You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Describe the bug
Running export_hugectr_ensemble with merlin-training:21.09 container for scaling-criteo example (04-Triton-Inference-with-HugeCTR.ipynb) gives the following error:
ValueError Traceback (most recent call last)
/tmp/ipykernel_36095/230262359.py in <module>
7 hugectr_params["embedding_vector_size"] = 128
8 hugectr_params["n_outputs"] = 1
----> 9 export_hugectr_ensemble(
10 workflow=workflow,
11 hugectr_model_path="./criteo_hugectr/1/",
/nvtabular/nvtabular/inference/triton/__init__.py in export_hugectr_ensemble(workflow, hugectr_model_path, hugectr_params, name, output_path, label_columns, version, cats, conts, max_batch_size, nvtabular_backend)
264 raise ValueError("Either cats or conts has to have a value.")
265
--> 266 workflow = _remove_columns(workflow, label_columns)
267
268 # generate the nvtabular triton model
/nvtabular/nvtabular/inference/triton/__init__.py in _remove_columns(workflow, to_remove)
739 # TODO: Handle selector sub-groups?
740 if column in node.selector.names:
--> 741 node.selector._names.remove(column)
742
743 return workflow.fit_schema(new_schema)
ValueError: list.remove(x): x not in list
The inference notebooks with the Criteo example were broken - and failed
to generate triton config with an in _remove_columns like ```ValueError: list.remove(x): x not in
list```. This was because the label column wa being inserted into a subgroup of the outputnode,
and wasn't getting removed from there.
Fix and add a basic unittest for _remove_columns that would have caught this.
ClosesNVIDIA-Merlin#1198
benfred
added a commit
to benfred/NVTabular
that referenced
this issue
Oct 21, 2021
The inference notebooks with the Criteo example were broken - and failed
to generate triton config with an in _remove_columns like ```ValueError: list.remove(x): x not in
list```. This was because the label column wa being inserted into a subgroup of the outputnode,
and wasn't getting removed from there.
Fix and add a basic unittest for _remove_columns that would have caught this.
ClosesNVIDIA-Merlin#1198
The inference notebooks with the Criteo example were broken - and failed
to generate triton config with an in _remove_columns like ```ValueError: list.remove(x): x not in
list```. This was because the label column wa being inserted into a subgroup of the outputnode,
and wasn't getting removed from there.
Fix and add a basic unittest for _remove_columns that would have caught this.
Closes#1198
The inference notebooks with the Criteo example were broken - and failed
to generate triton config with an in _remove_columns like ```ValueError: list.remove(x): x not in
list```. This was because the label column wa being inserted into a subgroup of the outputnode,
and wasn't getting removed from there.
Fix and add a basic unittest for _remove_columns that would have caught this.
Closes#1198
Describe the bug
Running
export_hugectr_ensemble
withmerlin-training:21.09
container forscaling-criteo
example (04-Triton-Inference-with-HugeCTR.ipynb
) gives the following error:Steps/Code to reproduce bug
Follow this: https://github.com/NVIDIA-Merlin/NVTabular/blob/main/examples/scaling-criteo/04-Triton-Inference-with-HugeCTR.ipynb
Expected behavior
Creates
config.pbtxt
fileEnvironment details (please complete the following information):
Docker : merlin-training:21.09
NVIDIA-SMI 450.119.04,Driver Version: 450.119.04,CUDA Version: 11.4
The text was updated successfully, but these errors were encountered: