Speeding up Your ML Workload: Pytorch Compile and Distributed Training
summary
Explains how torch.compile turns eager PyTorch into graph mode using torchdynamo, AOT Autograd and torch inductor and what each stage actually emits and optimizes. Shows numbers: MobileNet on ImageNet drops latency ~32% and raises throughput ~48%, A100 runs ~76% faster in official benchmarks, CPU wins up to ~1.5x though some models see many graph breaks (up to 44) yet still small gains. Describes a mini‑PyTorch hands‑on that implements data/tensor/pipeline parallelism for multi‑GPU distributed training.
AI-written summary · about
related talks
23mBlazing Fast GenAI Inference With Torch.compile
24mAdvancements in High-Performance AI/ML through PyTorch's Python Compiler
24mTraining large scale models using PyTorch
38mMPI Meets Machine Learning: Unlocking PyTorch distributed for scaling AI workloads
08mPyTorch Distributed: Towards Large Scale Training
32mTaming the Machine: Basics of ML Models Training and Inference Optimization
15mPyTorch 2.0: Unlocking the Power of Deep Learning with the Torch Compile API
26mThunder: Distribute and Optimize Your PyTorch Models With Zero...