update layernorm.md

burried -> buried

doc/layernorm/layernorm.md

@@ -3,7 +3,7 @@
 
 Quick tutorial. Let's look at how LayerNorm is handled, as one example layer in the model. We start with the [PyTorch docs for LayerNorm](https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html). LayerNorm of course comes from this original paper by [Ba et al. 2016](https://arxiv.org/abs/1607.06450), and was incorporated into the Transformer in [Vaswani et al.](https://arxiv.org/abs/1706.03762) famous paper Attention is All You Need. [GPT-2](https://d4mucfpksywv.cloudfront.net/better-language-models/language_models_are_unsupervised_multitask_learners.pdf) picked up the same architecture as the Transformer, but the position of the LayerNorm was famously moved into what is now called the pre-normalization version. That is, the residual path of the Transformer is kept clean, and the LayerNorms are now the first layer of each block of the Transformer. This positively improves training stability.
 
-The first thing to note when looking at [PyTorch LayerNorm](https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html) is that you will most likely not be able to find the actual implementation of the equation. That's because it is burried 30 layers deep in the code, behind an inscrutable dynamical dispatcher, in some possibly auto-generated CUDA code (for those who are interested in details, see [layer_norm.cpp](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/layer_norm.cpp) and  [layer_norm_kernel.cu](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/cuda/layer_norm_kernel.cu)). This is done because PyTorch really really cares about efficiency, fair enough. For our purposes though, we have to start by first implementing LayerNorm manually using simpler PyTorch operations. This will be a lot less efficient than just forwarding a `LayerNorm` module, but it is algorithmically instructive. So here is the direct implementation of the math of LayerNorm using simpler PyTorch operations:
+The first thing to note when looking at [PyTorch LayerNorm](https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html) is that you will most likely not be able to find the actual implementation of the equation. That's because it is buried 30 layers deep in the code, behind an inscrutable dynamical dispatcher, in some possibly auto-generated CUDA code (for those who are interested in details, see [layer_norm.cpp](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/layer_norm.cpp) and  [layer_norm_kernel.cu](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/cuda/layer_norm_kernel.cu)). This is done because PyTorch really really cares about efficiency, fair enough. For our purposes though, we have to start by first implementing LayerNorm manually using simpler PyTorch operations. This will be a lot less efficient than just forwarding a `LayerNorm` module, but it is algorithmically instructive. So here is the direct implementation of the math of LayerNorm using simpler PyTorch operations:
 
 ```python
 import torch