ALICE-SHARK User Meeting 2025
The ALICE-SHARK User Meeting 2025 is over. Thank you very much to all participants and speakers. We hope that you enjoyed the meeting and discussions as much as we do. We look forwared to staying in touch with you and to seeing you again at the meeting next year.
Announcement
We are excited to announce that the ALICE-SHARK User Meeting 2025 will take place on Tuesday, 3 June 2025 from 10:00 - 14:00 in the Pieter-de-la-Court building room 1B.01
If you are a current, past or possible future user of the ALICE HPC cluster at Leiden University or the SHARK HPC cluster at the Leiden University Medical Center (or both), this meeting is for you. Our goal is to bring the user communities of both clusters together to connect with each other and the support teams of both clusters. The meeting is also of interest to users who are not yet actively using one of the two clusters.
The meeting will include an overview and update for both clusters, a selection of talks from users and an interactive Q&A session with the support teams.
Attendance is possible in person and remotely. For in-person attendees, the meeting will be held in room 1B01 of the Pieter de la Court building. For remote attendees, all sessions will be streamed live with Microsoft Teams with the possibility to ask questions remotely.
Registration
The number of people that can attend in-person is limited by the space of the room and will be filled on a first-come-first-serve basis. But we still have seats left.
There is no registration fee for the meeting and lunch will be provided for in-person participants.
If you want to register for the meeting, please send an e-mail to the ALICE helpdesk (helpdesk@alice.leidenuniv.nl) with the information below:
ALICE or SHARK username (if available):
Have you already been using ALICE or SHARK actively? (Yes/No):
Do you plan to attend in person? (Yes/No):
If you plan to attend in person, do you have any dietary constraints or allergies?
Important dates
Abstract submission deadline (extended): Tuesday, 13 May 2025 at 23:59 CEST
Final programme: Friday, 16 May 2025
Deadline for registration: Sunday, 1 June 2025 at 23:59 CEST
User Meeting: Tuesday, 3 June 2025 from 10:00 till 14:00 CEST
When & Where
When: Tuesday, 3 June 2025 from 10:00 till 14:00 CEST
Where:
in-person participants: Peter De La Court building, room 1B01 and Microsoft Teams
remote participants: Microsoft Teams
Microsoft Teams space (for in-person and remote participants)
All participants (in-person and remote) will be added to a Microsoft Teams space created specifically for the meeting. This has worked well last year which is why we will use it again this year.
The advantage for us is that Teams provide both, live streaming (with chat for questions) and a separate chat functionality for all participants independent of the live stream in multiple channels in a single application. The tool is also officially supported by both organizations (LEI and LUMC).
The Teams space can be actively used by all participants. We strongly encourage you to make use of it before and during the meeting independent of how you attend the meeting. The General channel contains two tabs next to Files with information about the Teams space and the schedule.
The meeting space will be created about two weeks prior to the meeting. An invitation to join the Teams space will be send around to all participants.
Schedule
The following table shows a tentative schedule for the meeting:
Start Time | End Time | Title | Abstract | Speaker |
|---|---|---|---|---|
10:00 | 10:15 | Arrival Participants |
|
|
10:15 | 10:20 | Welcome |
|
|
10:20 | 10:40 | Overview/Update ALICE |
| ALICE Team |
10:40 | 11:00 | Overview/Update SHARK |
| SHARK Team |
11:00 | 11:05 | Handover speaker |
|
|
11:05 | 11:20 | slurm-viewer: A New Command Line Tool for Simplified SLURM Monitoring | Large SLURM clusters can be complex to monitor. | P. Mody & P. de Koning (LUMC) |
11:20 | 11:25 | Handover speaker |
|
|
11:25 | 11:40 | Designing Custom HPC Interfaces for Transparent Usage | HPC clusters provide massively greater computing resources compared to traditional desktop machines. Yet users are reluctant to adopt them over ad hoc solutions on their local devices. A common complaint about HPC clusters is that they are intimidating and difficult to understand, or otherwise obstruct the tasks at hand. This presentation, aimed both at beginning users and software engineers, explores the major problems preventing seamless use of HPC clusters, and several solutions ranging from dead simple to boutique software design. | C. Handy (LEI) |
11:40 | 11:45 | Handover speaker |
|
|
11:45 | 12:00 | Quantum Physics Computations on ALICE | Quantum Physics is mostly governed by a single equation, the Schrödinger equation. While easy to write down, it is a formidable to solve it. The main challenge is the exponential scaling of the description of quantum man-body systems. Our challenge as theoretical physicists is to find good approximations for this problem. One modern approach to the problem are tensor networks. In our talk, we will motivate the main computational hurdles, give a short introduction into tensor networks and demonstrate our workflow with an example project on the cluster. | K.S. Rai (LEI) & P. Emonts (LEI, Univ. Ulm) |
12:00 | 12:05 | Handover speaker |
|
|
12:05 | 12:45 | Lunch |
|
|
12:45 | 13:00 | From OCR to Multimodal LLMs: Building a Korean Textbook Corpus with ALICE | This talk presents a hybrid pipeline for processing historical Korean textbooks using local LLM resources on the ALICE HPC cluster. Our original workflow relied on traditional OCR (PDF → PNG → EasyOCR → text), followed by EXAONE-based LLM cleaning. We are now prototyping a second, cutting-edge pipeline that bypasses OCR entirely: scanned textbook images (or PDFs) are interpreted directly using a multimodal vision-language model, which is then followed by a high-performance, Korean-trained LLM (EXAONE-Deep 70B, Q5_K_M) for post-editing and formatting. We consider the trade-offs between accuracy, speed, and model performance across both pipelines. If feasible, the workflow would be optimized to run efficiently on a single ALICE node, leveraging quantized models and GPU-accelerated inference. This project demonstrates the potential of local LLM infrastructure to support complex, multilingual document analysis in the humanities and social sciences. | S. Denny (LEI) |
13:05 | 13:50 | Q&A with ALICE - SHARK Team |
| ALICE & SHARK Teams |
13:50 | 13:55 | Closing |
|
|