Choose parallelism and size your training GPUs.
What it does
Choose parallelism configurations for Megatron Bridge training, with guidance on GPU sizing, hardware topology, and configuration pitfalls.
When to use it
Use when choosing or sizing parallelism degrees, or investigating an out-of-memory error or regression after a parallelism configuration change.
How to use it
Give it a parallelism sizing or debugging request; it returns configuration instructions and shell commands.
What you need · 4
The combined-parallelism verification command runs Python; no runtime version is specified.
The combined-parallelism verification command uses uv to launch Python.
The combined-parallelism verification command requires the torch.distributed.run module.
Running the supplied verification requires four CUDA-visible GPUs and uses four ranks.
Access · 0
No additional actions listed.