Merge branch 'eddiey-main-patch-57472' into 'main'

fix potential race in `cuComputeGradInput` in `layer_norm_cuda_kernel.cu`

See merge request ADLR/megatron-lm!340

megatron/fused_kernels/layer_norm_cuda_kernel.cu

@@ -645,6 +645,8 @@ void cuComputeGradInput(
         k_grad_input[l] = static_cast<T>(f_grad_input);
       }
     }
+    // prevent race where buf is written again before reads are done
+    __syncthreads();
   }
 }