torchway 0.0.2

CLI tool for distributed PyTorch training on Kubernetes

Torchway

A CLI tool that takes any PyTorch training script and handles the entire distributed training stack in one command — no YAML, no boilerplate.

pip install torchway

What it does

• Generates and applies Kubernetes PyTorchJob manifests automatically
• Injects DDP/FSDP distributed training boilerplate
• Aggregates multi-pod logs in real time
• Auto-wires MLflow experiment tracking
• Surfaces plain-language fault explanations for OOM kills and NCCL errors

Status

🚧 Under active development — V1 coming soon.

Author

Suvan Kasina — GitHub