vLLM, Function Calling, and World Models explained
How vLLM, Function Calling, and World Models actually work
How does vLLM work?
In this blog, we will learn about how vLLM works. We will also see why we need it, how it manages memory so cleverly, and where it is used in the real world to serve large language models to many users at once.
Serving an LLM means running the model on a computer so that many users can send it questions and get answers at the same time. So, the whole game of serving is this: we want to serve as many users as possible, as fast as possible, on one GPU. So, the real bottleneck in serving an LLM is not the math speed. It is the KV cache memory.
vLLM is a high-throughput engine for serving LLMs. It is built to serve as many requests as possible on a GPU by managing the KV cache memory very efficiently. It solves the memory problem with two main ideas working together:
PagedAttention, which manages the KV cache in small fixed-size blocks instead of one giant block, so no memory is wasted.
Continuous batching, which keeps the GPU busy by swapping finished requests out and new ones in at every step.
In simple words, vLLM lets us serve more users, faster, and cheaper, on the same GPU, without changing the quality of the answers. The model still produces the same replies. vLLM just stops wasting memory and time. That’s the beauty of vLLM.
Read it here: https://outcomeschool.com/blog/how-does-vllm-work
How does Function Calling work in LLMs?
In this blog, we will learn about how Function Calling works in LLMs. We will see what it is, why we need it, the key insight behind it, and how it powers AI agents and assistants step by step.
Function Calling is a way to let an LLM use external tools, APIs, and functions to get things done. It is also called tool calling, and both names mean the same thing. We need it because an LLM only generates text. It cannot fetch live data, and it cannot take real actions on its own.
This is the most important idea in the whole blog: the model does not execute the function. It only outputs a structured request to run it. Our application runs the function. The model is the brain, and our application is the hands.
This split is what keeps everything safe and correct. Our code stays fully in control. The model can only suggest a function call. It can never run anything by itself. This is why Function Calling is the backbone of AI agents and assistants. It is the bridge between language and action.
Read it here: https://outcomeschool.com/blog/how-does-function-calling-work-in-llms
How do World Models work?
In this blog, we will learn about how World Models work. We will also see why we need them, how they actually learn an internal picture of an environment, and where they are used in real systems like robotics, game-playing, and video generation.
A World Model is an AI that learns an approximate internal copy of how an environment behaves, so that it can predict what happens next, given the current state and an action. In simple words, a World Model is a little simulator that the AI builds inside its own head. We can ask it a question like “if I am in this state and I take this action, what happens next?” and it gives back an answer: the next state, and how good or bad that outcome is.
Think of chess. Before we actually touch a piece, we imagine. We play the move in our head first, look at the imagined outcome, and only if it looks good, we make the real move. That little simulation inside our mind is exactly a World Model. So, a World Model gives an AI this same ability: the power to imagine the outcome of an action before taking it in the real world.
This is why it matters. A World Model is, at its core, a machine that predicts the future. When an AI can predict the future, it can plan. Because all of this imagining happens inside the model, the AI can practice millions of times very quickly and cheaply, without breaking any real robots and without waiting for the slow real world. This makes learning faster, cheaper, safer, and far more sample-efficient than learning by raw trial and error.
Read it here: https://outcomeschool.com/blog/how-do-world-models-work
New on YouTube
Two short explainers on LLM internals that trip up a lot of people:
The First-Token Latency Problem in LLMs:
Why is the context window limited in LLMs?
That’s it for now.