bindings.cpp

@@ -643,11 +643,13 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
 
 #ifndef USE_ROCM
   // Compute per-token-group FP8 quantized tensor and scaling factor.
+  // The dummy arguments are here so we can correctly fuse with RMSNorm.
   ops.def(
       "per_token_group_fp8_quant(Tensor input, Tensor! output_q, Tensor! "
       "output_s, "
       "int group_size, float eps, float fp8_min, float fp8_max, bool "
-      "scale_ue8m0) -> ()");
+      "scale_ue8m0, bool dummy_is_scale_transposed, bool dummy_is_tma_aligned "
+      ") -> ()");
   ops.impl("per_token_group_fp8_quant", torch::kCUDA,
            &per_token_group_quant_fp8);