ServerlessT2I: Efficient Text-to-Image Workflow Serving on a Serverless Platform
ServerlessT2I is an innovative framework that breaks down text-to-image (T2I) processes into loosely connected model functions. This design facilitates scaling for each model, allows for declarative workflow assembly, ensures transparent communication among GPU resources, and incorporates fairness-aware scheduling. It tackles the drawbacks of traditional monolithic setups on serverless platforms, where workflows are often deployed as opaque GPU functions, leading to structural obscurity, increased scaling costs, and restricted fairness. Tailored for serverless-native environments, the system permits the independent management and scheduling of models. The research paper can be found on arXiv with the ID 2607.26566.
Key facts
- ServerlessT2I decomposes T2I workflows into loosely coupled model functions.
- It enables per-model scaling and fairness-aware scheduling.
- The system addresses monolithic deployment inefficiencies.
- It allows transparent GPU-resident communication.
- The paper is published on arXiv with ID 2607.26566.
Entities
Institutions
- arXiv