Introduction to Cloud Service
Microsoft Azure
LLM APIs Available:
- Azure OpenAI Service: Access to the latest GPT-4.1, GPT-4.1-mini/nano, o3-mini, GPT-4o, and more, including multimodal models for text, image, and audio processing.
- Embeddings and Vision Models: Includes text embeddings (e.g., text-embedding-3-small/large), DALL-E for image generation, and support for video and audio models.
Other Services:
- Azure Cognitive Services: APIs for vision (object/text recognition), speech (speech-to-text, text-to-speech), language (translation, summarization), and document intelligence (text/table extraction).
- Azure Machine Learning: Full lifecycle platform for training, deploying, and managing ML models at scale.
- Azure AI Bot Service: Tools for building conversational AI experiences.
- Virtual Machines (VMs): Azure supports Windows Server 2025 and a wide variety of VM types for general, GPU, and high-performance workloads, with deep integration for hybrid and edge computing.
Google Cloud Platform
LLM APIs Available:
- Vertex AI Model Garden: Access Google’s Gemini 2.0, Gemini 1.5, Anthropic Claude 3.5 Sonnet, Meta Llama, Mistral Mixtral 8x7B, and AI21 Labs Jamba 1.5.
- Gemini API: Supports text, image, audio, and code generation, with up to 2 million token context windows and advanced multimodality.
- Model Customization: Vertex AI Model Builder allows for testing, fine-tuning, deployment, and monitoring of both Google and third-party models.
Other Services:
- Vision AI API: Image search and analysis.
- Virtual Machines: Next-gen VMs (C4, C4D, H4D, M4, Z3) for general, high-performance, and specialized workloads, with significant performance improvements and support for hybrid and multi-cloud deployments.
- Kubernetes Engine (GKE): Enhanced for large AI/HPC clusters, with AI-aware load balancing and serverless options.
- Storage and Networking: Rapid storage, zonal SSD caches, and global high-speed networking for low-latency AI workloads.
Alibaba Cloud
Alibaba Cloud
Alibaba Cloud has rapidly expanded its AI offerings, with a focus on proprietary and open-source LLMs and developer-friendly infrastructure.
LLM APIs Available:
- Qwen Series: The latest Qwen3 and Qwen2.5 (up to 72B parameters) models are available globally via Model Studio APIs, supporting hybrid reasoning for both complex and general-purpose tasks.
Other Services:
- Platform for AI (PAI): Distributed inference, multi-node architecture, and workflow/agent tools for advanced AI application development.
- Virtual Machines: 9th Generation Enterprise Elastic Compute Service (ECS).
Amazon Web Services (AWS)
LLM APIs Available:
- Amazon Bedrock: Provides access to a wide range of foundation models, including Amazon Titan, Anthropic Claude 3.5, Meta Llama, Mistral Mixtral 8x7B, and more. Perplexity is also available through the AWS Marketplace.
- Multi-LLM Routing: Supports dynamic routing between different LLMs for cost and performance optimization in applications.
Other Services:
- Virtual Machines: EC2 supports Windows Server 2025, Hyper-V virtualization, and a wide range of VM types for general, GPU, and high-performance workloads.
- Machine Learning: Amazon SageMaker for building, training, and deploying ML and foundation models at scale.
- Serverless Computing: AWS Lambda for API and AI workload deployment without server management.
- AI APIs: Vision, speech, language, and document processing APIs for a wide range of business and academic applications.
Application Process
For Innovation Academy Funding Scheme projects, please contact Innovation Academy (innoacademy@hku.hk) for applications.
Student Interest Groups (SIG) can submit applications to use Generative AI APIs, including GPT-4o-mini, o3-mini, DeepSeek-R1, and DeepSeek-V3. Each SIG can request one API key. Applications can be made here (https://forms.office.com/r/Hmfab9YmuU). Please contact Innovation Wing (innowing@hku.hk) for enquiries.
Waterjet CNC cutting machine