Fixing broken links (#1736) (#926)

* Fixing broken links

Co-authored-by: Sadaf Rasool <[email protected]>

Author: aws-rxgupta

libraries/neuronx-distributed/activation_memory_reduction.rst

@@ -96,7 +96,7 @@ In the activation memory equation, we have a quadratic term of `5abs^2`. As the
 faster rate. This quadratic term comes from the softmax computation. `Vijay Korthikanti et.al <https://browse.arxiv.org/pdf/2205.05198.pdf>`__ 
 propose `Selective activation checkpointing` where they only recompute the softmax and attention computation and thereby avoid saving the activations coming 
 from softmax and attention computation. This completely gets rid of the quadratic term and brings down the activation memory per layer to 
-`34sbh/t`. The LLama-7B example in `this tutorial <https://awsdocs-neuron.readthedocs-hosted.com/en/latest/libraries/neuronx-distributed/tutorials/training_llama2_7b.html#llama2-7b-tp-zero1-tutorial>`__ 
+`34sbh/t`. The LLama-7B example in `this tutorial <https://awsdocs-neuron.readthedocs-hosted.com/en/latest/libraries/neuronx-distributed/tutorials/training_llama_tp_zero1.html#llama2-7b-tp-zero1-tutorial>`__
 used selective activation checkpointing.
 
 

libraries/neuronx-distributed/tutorials/training_codegen25_7b.rst

@@ -3,7 +3,7 @@
 Training CodeGen2.5 7B with Tensor Parallelism and ZeRO-1 Optimizer (``neuronx-distributed``)
 ==============================================================================================
 
-In this tutorial, we showcase how to pretrain a CodeGen2.5 7B model for program synthesis. Since Codegen2.5's architecture is identical to the one of Llama2, you may want to take a look at our `Llama2 tutorial <https://awsdocs-neuron.readthedocs-hosted.com/en/latest/libraries/neuronx-distributed/tutorials/training_llama2_7b.html>`__ first.
+In this tutorial, we showcase how to pretrain a CodeGen2.5 7B model for program synthesis. Since Codegen2.5's architecture is identical to the one of Llama2, you may want to take a look at our `Llama2 tutorial <https://awsdocs-neuron.readthedocs-hosted.com/en/latest/libraries/neuronx-distributed/tutorials/training_llama_tp_zero1.html>`__ first.
 
 After setting up the environment and installing ``neuronx-distributed``, we need to download a data set containing source code (in this case Java code) and then preprocess and tokenize it to match the code-infill format (more about this below). Use the following commands to download the required files. Note, that we reuse our llama2 training files.