Serverless AI
Service by Nebius
Run containerized AI workloads without infrastructure overhead
A fully managed serverless platform for running containerized AI workloads with per-second billing. It enables developers to deploy interactive inference endpoints and background training jobs without managing underlying GPU infrastructure.
- Languages
- English
Category: Serverless
- Billing Granularity
- Per-second
- Workload Types
- Interactive Endpoints, Background Jobs
- Management Level
- Fully Managed (Serverless)
- Supported Workloads
- Inference, Fine-tuning, Training, Speech Synthesis
- Per-second billing for active compute only
- Supports both interactive endpoints and background jobs
- Zero infrastructure management or cluster setup required
- Built-in support for LLM fine-tuning and speech synthesis
Overview
Nebius Serverless AI is a cloud service designed for running containerized AI workloads without the need for manual infrastructure management. It serves AI developers and enterprises by providing a platform for both interactive inference endpoints and non-interactive background jobs, such as model training or data processing. The service eliminates the overhead of cluster management and the cost of idle GPUs through a per-second billing model where resources stop automatically after a task is finished. It is specifically engineered to support complex AI workflows including large language model fine-tuning with Axolotl and speech synthesis. By providing a "Serverless Cookbook" of runnable examples, it enables teams to move from experimentation to production deployment rapidly.
Compute Modes
- Interactive Endpoints: Deploy interactive endpoints that listen for requests and return results immediately for real-time applications.
- Background Jobs: Execute non-interactive workloads in the background that stop automatically upon task completion.
- No Infrastructure Management: Eliminate cluster management and infrastructure setup by running containerized workloads directly.
- Zero Idle Costs: Avoid costs for idle GPUs by paying only for the compute used during active processing.
AI Workflows and Tools
- LLM Fine-tuning: Fine-tune large language models using serverless jobs with tools like Axolotl.
- Speech Synthesis: Convert text to speech by creating models, running fine-tuning jobs, and deploying endpoints.
- Model Inference: Deploy and chat with open-source large language models through serverless endpoints.
- Serverless Cookbook: Access a GitHub repository of runnable examples for various AI workloads.
Use cases
- Developers deploy open-source large language models as interactive endpoints to power real-time chat applications.
- Data scientists run non-interactive background jobs to fine-tune foundation models using Axolotl without managing clusters.
- Engineers build speech synthesis pipelines by running fine-tuning jobs and deploying inference endpoints for text-to-speech conversion.
- Teams use the Serverless Cookbook examples to quickly prototype and scale specialized AI functions with zero idle GPU costs.
Specifications
- Billing Granularity
- Per-second
- Workload Types
- Interactive Endpoints, Background Jobs
- Management Level
- Fully Managed (Serverless)
- Supported Workloads
- Inference, Fine-tuning, Training, Speech Synthesis
How it works
- Choose Workload Type: Users select between an interactive endpoint for real-time results or a background job for long-running tasks.
- Configure Container: Prepare the containerized AI workload or use provided tutorials for models like open-source LLMs.
- Deploy and Scale: Deploy the workload via the Nebius web console or CLI to start processing requests or background tasks.
Pricing
- Serverless AI Compute
- Per-second billing based on usage
FAQ
What types of workloads can I run on Serverless AI?
Serverless AI supports two main types: interactive endpoints for immediate results and background jobs for non-interactive tasks that stop automatically.
How is Serverless AI billed?
The service uses per-second billing, ensuring you only pay for the exact compute time used without costs for idle infrastructure.
Can I fine-tune models using this service?
Yes, you can fine-tune large language models by using a Serverless AI job with tools like Axolotl.
Are there any templates or examples available?
Nebius provides a 'Serverless Cookbook' on GitHub with runnable examples for various AI workloads to help developers get started.
Added by AntSpace from public sources (docs.nebius.com).