In this video we will guide you through two different AI based workflows, involving ChatUI and CVAT apps.
You will learn how to:
Use advanced CVAT features including auto-annotation, algorithmic assistance, management and analytics.
Use Chat UI as a flexible interface for hosting of various LLM models, and interact via a chat or API environment.
Use ChatUI for semantic search in a knowledge base.
Use CVAT as a powerful annotation tool, including image classification, object detection, semantic and instance segmentation, and video / 3D annotations.
In this video we will guide you through the complete pipeline of transcribing audio files from speech to text and editing and classifying transcription segments.
In this session, you’ll learn how to:
Use Transcriber for transcribing audio/video files. Transcriber is based on Open AI’s Whisper language model. The app can transcribe speech audio to text in various formats and uses the WhisperX package to perform speaker recognition.
Navigate the new, simple, drag and drop Transcriber user interface to make it easier for you to use AI to transcribe audio files.
Edit and classify the transcriptions with Speech Analyzer. Speech Analyzer is an application built on top of Label Studio, specifically optimized for dialogue analysis. It enables you to label, edit, and annotate transcriptions generated using Transcriber.
Perform a comprehensive dialogue analysis on UCloud involving transcribing audio files using Transcriber, followed by transcription analysis with Speech Analyzer.
All workflows will be executed inside a UCloud project environment with access to GPU resources.
Target audience: Researchers across all Departments, particularly Digital Humanities and Social Science, Students, AI interested.
In this video we will guide you through the complete pipeline of fine-tuning large language models (LLMs) for specialised tasks such as medical question-answering using NeMo Framework and Triton Inference Server.
Prepare and preprocess open-source datasets for fine-tuning.
Apply Parameter-Efficient Fine-Tuning (PEFT) using LoRA with NVIDIA NeMo Framework.
Deploy optimised LLMs using NVIDIATriton Inference Server and TensorRT-LLM.
Generate a synthetic Q&A dataset using Label Studio connected to a live inference backend.
Fine-tune and evaluate your customised LLM for domain-specific applications.
All workflows will be executed inside a UCloud project environment with access to GPU resources.
Target audience: Machine learning practitioners, researchers, and engineers interested in LLM customisation, domain adaptation, or scalable model deployment.
In this video we will go through the process of developing a course on UCloud, using UCloud Courses – a tool for hosting and managing university courses on UCloud. Wnt to know more about this new feature, check out our webinar recording introducing UCloud Courses.
Introduction
00:00 – Outline of the workshop agenda 03:10 – Introduction to the UCloud Courses concept 04:23 – Advantages of using UCloud Courses 05:56 – Outline of the steps involved in developing a UCloud Course 09:54 – Requesting and planning a UCloud Course 10:20 – Example of the planning and structuring of an existing UCloud Course
Developing a UCloud Course step by step
Setting up the environment 15:53 – Preliminary remarks 17:15 – Showing the existing UCloud Course that will be re-developed in the workshop 18:02 – Showing the UCloud-Courses GitHub repository 18:40 – Software requirements and -recommendations 20:22 – Short introduction to Git and GitHub 24:27 – Cloning the repository 25:15 – Creating a working branch 26:12 – What should go in the UCloud-Courses repository and what shouldn’t 27:24 – Installing the required dependencies
Creating a new UCloud Course and modifying the templates 28:00 – Creating a UCloud Course using a prepared script 30:31 – Walk-through of the different auto-generated files/folders in the course folder 33:35 – Short introduction to Docker 38:15 – Modifying the Dockerfile 42:32 – What should go in the UCloud-Courses repository and what shouldn’t 45:28 – Building the Docker image locally using a prepared script 47:54 – Modifying the starting script 01:00:50 – What can, shouldn’t, and mustn’t be changed while the course is running
Building and testing the course locally 01:05:00 – Re-building the Docker image locally using a prepared script 01:05:25 – Running a Docker container locally using a prepared script 01:10:58 – Opening the JupyterLab interface on localhost
Finalising the course app 01:12:48 – Opening a pull request and requesting code review 01:19:44 – Testing the course app on UCloud before it’s deployed
Closing remarks
01:25:44 – Reusing/updating an existing UCloud course 01:28:19 – The financial model in brief 01:29:38 – Useful links and resources
I denne video introducerer vi den nye UCloud Courses-app – et værktøj til at hoste og administrere universitetskurser i UCloud.
Dr. Federica Lo Verso guider dig igennem appens koncept og baggrund. Du får en praktisk demonstration af, hvordan du tilgår og bruger værktøjet, udforsker dets integration med GitHub, og hører direkte fra Dr. Himanshu Khandelia, som deler sine erfaringer med at bruge UCloud Courses i virkelige undervisningssituationer.
Vi guider dig også gennem ansøgningsprocessen, viser dig de tekniske og økonomiske krav og giver dig en forsmag på et kommende praktisk kursus.
Tidskoder:
00:00 – Introduktion ved Dr. Federica Lo Verso 01:23 – Baggrund: Lanceringen af den nye UCloud Courses-app 04:12 – Gennemgang: Hvor du finder Courses-appen, og hvordan du bruger den 07:00 – Opsummering af demonstrationen af UCloud Courses-appen 07:56 – Gennemgang: Sådan fungerer GitHub-repositoriet 10:07 – Fordele ved at bruge Courses på UCloud 11:12 – Introduktion til brugsscenarie: Dr. Himanshu Khandelia 11:58 – Dr. Khandelia deler sine erfaringer med at bruge UCloud Courses 14:00 – Live-demo: Dr. Khandelia viser sin brug af Courses og GitHub-integration 23:00 – Sådan ansøger du: Ansøgningsprocessen for UCloud Courses 24:21 – Påkrævede ressourcer til at køre et UCloud-kursus 25:28 – Sådan genbruger og opdaterer du eksisterende UCloud Courses 26:20 – Økonomisk model: Omkostninger og tilgængelig støtte 27:12 – Outro og teaser til det kommende praktiske kursus
Udvikling af din egen UCloud-kursusapp med en nyudviklet skabelonbaseret tilgang
Deltag i en praktisk workshop, hvor vi guider dig gennem alle trin i udviklingen af en UCloud-kursusapp ved hjælp af vores nyudviklede skabelonbaserede tilgang. Konceptet går ud på at have en dedikeret app på UCloud til dit universitetskursus, som studerende kan bruge f.eks. i øvelses-/labsessioner og/eller derhjemme. En introduktion til tilgangen findes i denne webinaroptagelse.
I denne workshop lærer du at:
Oversætte din kursusstruktur til en struktur, der er kompatibel med en UCloud-kursusapp.
Opsætte kursusudviklingsmiljøet, hvilket indebærer at klone GitHub-repositoriet og køre et kursus-setup-script.
Tilpasse de medfølgende skabeloner for at bygge en UCloud-kursusapp, der inkluderer alle nødvendige komponenter – softwareafhængigheder, scripts, datasæt og mere.
Teste kurset på din egen computer under udviklingen for at sikre, at alt fungerer korrekt. Vi ser også, hvordan den færdige kursusapp ser ud, når den er lagt på UCloud.
Git/GitHub og Docker er essentielle værktøjer i udviklingen af UCloud-kursusapps. I workshoppen giver vi korte introduktioner til begge værktøjer, primært rettet mod deltagere uden forudgående erfaring. Deltagerne kan med fordel gennemgå introduktionsmateriale til Git og Docker på forhånd, men det er ikke et krav.
Dato: 11. juni 2025
Tidspunkt: 12:30–14:30 (CET)
Sted: Online, via Zoom (link følger)
Målgruppe: Forskere og undervisere fra alle afdelinger på alle danske universiteter
00:00 – Introduction and welcome 00:50 – Introduction to UCloud 05:27 – DeiC Interactive HPC website 06:07 – UCloud: Log in 07:06 – UCloud: HPC providers on the UCloud platform 11:08 – UCloud: Initial resource allocations in “My workspace” 12:27 – UCloud: Storage in “My workspace” 13:28 – UCloud: Resources and applications for new resource allocations 13:50 – UCloud: Completing the resource application 21:05 – UCloud: Resource pools and limits (what does it cost?) 23:05 – UCloud: Apps, the app store, applications index in UCloud docs 26:00 – UCloud: Advanced use cases and integration patterns
27:45 – UCloud: Transcriber: Intro and resource needs 30:08 – UCloud: Transcriber: Uploading files inside a project 31:20 – UCloud: Transcriber: Finding and launching Transcriber 32:00 – UCloud: Transcriber: Run Transcriber for the first time (Completing the app launch screen) 34:28 – UCloud: Transcriber: Running multiple Transcriber jobs simultaneously 35:00 – UCloud: Transcriber: Import previous Transcriber job parameters 35:35 – UCloud: Transcriber: Opening running jobs from the “Recent runs” pane 37:15 – UCloud: Transcriber: Transcriber output directories (Jobs folder) 38:26 – UCloud: Transcriber: Transcriber outputs in “Recent runs” 39:54 – UCloud: Transcriber: Output inspection and data download of zip file
41:36 – UCloud: Chat UI: Introduction 43:27 – UCloud: Chat UI: Run Chat UI for the first time (Completing the app launch screen) 48:40 – UCloud: Chat UI: First look at the Chat UI interface (disable new sign-ups and download a model) 52:36 – UCloud: Chat UI: Including documents to support Retrieval Augmented Generation (RAG) (i.e. supplementing the model with an additional document) 57:18 – UCloud: Chat UI: Extend the job time on any UCloud job (if needed) 58:28 – UCloud: Chat UI: Text-to-image generation (stable diffusion with a standard LLM model) 1:01:45 – UCloud: Chat UI: RAG for (best guess) document summarization (beware of model hallucinations)
1:04:45 – UCloud: Label Studio: Introduction 1:05:15 – UCloud: Label Studio: Run Label Studio for the first time (Completing the app launch screen) 1:07:55 – UCloud: Label Studio: First look at the Label Studio interface 1:09:40 – UCloud: Label Studio: Brief view of the Label Studio documentation 1:10:52 – UCloud: Label Studio: Introduction to the coming “Speech Analyser” application 1:15:45 – UCloud: Label Studio: Documentation
Streamline your workflow when using DeiC Interactive HPC
Have you ever experienced delays due to high demand for computing resources on UCloud (DeiC Interactive HPC)? The key to faster access and reducing system strain is efficient resource management.
Over-requesting can lead to unnecessary waiting because DeiC Interactive HPC operates by allocating resources to users as the requested resources become available, without the use of priority queues. Consequently, if you select a large machine for a relatively small task, you will need to wait for other tasks requested for these larger but fewer machines to finish, leading to prolonged waiting.
“Users with little or no experience in assessing the amounts of resources needed have a tendency to request more than they actually need. This causes unnecessary queues and frustration for other users. However, we also see experienced users choosing the large machines simply out of habit instead of starting small, and subsequently migrate to a larger machine only when necessary.”
Jes Elgin, Cloud Engineer at DeiC Interactive HPC
To avoid delays, users need to accurately assess and request only the necessary resources for their task. Starting with conservative estimates can expedite access and improve overall system efficiency.
“Choosing small will always give you a head start as there are more small machines, and you can always upgrade to a large machine if you need it. You don’t have to start over, and you won’t lose any data.”
Jes Elgin, Cloud Engineer at DeiC Interactive HPC
**TIP!** Did you know that the likelihood of obtaining a machine right away is higher the less resources you request?
So, choose your machine based on a qualified estimate of resources needed. If unsure start small and seek advice from experienced colleagues or the DeiC Front Office.
That’s how long it took to set up the new SSH access to DeiC Interactive HPC applications.
If SSH is of interest to you, you probably know that DeiC Interactive HPC applications have recently experienced limitations in providing a reliable and scalable solution for accessing their services using Secure Shell Protocol (SSH).
The challenges were attributed to a limited number of available IP addresses from the common pool on their platform, along with the implementation of a more scalable solution. The latter proved to be burdensome, as acquiring multiple new IP addresses would not provide the scalability required. However, DeiC Interactive HPC has launched a new solution for Secure Shell Protocol access that eliminates the need for multiple IP addresses. This new solution is based on ports which are much more scalable. Therefore, users can now access DeiC Interactive HPC applications using SSH with ease.
