Skip to content
Open
Show file tree
Hide file tree
Changes from 1 commit
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions CHANGELOG.rst
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,7 @@ Changelog
- Update HuggingFace checkpoint export to use name-based tied-weight deduplication instead of the previous address-based approach. The address-based deduplication could incorrectly drop an untied weight that happened to share memory with a tied one, producing an incomplete checkpoint (observed as a false positive on MiniMax-M2.7).
- Fix EAGLE-3 training with context parallelism (``--cp_size > 1`` in ``examples/speculative_decoding``), which failed to start on ``accelerate >= 1.13`` and then raised ``got mixed torch.Tensor and DTensor``.
- Polygraphy minimum dependency upgraded to ``0.53.4`` to solve ONNX AutoCast failures when marking optional graph outputs.
- Fix ``ModuleNotFoundError: No module named 'huggingface_hub'`` on any ``modelopt.torch`` import when the optional ``hf`` extra is not installed. ``modelopt.torch.opt.plugins`` imported the HuggingFace checkpointing plugin unconditionally; it is now guarded by ``import_plugin`` like the other third-party plugins. Affects 0.44 through 0.46.
Comment thread
coderabbitai[bot] marked this conversation as resolved.
Outdated

0.46 (2026-08-17)
^^^^^^^^^^^^^^^^^
Expand Down
3 changes: 2 additions & 1 deletion modelopt/torch/opt/plugins/__init__.py
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,8 @@

from modelopt.torch.utils import import_plugin

from .huggingface import *
with import_plugin("huggingface_hub"):
from .huggingface import *

with import_plugin("megatron core dist checkpointing"):
from .mcore_dist_checkpointing import *
Expand Down