unsloth/scripts/data
Daniel Han 6f8df52cbc
Acknowledge the Colab oracle drift, and stop the diff hiding entries (#11324)
* Acknowledge the Colab oracle drift, and stop the diff hiding entries

Notebooks CI has been red since at least 2026-09-16 on

    A rule-bearing Colab oracle drifted from its committed snapshot; run
    `notebook_validator.py refresh-colab --all --snapshot-dir scripts/data`
    to acknowledge.

That is the tripwire working: any drift in pip-freeze.gpu.txt is strict by
construction, because the rules resolve a notebook's installs against it. What
it wants is an acknowledged refresh, so this is one, with the diff read rather
than waved through.

What moved: 91 pip entries (3 new, 1 removed, 87 changed), notably
transformers 5.15.0 -> 5.16.1 and datasets 4.0.0 -> 4.8.5; the base image from
Ubuntu 22.04.5 to 24.04.1; 1261 apt entries. torch did not move, and neither
did the Python line, which is why only the pip oracle tripped --strict.

The evidence that acknowledging is safe: linting all 445 notebooks in
unslothai/notebooks against the old and the new oracle gives the same 82
findings, 0 new and 0 resolved. This is upstream churn no rule resolves
differently, not a compatibility change being papered over.

Robustification, from a defect this drift exposed. colab-diff caps its lists at
50/50/80 and elides the rest with a bare count, and transformers 5.15.0 ->
5.16.1 sat past the CHANGED cap -- reading the output, that package looked
untouched. Every entry in the pip oracle is rule-bearing, so an elided one is
exactly what a reviewer may be deciding on. Added --full, and the elision now
names it instead of leaving it to be guessed at. Two tests cover both halves;
removing the flag's effect fails both. tests/notebooks: 314 passed, 2 skipped.

Not addressed here: the nine "smoke install (Colab-shaped venv, opt-in)" jobs
in the same workflow also failed. Their logs have expired, so I have not
diagnosed them and this change does not claim to fix them.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
2026-09-19 00:04:01 -07:00
..
colab_apt_list.gpu.txt Acknowledge the Colab oracle drift, and stop the diff hiding entries (#11324) 2026-09-19 00:04:01 -07:00
colab_os_info.gpu.txt Acknowledge the Colab oracle drift, and stop the diff hiding entries (#11324) 2026-09-19 00:04:01 -07:00
colab_pip_freeze.gpu.txt Acknowledge the Colab oracle drift, and stop the diff hiding entries (#11324) 2026-09-19 00:04:01 -07:00
colab_to_cpu_pin.json Trim 2.59 GiB per generation from the Colab smoke job's pip cache (#11270) 2026-09-18 07:31:58 -07:00