Skip to content

fix: export fp8e4m3fnuz tensors with the correct onnx type #4843

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
rootkiller6788 wants to merge 2 commits into
base: main
Choose a base branch
from
+22 −1
Open

fix: export fp8e4m3fnuz tensors with the correct onnx type #4843

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fbfcfc1
fix: export fp8e4m3fnuz tensors with the correct onnx type
rootkiller6788 Aug 31, 2026
60981d9
test: add fp8e4m3fnuz constant export regression test
rootkiller6788 Aug 31, 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
2 changes: 1 addition & 1 deletion tools/onnx-graphsurgeon/onnx_graphsurgeon/exporters/onnx_exporter.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,7 +50,7 @@ def dtype_to_onnx(dtype: Union[np.dtype, "onnx.TensorProto.DataType"]) -> int:
ml_dtype_to_onnx_name = {
np.dtype(ml_dtypes.bfloat16): "BFLOAT16",
np.dtype(ml_dtypes.float8_e4m3fn): "FLOAT8E4M3FN",
np.dtype(ml_dtypes.float8_e4m3fnuz): "FLOAT8E4M3FN",
np.dtype(ml_dtypes.float8_e4m3fnuz): "FLOAT8E4M3FNUZ",
np.dtype(ml_dtypes.float8_e5m2): "FLOAT8E5M2",
np.dtype(ml_dtypes.float8_e5m2fnuz): "FLOAT8E5M2FNUZ",
np.dtype(ml_dtypes.uint4): "UINT4",
Expand Down
21 changes: 21 additions & 0 deletions tools/onnx-graphsurgeon/tests/test_exporters.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -156,6 +156,27 @@ def test_should_export_constant_tensor_with_ml_dtype(self) -> None:
assert onnx_tensor.data_type == onnx.TensorProto.FLOAT8E4M3FN
assert tuple(onnx_tensor.dims) == shape

def test_should_export_constant_tensor_with_ml_dtype_fnuz(self) -> None:
"""Test that `export_tensor_proto` exports a Constant with ml_dtypes.float8_e4m3fnuz values
using the correct ONNX data type."""
name = "constant_tensor"
shape = (3, 224, 224)
values = np.random.random_sample(size=shape).astype(ml_dtypes.float8_e4m3fnuz)

tensor = Constant(name=name, values=values)

onnx_tensor = OnnxExporter.export_tensor_proto(tensor)

assert onnx_tensor.name == name
assert np.all(
self._bytes_to_np_array(
onnx_tensor.raw_data, onnx_tensor.dims, ml_dtypes.float8_e4m3fnuz
)
== tensor.values
)
assert onnx_tensor.data_type == onnx.TensorProto.FLOAT8E4M3FNUZ
assert tuple(onnx_tensor.dims) == shape

def test_should_export_constant_tensor_with_ml_dtype_raise_error_when_onnx_dtype_not_supported(
self,
) -> None:
Expand Down