[doc] added reference to related works (#2994)

* [doc] added reference to related works

* polish code

docs/source/en/concepts/paradigms_of_parallelism.md

@@ -119,5 +119,6 @@ model on a single machine.
 </figure>
 
 Related paper:
+- [ZeRO-Offload: Democratizing Billion-Scale Model Training](https://arxiv.org/abs/2101.06840)
 - [ZeRO-Infinity: Breaking the GPU Memory Wall for Extreme Scale Deep Learning](https://arxiv.org/abs/2104.07857)
 - [PatrickStar: Parallel Training of Pre-trained Models via Chunk-based Memory Management](https://arxiv.org/abs/2108.05818)

docs/source/en/features/nvme_offload.md

@@ -5,6 +5,11 @@ Author: Hongxin Liu
 **Prerequisite:**
 - [Zero Redundancy Optimizer with chunk-based memory management](../features/zero_with_chunk.md)
 
+**Related Paper**
+
+- [ZeRO-Offload: Democratizing Billion-Scale Model Training](https://arxiv.org/abs/2101.06840)
+- [ZeRO-Infinity: Breaking the GPU Memory Wall for Extreme Scale Deep Learning](https://arxiv.org/abs/2104.07857)
+
 ## Introduction
 
 If a model has `N` parameters, when using Adam, it has `8N` optimizer states. For billion-scale models, optimizer states take at least 32 GB memory. GPU memory limits the model scale we can train, which is called GPU memory wall. If we offload optimizer states to the disk, we can break through GPU memory wall.

docs/source/en/features/zero_with_chunk.md

@@ -1,6 +1,7 @@
 # Zero Redundancy Optimizer with chunk-based memory management
 
 Author: [Hongxiu Liu](https://github.com/ver217), [Jiarui Fang](https://github.com/feifeibear), [Zijian Ye](https://github.com/ZijianYY)
+
 **Prerequisite:**
 - [Define Your Configuration](../basics/define_your_config.md)
 
@@ -9,9 +10,11 @@ Author: [Hongxiu Liu](https://github.com/ver217), [Jiarui Fang](https://github.c
 - [Train GPT with Colossal-AI](https://github.com/hpcaitech/ColossalAI/tree/main/examples/language/gpt)
 
 **Related Paper**
+
 - [ZeRO: Memory Optimizations Toward Training Trillion Parameter Models](https://arxiv.org/abs/1910.02054)
 - [ZeRO-Offload: Democratizing Billion-Scale Model Training](https://arxiv.org/abs/2101.06840)
 - [ZeRO-Infinity: Breaking the GPU Memory Wall for Extreme Scale Deep Learning](https://arxiv.org/abs/2104.07857)
+- [DeepSpeed: System Optimizations Enable Training Deep Learning Models with Over 100 Billion Parameters](https://dl.acm.org/doi/10.1145/3394486.3406703)
 - [PatrickStar: Parallel Training of Pre-trained Models via Chunk-based Memory Management](https://arxiv.org/abs/2108.05818)
 
 ## Introduction

docs/source/zh-Hans/concepts/paradigms_of_parallelism.md

@@ -87,5 +87,6 @@
 </figure>
 
 相关文章:
+- [ZeRO-Offload: Democratizing Billion-Scale Model Training](https://arxiv.org/abs/2101.06840)
 - [ZeRO-Infinity: Breaking the GPU Memory Wall for Extreme Scale Deep Learning](https://arxiv.org/abs/2104.07857)
 - [PatrickStar: Parallel Training of Pre-trained Models via Chunk-based Memory Management](https://arxiv.org/abs/2108.05818)

docs/source/zh-Hans/features/nvme_offload.md

@@ -5,6 +5,10 @@
 **前置教程:**
 - [基于Chunk内存管理的零冗余优化器 (ZeRO)](../features/zero_with_chunk.md)
 
+**相关论文**
+
+- [ZeRO-Offload: Democratizing Billion-Scale Model Training](https://arxiv.org/abs/2101.06840)
+- [ZeRO-Infinity: Breaking the GPU Memory Wall for Extreme Scale Deep Learning](https://arxiv.org/abs/2104.07857)
 ## 引言
 
 如果模型具有`N`个参数，在使用 Adam 时，优化器状态具有`8N`个参数。对于十亿规模的模型，优化器状态至少需要 32 GB 内存。 GPU显存限制了我们可以训练的模型规模，这称为GPU显存墙。如果我们将优化器状态 offload 到磁盘，我们可以突破 GPU 内存墙。

docs/source/zh-Hans/features/zero_with_chunk.md

@@ -3,9 +3,11 @@
 作者: [Hongxiu Liu](https://github.com/ver217), [Jiarui Fang](https://github.com/feifeibear), [Zijian Ye](https://github.com/ZijianYY)
 
 **前置教程:**
+
 - [定义配置文件](../basics/define_your_config.md)
 
 **示例代码**
+
 - [Train GPT with Colossal-AI](https://github.com/hpcaitech/ColossalAI/tree/main/examples/language/gpt)
 
 **相关论文**
@@ -13,8 +15,10 @@
 - [ZeRO: Memory Optimizations Toward Training Trillion Parameter Models](https://arxiv.org/abs/1910.02054)
 - [ZeRO-Offload: Democratizing Billion-Scale Model Training](https://arxiv.org/abs/2101.06840)
 - [ZeRO-Infinity: Breaking the GPU Memory Wall for Extreme Scale Deep Learning](https://arxiv.org/abs/2104.07857)
+- [DeepSpeed: System Optimizations Enable Training Deep Learning Models with Over 100 Billion Parameters](https://dl.acm.org/doi/10.1145/3394486.3406703)
 - [PatrickStar: Parallel Training of Pre-trained Models via Chunk-based Memory Management](https://arxiv.org/abs/2108.05818)
 
+
 ## 引言
 
 零冗余优化器 (ZeRO) 通过对三个模型状态（优化器状态、梯度和参数）进行划分而不是复制他们，消除了数据并行进程中的内存冗余。该方法与传统的数据并行相比，内存效率得到了极大的提高，而计算粒度和通信效率得到了保留。