👋 Welcome to Yue’s blog

In recent years, Large Language Models (LLMs) have achieved revolutionary breakthroughs in fields such as natural language processing, code generation, and even multimodal interaction. However, the powerful capabilities of these models come at the cost of enormous computational and memory overhead, especially during the inference stage. Efficiently deploying and running these models, which have billions or even trillions of parameters, has become a core challenge in scaling LLM technology for real-world applications. ...

As the parameters of Large Language Models (LLMs) continue to grow, deploying and serving these models presents significant challenges. vLLM is an open-source library designed for fast, convenient, and cost-effective LLM inference and online serving. Its core lies in the PagedAttention algorithm, which efficiently manages the KV Cache in the attention mechanism. Evaluation Metrics To evaluate the performance of LLM inference and serving engines, we primarily focus on the following metrics: ...

Humans interact with the world through multiple senses (vision, hearing, touch, etc.), with each sensory channel offering unique advantages in representing and communicating specific concepts. This multimodal interaction fosters our deep understanding of the world. One of the core goals in the field of artificial intelligence is to develop general-purpose assistants that can effectively follow multimodal instructions (such as visual and linguistic ones), enabling them to perform various real-world tasks like humans. In recent years, with the release of models like GPT-4o (OpenAI, 2024), Gemini 2.5 Pro (DeepMind, 2025), and o3/o4-mini (OpenAI, 2025), Multimodal Large Language Models (MLLMs) have made significant progress. They can not only understand information from multiple modalities like images, videos, and audio but also perform complex reasoning and generation. ...

DeepSeek AI successively released DeepSeek-V2 (DeepSeek-AI, 2024) and DeepSeek-V3 (DeepSeek-AI, 2024), two powerful Mixture-of-Experts (MoE) language models that significantly optimize training costs and inference efficiency while maintaining state-of-the-art performance. DeepSeek-V2 has a total of 236B parameters, activating 21B per token, while DeepSeek-V3 further expands to 671B total parameters, activating 37B per token. Both support a 128K context length. The core innovations of these two models lie in the adoption of Multi-head Latent Attention (MLA) and the DeepSeekMoE architecture (Dai et al., 2024). MLA drastically reduces GPU memory usage during inference by compressing the Key-Value (KV) cache into low-dimensional latent vectors, improving efficiency. DeepSeekMoE achieves stronger expert specialization capabilities and more economical training costs through fine-grained expert segmentation and shared expert isolation. Building upon V2, DeepSeek-V3 further introduces an Auxiliary-Loss-Free Load Balancing strategy (Wang et al., 2024) and the Multi-Token Prediction (MTP) (Gloeckle et al., 2024) training objective, further enhancing model performance and training efficiency. ...

LLaMA The LLaMA series of open-source models released by Meta AI has become one of the cornerstones of the large language model community, profoundly impacting the advancement of open research and applications. From the pioneering LLaMA released in early 2023, to the significantly improved LLaMA 2 later that year, to derivative models targeting specific domains (like code, safety), and the subsequent new generations LLaMA 3 and LLaMA 4 launched in 2024 and 2025 respectively, Meta has continuously committed to enhancing the performance of open-source models, gradually bringing them closer to state-of-the-art closed-source models. Below, we will introduce the key technical details of each major model in sequence. ...