[Performance] Quadratic behaviour in list operations with SequenceInsert in onnx #20696
Labels
performance
issues related to performance regressions
stale
issues that have not been addressed in a while; categorized by a bot
Describe the issue
When doing list operations e.g. splitting and concatenating a list, the runtime performance of the onnx exported model increases quadratically with the number of elements.
The expected behaviour is linear dependency on the number of elements as in the torch and torch script implementation.
The problem seems to be in the SequenceInsert onnx operator.
To reproduce
Urgency
This is currently slowing down our inference pipeline and we are trying to avoid all list operations.
Platform
Mac
OS Version
macOS-14.4.1-x86_64-i386-64bit
ONNX Runtime Installation
Released Package
ONNX Runtime Version or Commit ID
1.17.3
ONNX Runtime API
Python
Architecture
X64
Execution Provider
Default CPU
Execution Provider Library Version
CPUExecutionProvider
Model File
list_operation.onnx.zip
Is this a quantized model?
No
The text was updated successfully, but these errors were encountered: