Fix broken links in README

Author: li-yi-dong

README.md

@@ -35,7 +35,7 @@ Therefore, to facilitate the training of LLaMA-based models and reduce the cost
 
 # OverlappedDistributedOptimizer
 
-In the vanilla Megatron-LM, users can leverage [`DistributedOptimizer`]("https://github.com/NVIDIA/Megatron-LM/blob/main/docs/distrib_optimizer.md") to partition gradients and optimizer states to reduce GPU memory occupation. After accumulated all gradients in GA, `DistributedOptimizer` employs a `ReduceScatter` operation to scatter the gradients to the corresponding ranks. Each rank then updates the local parameters, and then collect the remaining parameters through an `AllGather` operation from all the other ranks. However, we observe a significant overhead on communication under small GA settings (over 50% time consumption without GA). 
+In the vanilla Megatron-LM, users can leverage [`DistributedOptimizer`](https://github.com/NVIDIA/Megatron-LM/blob/main/docs/distrib_optimizer.md) to partition gradients and optimizer states to reduce GPU memory occupation. After accumulated all gradients in GA, `DistributedOptimizer` employs a `ReduceScatter` operation to scatter the gradients to the corresponding ranks. Each rank then updates the local parameters, and then collect the remaining parameters through an `AllGather` operation from all the other ranks. However, we observe a significant overhead on communication under small GA settings (over 50% time consumption without GA). 
 
 To mitigate the overhead, we try to overlap the collective communication with computation, according to the partition strategy in DeepSpeed ZeRO Stage-2. This strategy fails to scale. It takes too many small `Reduce` operations at large scale, which makes it under-utilize the inter-connection bandwidth.
 
@@ -125,7 +125,7 @@ In particular, we recommend to increase the micro-batch size to fully occupy the
 | `--tokenizer-type=PretrainedFromHF` | Use a Tokenizer from Huggingface (would be loaded via `transformers.AutoTokenizer`) |
 | `--distributed-checkpointing` | Distributed saving of checkpoint files. |
 
-Megatron-LLaMA supports the canonical [data prepocessing]("https://github.com/NVIDIA/Megatron-LM/blob/main/README.md#data-preprocessing") and [evaluation]("https://github.com/NVIDIA/Megatron-LM/blob/main/README.md#evaluation-and-tasks") as mentioned in the Megatron-LM library.
+Megatron-LLaMA supports the canonical [data prepocessing](https://github.com/NVIDIA/Megatron-LM/blob/main/README.md#data-preprocessing) and [evaluation](https://github.com/NVIDIA/Megatron-LM/blob/main/README.md#evaluation-and-tasks) as mentioned in the Megatron-LM library.
 
 ### Future work
 
@@ -144,9 +144,9 @@ Megatron-LLaMA is developed by Aicheng Technology, Alibaba Group and is based on
 
 The following repositories are used in Megatron-LLaMA, either in close to original form or as an inspiration:
 
-[Megatron-LM]("https://github.com/NVIDIA/Megatron-LM")
+[Megatron-LM](https://github.com/NVIDIA/Megatron-LM)
 
-[LLaMA]("https://github.com/facebookresearch/llama")
+[LLaMA](https://github.com/facebookresearch/llama)
 
-[DeepSpeed]("https://github.com/microsoft/DeepSpeed")
+[DeepSpeed](https://github.com/microsoft/DeepSpeed)
 

README_zh.md

@@ -36,7 +36,7 @@ LLaMA是目前大语言模型开源社区中一项重要工作。LLaMA在LLM的
 
 ## 2. Megatron-LLaMA中`OverlappedDistributedOptimizer`简介
 
-在原生Megatron-LM中，用户可以使用[`DistributedOptimizer`]("https://github.com/NVIDIA/Megatron-LM/blob/main/docs/distrib_optimizer.md")来切分梯度和优化器状态，以减少训练中的显存占用。`DistributedOptimizer`在每次获得预设的梯度聚合组梯度后，通过`ReduceScatter`算子，将之前累积的全部梯度分发到不同的Rank。每个Rank更新完属于自己的参数后，再通过`AllGather`算子将更新后的参数复制到所有Rank。在实际训练中，我们观察到`DistributedOptimizer`的集合通信在梯度聚合较小的情况下，将引入极大的额外开销。极端情况下，不使用梯度聚合，将引入超过整体耗时50%的额外开销。
+在原生Megatron-LM中，用户可以使用[`DistributedOptimizer`](https://github.com/NVIDIA/Megatron-LM/blob/main/docs/distrib_optimizer.md)来切分梯度和优化器状态，以减少训练中的显存占用。`DistributedOptimizer`在每次获得预设的梯度聚合组梯度后，通过`ReduceScatter`算子，将之前累积的全部梯度分发到不同的Rank。每个Rank更新完属于自己的参数后，再通过`AllGather`算子将更新后的参数复制到所有Rank。在实际训练中，我们观察到`DistributedOptimizer`的集合通信在梯度聚合较小的情况下，将引入极大的额外开销。极端情况下，不使用梯度聚合，将引入超过整体耗时50%的额外开销。
 
 在尝试实现通信和计算并行的过程中，我们尝试了DeepSpeed ZeRO2中对梯度以及优化器状态的切分方式。在超大规模的场景下，我们观察到其切分方式需要大量细碎的通信Kernel，无法充分利用通信带宽，造成了通信耗时过长，模型的计算量不足以与通信充分并行。
 
@@ -142,8 +142,8 @@ Megatron-LLaMA使用Apache 2.0开源协议，允许用作商业用途。详情
 
 ### 参考工作
 
-[Megatron-LM]("https://github.com/NVIDIA/Megatron-LM")
+[Megatron-LM](https://github.com/NVIDIA/Megatron-LM)
 
-[LLaMA]("https://github.com/facebookresearch/llama")
+[LLaMA](https://github.com/facebookresearch/llama)
 
-[DeepSpeed]("https://github.com/microsoft/DeepSpeed")
+[DeepSpeed](https://github.com/microsoft/DeepSpeed)