Connecting the world and beyond

Geneva Digital Week

Foundation Model Platforms & Engineering Capabilities

​​​2.0.jpgThis category covers the full-lifecycle software platform foundation and core engineering technical specifications supporting AI and foundation models from data preparation, algorithm development, model training, fine-tuning, and deployment, to monitoring and operations. These standards aim to improve resource utilization and platform collaboration efficiency in cloud and edge computing environments.

2.1 FM Platforms & Cluster Management​

Specifically regulates heterogeneous computing cluster systems, model orchestration frameworks, and observability platforms required for foundation model training and inference, ensuring efficient and controllable platform operations.​

Standardizes the overall framework of foundation model platforms, covering data engineering, distributed pre-training, fine-tuning optimization, model delivery, and service operations management modules.
2.1.png
F.748.43 Framework of the foundation model platform     

Specifies comprehensive requirements for cluster systems used in foundation model training and inference regarding general/accelerated computing hardware units, network communication, and resource management.

Standardizes the technical architecture of observability platforms for foundation model clusters, enabling real-time monitoring and analysis of internal system states and telemetry data.

Regulates the framework system for the joint orchestration and collaboration between foundation model platform services and external plug-ins or dedicated models.

Specifies the software platform architecture and full lifecycle management requirements supporting time-series foundation models for pre-training, fine-tuning, and downstream task forecasting.

2.2 General AI Cloud Platform & Lifecycle​

Covers the basic technical architecture of traditional and general AI cloud platforms, along with full-link standards for resource management, data processing, intelligent annotation, inference optimization, and performance testing benchmarks.​

Defines the overall reference architecture of AI cloud platforms and basic specifications for management platform task logs and data analysis.

Standardizes the functional support requirements of AI cloud platforms in the stages of algorithm development, data annotation, and model training.

Standardizes the core modules and workflows for AI model deployment frameworks, format conversion, and online testing/monitoring on cloud platforms.
2.2.png
F.748.48​​ Framework of the foundation model platform     ​

Specifies the requirements for data import, manual annotation tools (e.g., polygons, b​ounding boxes), and intelligent assisted annotation methods in the AI cloud platform data annotation process.

Formulates the overall architecture and normative technical requirements for data processing on AI cloud platforms.

Standardizes the requirement system for computing resource allocation, monitoring, and scheduling within AI platforms.

Standardizes the metric requirements for high availability and disaster recovery monitoring of large-scale AI cloud platforms.

Provides general optimization architecture technical support for model inference, including parallel acceleration strategies at the software and hardware levels.

Standardizes the system functional requirements for Automated Machine Learning (AutoML) in hyperparameter optimization and feature engineering automation.

Provides an evaluation benchmark framework for AI cloud platform computing resource node configuration descriptions and model execution performance testing.

2.3.jpg2.3 Model Frameworks & Core Optim​ization Tech 
Focuses on underlying optimization technologies that enhance model execution and distribution efficiency, including deep learning framework evaluation, edge-side lightweighting, model compression representation, operator specifications, and foundation model knowledge distillation requirements.

ITU-T F.748.12: Deep learning software framework evaluation methodology (2021)
Provides an evaluation system and testing scenario references for deep learning software frameworks (training and inference) in terms of ecosystem construction, ease of use, performance, and hardware support.

ITU-T F.748.36: Requirements and framework of multi-algorithm scheduling systems (2024)
Standardizes the system requirements for uniformly scheduling and distributing multiple AI algorithm models across heterogeneous computing clusters.

ITU-T F.748.49: Architecture and protocols of multi-algorithm scheduling systems (2025)
Defines the internal logical structure of the Multi-Algorithm Scheduling (MAS) system and its underlying HTTP-based algorithm package management API protocol specifications.

ITU-T F.748.68 (ex F.EDS): Requirements of edge domain inference systems for foundation models (2025)
Specifies the software/hardware system framework and fine-tuning capability requirements for deploying compressed foundation models in edge domain environments for inference.

ITU-T F.748.69 (ex F.FMD): General technical framework and requirements of foundation models for mobile devices (2025)
Formulates the system adaptation, capability optimization, and service testing requirements for foundation models deployed on mobile devices such as smartphones, tablets, and robots.


ITU-T F.748.53: Representation and compression methods of artificial intelligence models (2025)​
Provides standard serialized representation and distribution encapsulation for model structural expression and network compression operators like quantization, pruning, and structured matrices.
2.3.png
F.748.53 The structure schema of representation and compression methods of artificial intelligence models     ​

ITU-T F.LLMO: Requirements and evaluation metrics for large language model operators (2025)
Proposes adaptation development specifications and evaluation metrics for core basic mathematical operators and neural network fusion operators (e.g., attention mechanism operators) executed in large language models.

ITU-T F.RF-MDC: Requirements and framework of model distillation capabilities (2025)
Standardizes the construction of model distillation capabilities, providing standardized guidance for transferring knowledge from large "teacher models" to lightweight "student models."

ITU-T F.AICP-MoE: General technical requirements and framework for artificial intelligence cloud platform - Mixture of Experts (MoE) capability (2025)
Specifies the technical requirements for AI cloud platforms to support parameterized configuration, distributed pre-training, and fine-tuning capabilities of Mixture of Experts (MoE) model structures.

ITU-T TR.WM: Technical framework and essential capabilities of World Models (2025)​
Technical report. Explores the technical framework foundations and key capability scenarios of World Models with the ability to understand physical laws and causal relationships in multimedia tasks.​

​




© ITU All Rights Reserved

Back to top