Skip to content

[Relax][Frontend][ONNX] Fix FastGelu when bias does not set #18358

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

Merged
tlopex merged 4 commits into from
Oct 7, 2025
Merged

[Relax][Frontend][ONNX] Fix FastGelu when bias does not set #18358

Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
31e3bda
[#17877][Relax][Frontend][ONNX] Fix when bias does not set
cchung100m Oct 2, 2025
05d9868
[#17877][FRONTEND][ONNX] Fix Error converting operator FastGelu, with…
cchung100m Oct 2, 2025
e0adf48
[#17877][FRONTEND][ONNX] Fix Warning: Detected pow(x, y) where y >= 3…
cchung100m Oct 2, 2025
a686151
[#17877][FRONTEND][ONNX] Fix tvm.error.InternalError: Check failed: (…
cchung100m Oct 7, 2025
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
Prev Previous commit
[#17877][FRONTEND][ONNX] Fix tvm.error.InternalError: Check failed: (… 
…ptr) is false: The struct_info is not populated, check if you have normalized the expr
  • Loading branch information
@cchung100m
cchung100m committed Oct 7, 2025
commit a68615179047b3ac6c970670b794e424190caeea
14 changes: 7 additions & 7 deletions python/tvm/relax/frontend/onnx/onnx_frontend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1160,7 +1160,7 @@ def _impl_v1(cls, bb, inputs, attr, params):
bias = inputs[1]
bias_shape = bias.struct_info.shape
assert len(bias_shape) == 1, "bias term must be a 1D tensor"
x = relax.op.add(x, bias)
x = bb.emit(relax.op.add(x, bias))

# Declare consts
const_dtype = x.struct_info.dtype
Expand All @@ -1170,13 +1170,13 @@ def _impl_v1(cls, bb, inputs, attr, params):
const2 = relax.const(0.044715 * math.sqrt(2 / math.pi), dtype=const_dtype)

# Compute FastGelu
term1 = relax.op.multiply(half, x)
term2 = relax.op.multiply(const1, x)
term1 = bb.emit(relax.op.multiply(half, x))
term2 = bb.emit(relax.op.multiply(const1, x))
# use x^3 = x * x * x instead of pow(x, 3) for better performance
x_cubed = relax.op.multiply(relax.op.multiply(x, x), x)
term3 = relax.op.multiply(const2, x_cubed)
tanh = relax.op.tanh(relax.op.add(term2, term3))
return relax.op.multiply(term1, relax.op.add(one, tanh))
x_cubed = bb.emit(relax.op.multiply(relax.op.multiply(x, x), x))
term3 = bb.emit(relax.op.multiply(const2, x_cubed))
tanh = bb.emit(relax.op.tanh(relax.op.add(term2, term3)))
return bb.emit(relax.op.multiply(term1, relax.op.add(one, tanh)))


class BiasGelu(OnnxOpConverter):
Expand Down
Loading