colabcola 0.1.0

Colab Cola packages Python scripts, modules, dependencies, and inputs into reproducible jobs. Submit them to Google Colab, monitor execution, and recover verified results from Google Drive — without building a notebook launcher yourself.

python -m pip install colabcola

Requires Python 3.12+. See Development for source installation and testing.

Quick start

1. Connect your Google account

colabcola login --profile research

Colab Cola guides you through execution-project setup and authorization. New profiles use the bundled production Desktop OAuth client by default; providing your own is optional. First-time Google consent and some Console steps require your participation. Setup guide

2. Package an existing Python workload

colabcola package --profile research \
  --save-colaml experiment.colaml -- python train.py

The packaging guide collects data, dependencies, hardware needs, outputs, and optional checkpoint configuration. It creates a .colajob plus a reusable .colaml recipe, without executing your code.

3. Run it on Colab

colabcola run --profile research \
  --colaml experiment.colaml \
  --run-dir ./runs/experiment-1

Colab Cola checks the job and account, stages inputs, supervises the Colab runtime, publishes results to Drive, verifies recovered files, and releases the runtime it owns. Keep the supervising computer online while the run is active.

To prepare offline, add --prepare-only. To verify launch prerequisites without uploading or allocating compute, use --preflight-only. Repeating the same command with the same run directory reconnects to the existing run rather than starting a new one.

What Colab Cola provides

Python workload + inputs
          │
          ▼
  colabcola package ──► .colajob / .colaml
                               │
                               ▼
                       colabcola run
                               │
                               ▼
                        Google Colab
                               │
                               ▼
                  Google Drive + verified results

Platform support

Requirements: Python 3.12 or newer, a Google account with access to Colab and Drive, and any one-time OAuth/Cloud project setup described in the setup guide. Accelerator availability depends on Google Colab. The Windows recovery acceptance test used CPU runtimes; it does not establish that every GPU or runtime configuration works.

Checkpoint recovery is opt-in: your workload must write restorable state and accept a checkpoint manifest. A runtime lost before publishing its latest checkpoint cannot recover its unsaved in-memory progress. Compute limits are bounded requests, not guaranteed billing ceilings.

Documentation

Development

Colab Cola is distributed as a Python package. Release checks cover clean wheel/source installs, complete regression tests, platform behavior, license notices, and the verified runtime lifecycle. See Development and Packaging for current details.

Colab Cola is an independent project and is not affiliated with or endorsed by Google. Google Colab and Google Drive remain subject to their own availability, policies, and limits.