Systems and efficient execution · MLD-0892
Pipeline parallelism
Assigning different layers or stages to different devices and moving intermediate activations between them.
Original reference page · detailed expansion pending. Original definitions, examples, notes and citations are preserved below. New examples are illustrative, not benchmark results.
Original example
Train a deep model in staged micro-batches.
Original caveat
Pipeline fill and drain periods can leave devices idle.
Context and aliases
Level: Advanced. Stages: Training · Inference · Deployment.
Tags: DL · LLM · SLM · Systems.
Aliases: PP.
Related concepts
Concepts linking here
Original references
- PyTorch Fully Sharded Data Parallel ↗ · Consulted
- PyTorch documentation ↗ · Reference link
- vLLM documentation ↗ · Consulted
- Original reference URL ↗