ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix: set __spec__ on mod stubs so importlib.util.find_spec doesn't raise

Browse source 
Manually-injected sys.modules entries have __spec__=None by default.
importlib.util.find_spec() raises ValueError when it finds a module in
sys.modules with __spec__=None (transformers.utils.import_utils hits this
when checking if torchao is available). Give every stub a minimal
ModuleSpec(name, loader=None, is_package=True) to satisfy find_spec.
This commit is contained in:
LeoBorcherding 2026-05-11 06:16:05 -05:00
commit 3e57133c84
1 changed files with 7 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

7
studio/backend/core/training/worker.py
View file

@ -1104,10 +1104,17 @@ def run_training_process(
# any attempt to import a submodule (e.g. "torch.distributed.tensor._foo")
# raises "is not a package" because Python checks __path__ before looking
# in sys.modules for the child.
import importlib.machinery as _ilm
def _make_mod_stub(mod_name):
m = _types.ModuleType(mod_name)
m.__path__ = [] # marks this as a package to the import system
m.__package__ = mod_name
# importlib.util.find_spec() raises ValueError when __spec__ is None
# on a module that's already in sys.modules (our manually-injected
# stubs). Give every stub a minimal ModuleSpec so find_spec() returns
# a spec object (non-None) instead of raising.
m.__spec__ = _ilm.ModuleSpec(mod_name, loader=None, is_package=True)
def _ga(attr, _m=m, _n=mod_name):
if attr.startswith("__"):
raise AttributeError(attr)