Overview
Alibaba Cloud’s full-stack solution for generative AI (GenAI) provides whole-process services for foundation models (FMs) and other AI development tasks. This solution helps you build and optimize FMs, fine-tune them according to your business preferences, and deploy them easily as online services, all on purpose-built AI infrastructure optimized for performance and efficiency. Regardless of the scale and stage of your business, this solution enables you to create new and intelligent customer experiences and drive business transformation with innovations in generative AI.
Disclaimer
Some images displayed in this page are solely for the purpose of demonstrating product functions. All information shown in these images has been generated by AI models and is not verified by Alibaba Cloud. Alibaba Cloud makes no warranties, expressed or implied, as to the authenticity, accuracy and completeness of all such information.
AI Infrastructure Built for Performance
You can perform compute-intensive AI tasks such as FM training and customization on our performance-optimized AI infrastructure, or choose our Cloud-Native AI suite to improve the speed and efficiency of your AI workloads on Kubernetes.
• Unified Hardware and Software Acceleration for AI
1. GPUs for Model Training and Inference
Model Training: gn7 series of ECS instances power large-scale training tasks with high-performance GPUs.
Model Inference: gn6 series of ECS instances provide a cost-effective choice for model inference tasks
2. AI Acceleration
You can leverage GPU Accelerator AIACC to speed up AI training tasks by up to 70% and inference tasks by 2-3 times according to Stanford DAWN Deep Learning Benchmark.
• Platform for AI (PAI): Whole-Process AI Engineering
PAI provides an end-to-end optimization solution to streamline AI engineering including PAI-iTAG for data labeling, PAI-DSW for model building, PAI-DLC for model training, and PAI-EAS for model inference and deployment.
PAI-Lingjun Intelligent Computing Service (Currently available only in China (Ulanqab) and Singapore) provides high-intensity heterogeneous computing services and AI capabilities for large-scale generative AI tasks based on the integrated optimization technology of software and hardware. It also leverages pre-packaged GPUs and Platform for AI (PAI) to provide full-link performance acceleration for AI training and inference tasks in one stop.
Data Preparation
Ready your data for model training with intelligent, customizable, and highly efficient multimodal data labeling services
Model Development
Build foundation models with our one-stop visualized modeling tool - PAI-Designer, or perform interactive development with Notebook in PAI-DSW
Model Training
Train models with PAI-DLC, our one-stop platform for cloud-native deep learning and training compatible with predefined and customized algorithm frameworks
Model Deployment
Deploy your model as an online service or a web app with PAI-EAS, which supports push-button deployment of large-scale complex models
A Wide Selection of Open-Source FMs
You can train and fine-tune popular FMs including Alibaba Cloud’s Qwen series in Platform for AI (PAI) in a few clicks, and easily deploy them as online services with PAI-EAS.
Tongyi Qianwen (Qwen)
Alibaba Cloud provides a series of open-source Tongyi Qianwen models: Qwen-Max and Qwen-Plus, the flagship LLMs; Qwen-Omni, Qwen-VL, and Qwen-Audio, the multimodal models; QVQ and QwQ, the reasoning models; Qwen-Coder and Qwen-Math, the vertical models, and more. Qwen models are pre-trained on multilingual data covering various industries and domains, and offer a wide range of capabilities, including multimodal understanding and generation, state-of-the-art image processing, and fully managed APIs to support your innovation in generative AI. The latest Qwen3 adopts unique “dual thinking modes” that let you flexibly switch between “Thinking” (for deep reasoning) and “Non-thinking” (for fast responses) modes to balance inference depth and speed while saving costs. It’s lightweight yet powerful, outperforming peers in accuracy and versatility among same-sized models, and breaks language barriers with multilingual capabilities in 119 languages and dialects.
You can easily fine-tune Qwen models with your enterprise data and deploy them as online services that understand your business.
Wan2.1
Wan2.1 is an open-source, versatile video foundation model suite for text-to-video and image-to-video generation. It excels at generating realistic visuals by accurately handling complex movements, enhancing pixel quality, adhering to physical principles, and optimizing the precision of instruction execution.
DeepSeek
DeepSeek-V3 is a high-performence LLM featuring a mixture of experts (MoE) architecture. DeepSeek-R1 is trained based on DeepSeek-V3-Base. The Model Gallery of PAI provides accelerated deployment options, such as BladeLLM, SGLang, and vLLM, so you can deploy DeepSeek models with one click.
Llama 3
LLaMA 3 is a powerful open-source LLM with a large set of training data. It focuses on innovation, scalability, and simplicity with several architectural improvements over its predecessor, LLaMA 2. You can access, fine-tune, and deploy LLaMA 3 with Platform for AI (PAI) in a few simple steps.
GenAI Development on Alibaba Cloud Model Studio
Accelerate model customization and GenAI application development on a one-stop platform that hosts industry-leading AI models such as Qwen-Max
Learn More About Model Studio
Built-In Model Inference and Evaluation Workflows
Speed up model development workflows with comprehensive tools designed to support SFT and LoRA, built-in model compression and inference acceleration, multi-dimensional model evaluation in visualized templates, and one-click model deployment
One-Click RAG Setup with AnalyticDB
Model Studio jointly applies in-depth retrieval optimization with AnalyticDB, which provides retrieval capability of over 10 billion vectors and is compatible with a variety of Alibaba Cloud AI products. Learn more about RAG >
Comprehensive Security Measures
Secure your enterprise data in storage and transmission by completing model and app development in your dedicated Virtual Private Cloud (VPC) network and accessing data with PrivateLink, apply customizable content governance to prompts and content, and combine responsible AI principles with tools for human accountability
Innovative Applications Based on Generative AI
Alibaba Cloud provides GenAI applications to boost productivity, enable innovative content creation, or fulfill your other business needs based on state-of-the-art GenAI capabilities such as Qwen models.
Smart Studio
Smart Studio is an all-in-one platform powered by GenAI that enables diverse content creation in one workspace. Its out-of-the-box features and pre-defined workflows are tailored for various industry scenarios, allowing you to create stunning image and video content effortlessly.
AI Doc
AI Doc revolutionizes digital document management for enterprises with the power of LLMs, including Qwen models. It efficiently parses various documents, accurately extracts information based on business requirements, and swiftly generates tailored documents.
Automatic Speech Recognition
This solution offers high-precision capabilities to convert speech from audio and video files to text in complex environments and identify English, Mandarin, and Cantonese speech across multilingual contexts, with more languages to come.
Lingma
Alibaba Cloud's Lingma is a coding copilot powered by Qwen models. It provides features such as intelligent code generation, AI Chat for developers, multi-file code modification, and automatic code execution capabilities.
