MosaicML Composer

Training efficiency library with algorithmic speedups and multi-GPU orchestration

Overview

MosaicML Composer is a PyTorch training library that provides algorithmic speedups, efficient multi-GPU training, and composable training recipes for faster model convergence.

Key Features

• Algorithmic speedups — Mix of training efficiency methods (e.g., progressive resizing, label smoothing)
• Composable recipes — Stack multiple methods without conflicts
• Multi-GPU — Native distributed training support
• Efficient data loading — Streaming datasets from object storage