High-performance computing (HPC) infrastructure and support from our specialists accelerate the development of AI-based projects and processes.
The HPC environment offers the highest scalability efficiency when training models on massive datasets. It also requires expertise in responding to potential issues, profiling performance, and identifying areas for improvement. We provide all of this in one place.
What sets us apart is our own physical, secure data processing infrastructure and our specialization in optimizing its use.
Our AI resources
HPC clusters (supercomputers):
- Helios: 440 GPU NVIDIA GH200
- Athena: 384 GPU NVIDIA A100 40GB
Virtual platforms and tools:
- The LLM Lab platform is available to PLGrid infrastructure users as a dedicated service, enabling API-based access to large language models tailored to diverse needs such as data analysis, modeling, and automation of research processes.
More information: LLM Lab platform
- Cluster environments (a complete set of software and drivers enabling immediate and efficient operation)
Developed as part of the Meetween project:
- SPEECHM - a platform for testing AI models within challenges
More information: https://speechm.cloud.cyfronet.pl/
- MLFlow - implementation of a platform for monitoring model training progress
Our expertise
- Management of AI computing environments (libraries, their dependencies, environment migration).
- Proficiency in leading AI frameworks and distributed model training processes.
- Use of tools for profiling AI tasks in terms of performance and analysis of the results.
- Optimization of data processing pipelines for AI tasks (efficient loading, optimal use of available file systems).
- Scalable serving of trained AI models.
Based on our resources and expertise, we offer comprehensive services. At every stage of implementation, clients receive professional support from our experts - practitioners specializing in the use of HPC for AI development.
Our AI services
- Business problem analysis and selection of the appropriate approach (choice of machine learning technique).
- Implementation requirements analysis (what data is needed, how much, how to process it, estimated computational resource needs).
- Creation and maintenance of an environment for developing such solutions.
- Support in designing the training process.
- Monitoring and profiling of training runs, optimization.
- Designing the model deployment process in a production environment.
- Consulting, training, and advisory services for specific challenges.
Contact us: helpdesk (at) plgrid.pl
Polish LLMs
Cyfronet provides computing resources and actively contributes to the development of two Polish Large Language Models (LLMs): Bielik and PLLuM (HIVE AI). We assist in preparing and launching large-scale optimized training runs, monitor the training process, and ensure its efficiency. We also maintain the environments and indirectly support the model serving process.
More information: LLM