ML Training Performance Profiling
Profile ML training to identify bottlenecks in data loading, model computation, and memory usage.
Quick Start
# Simple profiling (built-in)
python src/train.py trainer.profiler=simple trainer.max_epochs=1
# Detailed profiling (PyTorch Profiler)
python src/train.py trainer.profiler=pytorch trainer.max_steps=100
# View in TensorBoard
tensorboard --logdir=./tb_logs/profiler
# Monitor GPU utilization
watch -n 1 nvidia-smi
Profiling Tools
[Description truncada. Veja o README completo no GitHub.]