Fine-tune LLMs from one YAML. Layer streaming trains an 8B model on a 4 GB laptop GPU.
Python cliconsumer-gpudpofine-tuningggufhuggingfacellmllmopslocal-ailocal-llmloralow-vrammachine-learningollamapeftpythonpytorchqlorasfttransformers
"""The recipe count, written down once (#1016).
Adding a recipe changes ``len(RECIPES)``. The milestone assertions that pin the
Python dictionary compare against ``EXPECTED_RECIPE_COUNT`` here instead of
carrying their own literal, so that is the one number to edit in ``tests/``.
The documentation sites that state the count are listed, and checked against
the catalog, in ``tests/test_recipe_count_is_synced.py``.
"""
from __future__ import annotations
EXPECTED_RECIPE_COUNT = 171
def recipe_count_hint(actual: int) -> str:
"""The failure message a contributor adding a recipe sees."""
return (
f"RECIPES has {actual} entries but tests/recipe_count.py pins "
f"EXPECTED_RECIPE_COUNT = {EXPECTED_RECIPE_COUNT}. If you added or removed a "
"recipe on purpose, set EXPECTED_RECIPE_COUNT to the new count, then run "
"python scripts/sync_recipe_count.py to update the documentation sites "
"that state it (tests/test_recipe_count_is_synced.py lists them, and names "
"each file and line that still holds the old number)."
)