This training introduces PLGrid LLM Forge, an LLM inference service deployed on Cyfronet supercomputers. The platform allows users to access large language models (such as Bielik or Llama) via an OpenAI-compatible API. It simplifies the setup of an inference environment and provides monitoring of usage and accounting. During this session, you are going to learn how to apply for this resource, generate an API token, and use it in basic examples with Python. The training will also cover more complex hands-on exercises using the PLGrid LLM Forg service. The course is intended for PLGrid users who want to start using LLMs without the complicated setup.
Technical prerequisites:
* PLGrid account at https://portal.plgrid.pl/;
* an active affiliation at https://portal.plgrid.pl/users/affiliations;
* access to the Ares supercomputer,
* if a participant has access, no further action,
* if not, one needs to apply at https://portal.plgrid.pl/services/103, and the access will be granted when joining the team (see below),
* participants are not required to have their own grant allocations;
* access to JupyterHubHPC, apply at https://portal.plgrid.pl/services/110;
* membership in the "plggtraining" team - the activation link will be sent after registration.
The above prerequisites are required to actively participate in the hands-on exercises.
In the absence of an active affiliation or PLGrid account, please contact the organisers at training@cyfronet.pl.
Participants' computers should have SSH client software installed. On Windows, it is recommended to use Windows Terminal after checking if OpenSSH is installed. We advise against using Putty.
Duration
2 hours
Trainers
Szymon Mazurek (ACC Cyfronet AGH): Deep Learning Engineer in HPC Software Department, specializing in model development, performance profiling, optimization, and benchmarking.
Oskar Klimas (ACC Cyfronet AGH): HPC trainer, computational chemistry specialist at Helpdesk, with experience in using a variety of applications, including Gaussian, AMS ADF, GROMACS, and LAMMPS.
Training language
English
