How to

1st Thailand LLM Bootcamp

Matichon Maneegard

Matichon Maneegard

— 2 min read
1st Thailand LLM Bootcamp

Float16.cloud has been a part of hosting the Bootcamp with OpenACC, NVIDIA, Siam.ai, and AIEI.

The Thailand LLM Bootcamp will take place over 2 days. Day 1 (February 28) will be held online, and Day 2 will be held in person at Central World, Thailand.

https://www.openhackathons.org/s/siteevent/a0CUP00001Yalpp2AB/se000400

Participants will receive dedicated access to an H100, with one instance per person, provided by Siam.ai (NCP).

Agenda

This bootcamp will provide hands-on experience with an end-to-end LLM project. It is divided into 3 parts

1.Continue Fine-Tuning LLM Models

Learn how to continue fine-tuning LLM models, such as the Llama model, with NVIDIA NeMo while leveraging PEFT techniques like QLoRA.

2.Optimize LLM Models

The second part will help you reduce inference resources for LLM models. We will focus on reducing and balancing the model's ability with performance (model intelligence vs. tokens per second). We will explore quantization using NVIDIA TensorRT-LLM (TRT-LLM).

3.Serving the LLM Model

The final part will guide you in deploying the model in a production environment using best practices from the NVIDIA team. We will leverage the serving engine with Triton Inference Server and use advanced techniques like inflight batching, context caching, and paged KV cache.

Float16 teams

Our Float16 team is responsible for setting up the resources and providing Jupyter notebooks to the participants. In addition to setting up the resources, we will also serve as instructors and TAs for the event.

Lastly, this bootcamp is the first LLM bootcamp in Thailand. Float16 plans to localize and improve the bootcamp throughout the year and to continue holding LLM bootcamps across Southeast Asia (SEA), not limited to Thailand.

Contact Float16

Read more