Industrial Strength DALLE-E:Scaling Complex Large Text & Image Models
summary
Demo shows a reproducible pipeline that productionises a DALL-E image model hooked to GPT2, with runnable code, Docker files and a Kubernetes manifest. It uses DALL-E powered applications fed by GPT2 outputs, served by MLServer + HuggingFace runtime with a custom Python runtime for the DALL-E mini, environment pinned with poetry, and orchestration via seldom core for Kubernetes. Models load in one endpoint, generate image tensors deterministically with a seed, and the local demo works end to end.
AI-written summary · about
chapters (8)
- 0:00Industrial DALL·E scaling
- 3:01Why large X models matter
- 7:16Production challenges
- 10:07Reproducibility in serving
- 13:59Custom DALL·E runtimes
- 15:44Orchestrating with Seldon Core
- 18:11Demo: GPT-2 & DALL·E
- 26:42Resources & closing
related talks
34mDeep Learning at Scale
36mAccelerating High-Performance Machine Learning at Scale i...
27mDeveloping and Deploying Models at Scale — without Spark!
42mDeep learning beyond the learning
36mAccelerating Machine Learning at Scale with
13mMLOps Pipelines to Deliver Tf.Lite Models to Distributed Compute Pools at the Edge
26mOne Pipeline to Rule Them All: Unifying Multimodal and AI Data Processing with Daft
30mProductionizing an unsupervised machine learning model to...