Enhance NVFP4 blog content with code exercises

- Updated the NVFP4 pretraining blog to include hands-on code exercises for better understanding of the concepts.
- Improved links to direct users to relevant resources, including a new exercise notebook for practical application.

Author: vukrosic

public/content/pretrain-llm-with-nvfp4/pretrain-llms-with-fp4-content-zh.md

@@ -7,7 +7,7 @@ tags:
 - "📄 研究论文"
 ---
 
-[研究论文](https://arxiv.org/pdf/2509.25149) • [实现 PR](https://github.com/NVIDIA/TransformerEngine/pull/2177)
+[📄 研究论文](https://arxiv.org/pdf/2509.25149) • [⚙️ 实现 PR](https://github.com/NVIDIA/TransformerEngine/pull/2177) • [🧪 代码练习](https://colab.research.google.com/gist/vukrosic/2c0117344dd269263adf0b6e5382889f/excercise.ipynb)
 
 # 使用 NVFP4 预训练大语言模型的技术指南
 
@@ -229,6 +229,12 @@ H =     [ 0.5, -0.5,  0.5, -0.5 ]
 
 结合指定训练方法的 NVFP4 能够实现大语言模型在 4 位精度下的稳定、准确预训练。该方法在计算吞吐量和内存使用方面提供显著效率增益，同时不损害模型性能。NVIDIA Transformer Engine 已全面支持 NVFP4。
 
+## 代码练习
+
+为了加深您对 NVFP4 概念和实现的理解，我们准备了实践练习，演示本文讨论的关键技术：
+
+**[🧪 NVFP4 实现练习](https://colab.research.google.com/gist/vukrosic/2c0117344dd269263adf0b6e5382889f/excercise.ipynb)**
+
 ---
 
 ***来源***：*本指南是对技术报告《[Pretraining Large Language Models with NVFP4](https://arxiv.org/pdf/2509.25149v1)》的总结。完整细节请参阅原始出版物。*

public/content/pretrain-llm-with-nvfp4/pretrain-llms-with-fp4-content.md

@@ -7,7 +7,7 @@ tags:
 - "📄 Research Article"
 ---
 
-[Research Paper](https://arxiv.org/pdf/2509.25149) • [Implementation PR](https://github.com/NVIDIA/TransformerEngine/pull/2177)
+[📄 Research Paper](https://arxiv.org/pdf/2509.25149) • [⚙️ Implementation PR](https://github.com/NVIDIA/TransformerEngine/pull/2177) • [🧪 Code Exercises](https://colab.research.google.com/gist/vukrosic/2c0117344dd269263adf0b6e5382889f/excercise.ipynb)
 
 # A Technical Guide to LLM Pretraining with NVFP4
 
@@ -232,6 +232,12 @@ To achieve the same final training loss as the model trained with NVFP4, the mod
 
 NVFP4, when combined with the specified training methodology, enables stable and accurate pretraining of large-scale language models in 4-bit precision. This approach offers significant efficiency gains in terms of computational throughput and memory usage without compromising model performance. Full support for NVFP4 is available in NVIDIA's Transformer Engine.
 
+## Code Exercises
+
+To deepen your understanding of NVFP4 concepts and implementation, we've prepared hands-on exercises that demonstrate the key techniques discussed in this article:
+
+**[🧪 NVFP4 Implementation Exercises](https://colab.research.google.com/gist/vukrosic/2c0117344dd269263adf0b6e5382889f/excercise.ipynb)**
+
 ---
 
 ***Source:*** *This guide is a summary of the technical report "[Pretraining Large Language Models with NVFP4](https://arxiv.org/pdf/2509.25149v1)". For complete details, please refer to the original publication.*