Tweak rmsnorm operator regarding kernel name (flagos-ai#2056)

Author: tengqm

src/flag_gems/experimental_ops/rmsnorm.py

@@ -10,7 +10,7 @@
 
 
 @triton.jit
-def rmsnorm(
+def rmsnorm_kernel(
     input_ptr,  # *Pointer* to the input tensor flattened to 2D [M, N]
     weight_ptr,  # *Pointer* to the weight tensor [N]
     output_ptr,  # *Pointer* to the output tensor flattened to 2D [M, N]
@@ -48,10 +48,6 @@ def rmsnorm(
         col_start += BLOCK_SIZE
 
 
-# Keep a handle to the Triton kernel before defining the Python wrapper with the same name
-rmsnorm_kernel = rmsnorm
-
-
 def rmsnorm(input_tensor: torch.Tensor, weight: torch.Tensor, eps: float = 1e-6):
     assert input_tensor.is_cuda and weight.is_cuda, "Tensors must be on CUDA device."
     assert (