ml platform

◆ ml platform

marin speedrun control panel · Modal A10G

Configure run

Model architecture Training steps Learning rate Batch size Weight decay Steps per eval

Runs a real GPU job on Modal (A10G), trains on FineWeb-Edu, evaluates on Paloma, logs to W&B project marin-speedrun.

Live output

idle

Configure parameters and click "Run speedrun" to start.

Run history