update links

blog/2024-07-25-sglang-llama3.md

@@ -5,9 +5,9 @@ date: "Jul 25, 2024"
 previewImg: /images/blog/sglang_llama3/preview.png
 ---
 
-At LMSYS.org, we've been running the [Chatbot Arena](https://chat.lmsys.org/) platform for over a year, serving millions of users. We know firsthand how crucial efficient serving is for AI products and research. Through our operational experiences and in-depth research, we've continuously enhanced the underlying serving systems, spanning from the high-level multi-model serving framework, [FastChat](https://github.com/lm-sys/FastChat/tree/main), to the efficient serving engine, [SGLang Runtime (SRT)](https://github.com/sgl-project/sglang/tree/main).
+At LMSYS.org, we've been running the [Chatbot Arena](https://chat.lmsys.org/) platform for over a year, serving millions of users. We know firsthand how crucial efficient serving is for AI products and research. Through our operational experiences and in-depth research, we've continuously enhanced the underlying serving systems, spanning from the high-level multi-model serving framework, [FastChat](https://github.com/lm-sys/FastChat/tree/main), to the efficient serving engine, [SGLang Runtime (SRT)](https://github.com/sgl-project/sglang).
 
-This post focuses on SGLang Runtime, a general-purpose serving engine for LLMs and VLMs. While existing options like TensorRT-LLM, vLLM, MLC-LLM, and Hugging Face TGI have their merits, we found them sometimes hard to use, difficult to customize, or lacking in performance. This motivated us to develop SGLang v0.2, aiming to create a serving engine that is not only user-friendly and easily modifiable but also delivers top-tier performance. While SGLang includes frontend language features, this post will focus solely on the backend runtime and use "SGLang" and "SGLang Runtime" interchangeably to refer to the runtime.
+This post focuses on [SGLang Runtime](https://github.com/sgl-project/sglang), a general-purpose serving engine for LLMs and VLMs. While existing options like TensorRT-LLM, vLLM, MLC-LLM, and Hugging Face TGI have their merits, we found them sometimes hard to use, difficult to customize, or lacking in performance. This motivated us to develop SGLang v0.2, aiming to create a serving engine that is not only user-friendly and easily modifiable but also delivers top-tier performance. While SGLang includes frontend language features, this post will focus solely on the backend runtime and use "SGLang" and "SGLang Runtime" interchangeably to refer to the runtime.
 
 Compared to TensorRT-LLM and vLLM, SGLang Runtime consistently delivers superior or competitive performance in both online and offline scenarios, handling models from Llama-8B to Llama-405B, and on A100 and H100 GPUs, using FP8 and FP16. **SGLang consistently outperforms vLLM, achieving up to 3.1x higher throughput on Llama-70B. It also often matches or sometimes outperforms TensorRT-LLM**. More importantly, SGLang is fully open-source, written in pure Python, with the core schedulers implemented in fewer than 4K lines of code.