pytorch / TensorRT

PyTorch/TorchScript/FX compiler for NVIDIA GPUs using TensorRT
https://pytorch.org/TensorRT
BSD 3-Clause "New" or "Revised" License
2.6k stars 351 forks source link

switch from fx.symbolic_trace to dynamo_trace for converter test part-1 #3261

Closed lanluo-nvidia closed 3 weeks ago

lanluo-nvidia commented 1 month ago

Description

Switch from fx.symbolic_trace to dynamo trace for converter test part-1

Since we ~2000 test cases I splited into batches to migrate them off fx.symbolic_trace

current batch: tests/py/dynamo/conversion/test_a*.py

There mainly two types of testcases require change:

  1. for ops require custom dynamic shape support, will be changed to use dynamic_shapes instead of Input
  2. for ops will be decomposed into multiple ops, these op test will be added in the test_composite_aten_op.py

Type of change

Please delete options that are not relevant and/or add your own.

Checklist: