Pipeline Parallelism

1 article about pipeline parallelism.

sPipe: Hybrid GPU and CPU Pipeline for Training LLMs Under Memory Pressure

ยท12 min read

How sPipe combines GPU and CPU resources in a hybrid pipeline to train large language models when GPU memory is limited. Architecture, offloading strategies, and practical benchmarks.

spipellm-traininggpu-memorypipeline-parallelismcpu-offloadingmachine-learning

Browse by Topic

Ai Agents (303)Automation (230)Macos (184)Ai Agent (179)Productivity (172)Claude Code (153)Desktop Agent (118)Developer Tools (85)Reliability (83)Accessibility Api (77)Parallel Agents (75)Mcp (73)Desktop Automation (65)Multi Agent (64)Open Source (56)Ai Coding (56)Security (52)Workflow (51)Architecture (50)Memory (50)