Google Colab support for both fit and evals and CLI improvements

[david-rfai]

Changes

• Re-Support run_fit() on Google Colab
• Support run_evals() on Google Colab
• Add rapidfireai jupyter command, including outputting tunneling recommendations based on if in VSCode or not
• Jupyter default URL is /tree
• Automatic check if running in Google Colab for simpler install/init process
• Add some checks to rapidfireai doctor to give warning or errors if some required items not installed/compatible
• Add additional Python packages to rapidifreai doctor
• Add Torch version and Torch CUDA version to rapidfireai doctor
• Support for evals of newer versions of Torch based on Cuda version
• Add rapidfireai --test-notebooks to copy test notebook to tutorial_notebooks folder
• New FIT test notebook runs under 5 minutes
• Move OPENAI_API_KEY to first line of Notebooks
• Set default Ray console port to 8855, and allow customizing the port RF_RAY_PORT

Testing

• I have added tests that prove my fix is effective or that my feature works
• I have tested this change manually
• I have tested this change in the following environments:
  • AWS Ubuntu, evals
  • AWS Linux , evals
  • Colab, evals
  • PBD, evals
  • AWS Ubuntu, fit
  • AWS Linux, fit
  • Colab, fit
  • PBD, fit

Screenshots (if applicable)

Checklist

• My code follows the project's style guidelines
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published

Performance Impact

If this PR affects performance, describe the impact and any optimizations made.

Related Issues

Fixes #(issue number)
Closes #(issue number)
Related to #(issue number)

---

Note

Adds full Google Colab support for fit and evals with a new rapidfireai jupyter command, enhanced doctor/init, dispatcher/CORS and Ray dashboard config, and updated packaging/notebooks.

• Platform/Env Support (Colab):
  • Add rapidfireai.utils.colab and shared utils.constants; detect Colab and provide auth token helpers.
  • Adjust GPU/CPU allocation and Ray init (dashboard on 8855, RF_RAY_PORT) for Colab.
  • Early CUDA init in actors to avoid CUBLAS_STATUS_NOT_INITIALIZED.
• CLI:
  • New rapidfireai jupyter command (sets /tree, prints port-forward tips).
  • rapidfireai doctor: show site-packages, more packages, Torch/CUDA versions, statuses.
  • init installs env-specific reqs (fit/evals, colab/local); --test-notebooks to copy test notebooks.
• Evals/Dispatcher:
  • Dispatcher binds 0.0.0.0, permissive CORS with credentials in Colab; helpers get_dispatcher_url/get_dispatcher_headers.
  • Notebook UI uses Colab proxy and credentials; polling and clone/IC flows refined.
  • Controller/Scheduler: context build parallelism, CI-aware resource sizing; improved final/live metrics handling.
• Fit:
  • Dispatcher/Frontend import constants from utils; startup scripts handle Colab/TensorBoard mode and PID/port cleanup.
• Packaging/Setup:
  • Expand package data to include utils, setup requirements, tests; add optional deps groups; add setup requirements-*.txt for fit/evals.
• Docs/Notebooks/README:
  • README: add rapidfireai jupyter, consolidated SSH forwarding, extra port-kill tips.
  • New/updated tutorial and test notebooks (Colab-friendly imports, versions, API key prompts).

^{Written by Cursor Bugbot for commit 0456e55. This will update automatically on new commits. Configure here.}