--- title: "Siam Inference | AI Inference Community in Thailand" description: "Siam Inference is a Thai community for AI inference: running and serving LLMs, GPUs and accelerators, vLLM, quantization and Thai-language models." url: "https://siaminference.ai/en/" lang: "en" alternate: "https://siaminference.ai/index.md" --- # A Thai community for AI inference. A place to share what it takes to run AI models fast, affordably and in production, from hardware and engines to applications. ## What is AI inference, and why a community? AI inference is the moment a trained AI model is put to work. Every chatbot reply, every summarised document and every image a model reads is AI inference. ### What AI inference is **The part where the model works for its users** AI inference is what comes after training: a trained model, such as a large language model (LLM), takes a question and produces an answer. Training teaches the model. AI inference is the part people actually touch, and the part that has to stay available. ### Why AI inference matters **Speed and cost are decided here** How long a person waits for an answer, how many people a system can serve at once and what each answer costs are all decided at the AI inference step. They come down to the hardware, the engine and the settings you choose. ### Why a community **Learn from people who have tried it** AI inference tools change quickly, and the knowledge that holds up usually comes from someone who has already tried. Siam Inference is a place for people in Thailand who work on AI inference to share what they tried, what they measured and what they learned. ## Three things Siam Inference talks about Siam Inference talks about three things: the machines models run on, the software that makes them answer quickly, and the people doing the work. ### Compute *Where models run* Compute is the machine an AI model runs on: a GPU, a purpose-built accelerator, a server in your own building (on-premises) or the cloud. Each choice trades speed, price and upkeep differently. - **GPUs and accelerators**: Matching the machine to the model’s size and the budget. - **On-prem and cloud**: Running it yourself, renting it, or mixing the two. ### Inference engines *What makes it fast* An inference engine is the software that loads a model and serves it to users; vLLM is one example. Techniques such as batching requests, shrinking the model (quantization) and managing memory let the same machine answer sooner and serve more people. - **LLM serving**: Serving one model to many people at once. - **Quantization**: Shrinking a model so it needs less memory. ### Community *The people who build* Knowledge about AI inference lives with the people doing the work: measuring in a way others can compare (benchmarking), using models for Thai-language tasks, and the lessons from putting a system into production. - **Benchmarking**: Measuring, and saying how, so results can be compared. - **Thai-language models**: What it is like to use models for Thai-language work. ### Technical terms, explained - **AI inference**: Using a trained model to produce an answer. - **LLM (large language model)**: An AI model that reads and writes text, such as the model behind a chatbot. - **GPU / accelerator**: A chip that does a model’s arithmetic faster than a general-purpose CPU. - **Inference engine**: Software that loads a model and answers requests, such as vLLM. - **Batching**: Processing several requests together in one pass. - **Quantization**: Storing a model’s numbers at lower precision so it needs less memory. - **TTFT (time to first token)**: How long a person waits before an answer starts to appear. - **Throughput**: How much work a system gets through in a given time. - **Benchmark**: Measuring systems the same way so they can be compared. ## Sites in the Siam Inference network Siam Inference is in the same network as these sites. Each one has its own content and its own focus. ### [vLLM Bangkok](https://vllmbangkok.com/en/) *Community* The Thai community around vLLM, the open-source engine that serves large models fast, for the people who build AI inference and infrastructure. - [Facebook](https://www.facebook.com/vllmbangkok) ### [Cerebras Thailand](https://cerebras.siaminference.ai/en/) *Community* A community for Cerebras users and developers in Thailand. - [Facebook](https://www.facebook.com/cerebrasthailand) ### [AGICAFET](https://agicafet.com/) *Company* A Thailand-based company that sets up private AI that works where you are: Local AI that runs in your office, Hybrid AI that adds the cloud when you choose, and Physical AI inside machines and robots. ### [PreMize](https://premize.agicafet.com/) *Product* AGICAFET’s product: private AI in one box, a ready-to-run AI system on NVIDIA DGX Spark. ## Frequently asked questions ### What is AI inference? AI inference is using a trained AI model to do real work: answering a question in a chatbot, summarising a document or reading an image. Every time someone uses a model, that is AI inference. ### How is AI inference different from training? AI inference uses a trained model to answer new requests without learning anything new. Training is the earlier step, which teaches a model by adjusting its internal values on large amounts of data. The wait a user feels and the cost of each answer are therefore decided at the AI inference step. ### Can I run an LLM on my own computer? You can run an LLM on your own computer if the machine has enough memory for the model, especially the graphics card’s VRAM. Quantization lets a model use less memory. Tools such as Ollama and llama.cpp run models on a personal machine, while vLLM is designed to serve a model to many users at once. ### Who is Siam Inference for? Siam Inference is for anyone who works on, or is curious about, running AI models: engineers, developers, researchers, students and people choosing a system for their organisation. ### Does it cost anything to join Siam Inference? The Siam Inference website has no sign-up and no fees. Ways to join Siam Inference are coming soon. For events, follow each community in the network on its own channels. ### How is Siam Inference related to vLLM Bangkok, Cerebras Thailand and AGICAFET? Siam Inference, vLLM Bangkok, Cerebras Thailand and AGICAFET are sites in the same network, each with its own content. vLLM Bangkok is the Thai community around vLLM (vllmbangkok.com). Cerebras Thailand is a community for Cerebras users and developers in Thailand (cerebras.siaminference.ai). AGICAFET is a Thailand-based company that sets up private AI, and PreMize is its product (agicafet.com). ### Where can I follow Siam Inference? Siam Inference’s own channel is coming soon. For now, follow vLLM Bangkok (facebook.com/vllmbangkok) and Cerebras Thailand (facebook.com/cerebrasthailand) on Facebook. ## Join Siam Inference **Coming Soon** Siam Inference’s own channel is coming soon. For now, follow the vLLM Bangkok and Cerebras Thailand pages on Facebook. - [vLLM Bangkok on Facebook](https://www.facebook.com/vllmbangkok) - [Cerebras Thailand on Facebook](https://www.facebook.com/cerebrasthailand) --- Siam Inference is a Thai community for AI inference: running and serving LLMs, GPUs and accelerators, vLLM, quantization and Thai-language models. vLLM, NVIDIA, Cerebras and other names are trademarks of their respective owners.