User Guide

• End-to-End Use Case: TransformerEngine-FL + Megatron-LM-FL + FlagScale
  • 1. Docker Environment
  • 2. Prepare FlagScale
  • 3. Prepare Megatron-LM-FL
  • 4. Prepare TransformerEngine-FL
  • 5. Prepare Dataset and Tokenizer
  • 6. Qwen3 Training Test
  • 7. DeepSeek-V3 Training Test
  • 8. Notes
• TE-FL Custom Backend Examples
  • Two Approaches
  • Quick Start
  • In-tree Approach (3 Steps)
  • Out-of-tree Approach (Plugin Package)
  • Environment Variables
  • Examples
  • Expected Output