torch-pipeline-parallelism
OfficialScale training with PyTorch pipeline parallelism.
Authorletta-ai
Version1.0.0
Installs0
System Documentation
What problem does it solve?
This Skill provides guidance for implementing PyTorch pipeline parallelism for distributed training of large language models. It covers model partitioning, inter-rank communication, gradient flow management, and common pitfalls.
Core Features & Use Cases
- Model partitioning: Split transformer layers across ranks.
- Inter-rank communication: Send/recv and activation caching.
- Gradient flow management: AFAB scheduling, loss scaling, and backward synchronization.
Quick Start
Implement a 3-stage pipeline across 4 GPUs and validate gradient flow.
Dependency Matrix
Required Modules
None requiredComponents
Standard package💻 Claude Code Installation
Recommended: Let Claude install automatically. Simply copy and paste the text below to Claude Code.
Please help me install this Skill: Name: torch-pipeline-parallelism Download link: https://github.com/letta-ai/skills/archive/main.zip#torch-pipeline-parallelism Please download this .zip file, extract it, and install it in the .claude/skills/ directory.