Skip to content

Fix legacy ONNX export crash with transformers >= 5.0 #2381

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Lidang-Jiang wants to merge 2 commits into
base: main
Choose a base branch
from
+25 −0
Open

Fix legacy ONNX export crash with transformers >= 5.0 #2381

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
bee8db7
Fix legacy ONNX export crash with transformers >= 5.0
Lidang-Jiang Apr 5, 2026
139586e
Merge branch 'main' into fix/auto-opt-transformers5
shaahji Apr 8, 2026
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
25 changes: 25 additions & 0 deletions olive/passes/onnx/conversion.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -189,6 +189,11 @@ def _patch_model_if_necessary(pytorch_model: torch.nn.Module):
if transformers_version < version.parse("4.45"):
return

# from_legacy_cache / to_legacy_cache were removed in transformers 5.0.
# This patch is only needed for transformers 4.45 <= version < 5.0.
if transformers_version >= version.parse("5.0"):
return

orig_forward_name = "forward" if hasattr(pytorch_model, "forward") else "call"
orig_forward = getattr(pytorch_model, orig_forward_name)
signature = inspect.signature(orig_forward)
Expand Down Expand Up @@ -713,6 +718,26 @@ def _convert_model_on_device(
pytorch_model = model.load_model(cache_model=False)
pytorch_model.eval()

# Check transformers >= 5.0 compatibility with legacy ONNX export.
# transformers 5.0 removed DynamicCache.from_legacy_cache() and changed the cache API
# (e.g., past_key_values must be a Cache object, not a list of tuples), making the
# legacy (non-dynamo) ONNX export path incompatible for models with KV cache.
if (
not config.use_dynamo_exporter
and version.parse(transformers.__version__) >= version.parse("5.0")
and isinstance(model, HfModelHandler)
):
raise RuntimeError(
f"Legacy ONNX export (use_dynamo_exporter=False) is not compatible with "
f"transformers {transformers.__version__}. transformers >= 5.0 changed the "
"DynamicCache API, which breaks the non-dynamo export path for models with "
"KV cache. Please use one of the following options:\n"
" 1. Add --use_model_builder --use_ort_genai to use the model builder (recommended)\n"
" 2. Add --use_dynamo_exporter to use the dynamo-based ONNX export\n"
" 3. Downgrade transformers below 5.0\n"
"Example: olive auto-opt --model_name_or_path <model> --use_model_builder --use_ort_genai ..."
)

# get dummy inputs
dummy_inputs = _get_dummy_inputs(model, config)
io_config = model.io_config
Expand Down
Loading