faultline-sdk 0.24.0

ML training checkpoints and recovery for Faultline Cloud (import faultline)

Faultline Python SDK

Client library for Faultline Cloud — stream training metrics and checkpoints from your laptop, GPU VM, Slurm job, or Colab notebook.

pip install faultline-sdk

import os
import faultline

run = faultline.start(
    "my-run",
    project="demo",
    api_key=os.environ["FAULTLINE_API_KEY"],
    base_url=os.environ["FAULTLINE_API_URL"],
)

for step in range(100):
    run.log(train_loss=0.5, step=step)
    if step % 10 == 0:
        run.save(step=step, model=model, optimizer=optimizer)

run.complete()

Setup

1. Sign up on your Faultline dashboard and create an API key (Account page).
2. Set environment variables on the machine that runs training:

export FAULTLINE_API_KEY=fl_...
export FAULTLINE_API_URL=https://your-api.example.com

Training runs locally; Faultline stores metrics and checkpoints remotely.

Optional extras

pip install "faultline-sdk[grpc]"    # local Rust runtime over gRPC
pip install "faultline-sdk[psutil]"  # host/GPU telemetry helpers

Note on package name

The PyPI project is faultline-sdk because the name faultline is already registered by another package. Import in Python is still:

import faultline