Skip to content

Can't generate Constant nodes with attributes as vectors #1282

Open

Description

I find there are some bugs when trying to use attributes as vector for constant node input inside a local function, i.e.

These codes could work:

from onnxscript import opset18 as op
from onnxscript import script

import numpy as np


@script()
def LocalFunc(X, alpha: int):
    alphaX = op.Constant(value_int=alpha)
    gammaX = op.Constant(value_ints=[1, 2])
    Add = alphaX + X + gammaX
    return Add


v = np.array([[0, 1], [2, 3]], dtype=np.int64)
result = LocalFunc(v, 1)
print(result)

But this snippet would report errors:

@script()
def LocalFunc1(X, alpha: int, gamma: int):
    alphaX = op.Constant(value_ints=[alpha, gamma])
    Add = alphaX + X
    return Add


v = np.array([[0, 1], [2, 3]], dtype=np.int64)
result = LocalFunc1(v, 1, 2)
print(result)

Errors are like

...
onnxscript\converter.py", line 471, in _eval_constant_expr
    return eval(cpl, self.globals, locals)  # pylint: disable=eval-used
  File "<ast>", line 3, in <module>
NameError: name 'alpha' is not defined
...

NameError: ERROR: Missing names, globals contains ['__name__', '__doc__', '__package__', '__loader__', '__spec__', '__annotations__', '__builtins__', '__file__', '__cached__', 'op', 'script', 'np', 'LocalFunc', 'v', 'result'], locals [].
at: Function 'LocalFunc1', line 3
    alphaX = op.Constant(value_ints=[alpha, gamma])
                                    ^


I think there might be a bug in the Constant node conversion, and such bug makes it hard for us to use multiple attributes directly in local function design.
Could you fix it?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Assignees

No one assigned

    Labels

    questionFurther information is requested

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions