Getting Started

MVP One — frictionless distributed LoRA

From zero to running jobs in minutes. No kubernetes wrangling, no custom infra.

1) Initialize (guided)

Charter setups up your salad account and creates a new project with access to all HuggingFace models

python scripts/mvp.py init

2) Deploy

One click creation of a network of workers with a centralized coordinator (Intelligently distributed without constant oversight)

python scripts/mvp.py job deploy

3) Monitor & Receive Weights

Check lossed and training metrics and progess. Immediatly receive model training weights on completion.

python scripts/mvp.py monitor

What you get

  • Async aggregator service with exactly-once progress
  • Stateless worker launcher with budget guard
  • Simple cost tracking and live job monitoring
Full README and source on GitHub: github.com/WencySuo/mvp_one