unsloth/studio/backend/core/training
LeoBorcherding 4d09cbbda4 fix(rocm/win): stub class metaclass for ProcessGroup.BackendType; amd-smi circuit breaker
torchao.float8.inference accesses ProcessGroup.BackendType as a class-level
attribute.  Plain type() stubs have no __getattr__ on the metaclass so this
raises AttributeError.  Introduce _StubClassMeta whose __getattr__ returns
child stub classes, fixing the torchao import chain.

Add an amd-smi circuit breaker in amd.py: after 3 consecutive failures the
module stops spawning the process, eliminating the repeated Windows UAC /
DiskPart elevation prompts caused by polling a non-functional amd-smi.

Also guard BNB_ROCM_VERSION=72 behind a DLL existence check so bitsandbytes
fails with its own detection message rather than a harder "DLL not found" when
the Windows ROCm bnb wheel is not yet installed.
2026-05-11 05:22:03 -05:00
..
__init__.py Final cleanup 2026-03-12 18:28:04 +00:00
resume.py Studio: Add checkpoint resume for stopped training runs (#5255) 2026-05-04 00:34:46 +04:00
trainer.py feat(studio): add Continued Pretraining (CPT) as a training method (#4677) 2026-05-06 13:38:35 +04:00
training.py feat(studio): add Continued Pretraining (CPT) as a training method (#4677) 2026-05-06 13:38:35 +04:00
worker.py fix(rocm/win): stub class metaclass for ProcessGroup.BackendType; amd-smi circuit breaker 2026-05-11 05:22:03 -05:00