Repository navigation
[Bug][ONNX] SimplifiedLayerNormalization and SkipSimplifiedLayerNormalization ops unsupported #20625
Description
Expected behavior
Models Xenova/sweep-next-edit-1.5B and SwinliQ-AIs/SmolLM2-1.7B-Instruct should be executed end-to-end in TVM Relax.
Actual behavior
Model import fails with the following error:
Traceback (most recent call last):
File "/home/mcw/netra/tvm-build-workspace/tvm-relax/netra-files/onnx-model-test.py", line 138, in <module>
mod_onnx = from_onnx(
File "/home/mcw/netra/tvm-build-workspace/tvm-relax/python/tvm/relax/frontend/onnx/onnx_frontend.py", line 6806, in from_onnx
return g.from_onnx(graph, opset)
File "/home/mcw/netra/tvm-build-workspace/tvm-relax/python/tvm/relax/frontend/onnx/onnx_frontend.py", line 6331, in from_onnx
self._check_for_unsupported_ops(graph)
File "/home/mcw/netra/tvm-build-workspace/tvm-relax/python/tvm/relax/frontend/onnx/onnx_frontend.py", line 6477, in _check_for_unsupported_ops
raise tvm.error.OpNotImplemented(msg)
tvm.error.OpNotImplemented: The following operators are not supported for frontend ONNX: GroupQueryAttention, SimplifiedLayerNormalization, SkipSimplifiedLayerNormalization
Environment
- TVM version: 0.28.dev114
- Frontend: Relax ONNX frontend
- Relevant files:
python/tvm/relax/frontend/onnx/onnx_frontend.py
tests/python/relax/test_frontend_onnx.py
Steps to reproduce
from huggingface_hub import hf_hub_download
import onnx
from tvm import relax
from tvm.relax.frontend.onnx import from_onnx
model_path = hf_hub_download(
repo_id="Xenova/sweep-next-edit-1.5B",
filename="onnx/model.onnx",
)
data_path = hf_hub_download(
repo_id="Xenova/sweep-next-edit-1.5B",
filename="onnx/model.onnx_data",
)
data_1_path = hf_hub_download(
repo_id="Xenova/sweep-next-edit-1.5B",
filename="onnx/model.onnx_data_1",
)
data_2_path = hf_hub_download(
repo_id="Xenova/sweep-next-edit-1.5B",
filename="onnx/model.onnx_data_2",
)
print("Model:", model_path)
onnx_model = onnx.load(model_path)
shape_dict = {
"input_ids": [1, 128],
"attention_mask": [1, 128],
"token_type_ids": [1, 128],
}
mod_onnx = from_onnx(
onnx_model,
shape_dict=shape_dict,
keep_params_in_input=True,
)
mod_onnx, params_onnx = relax.frontend.detach_params(mod_onnx)
mod_onnx.show()Triage
- needs-triage
- bug
Activity
[-][FRONTEND][ONNX] OpNotImplemented: SimplifiedLayerNormalization and SkipSimplifiedLayerNormalization[/-][+][Bug][ONNX] SimplifiedLayerNormalization and SkipSimplifiedLayerNormalization ops unsupported[/+]Netra-Srinivasan-McW commentedon Oct 14, 2026
I have raised a PR addressing this issue. #20607
Netra-Srinivasan-McW commentedon Oct 14, 2026
@tqchen @tlopex Could you please take a look into this PR when you have time. Thanks.