minnervva / torchdetscan

This is a tool for finding non-deterministic functions in your pytorch code.
https://github.com/minnervva/torchdetscan
MIT License
1 stars 1 forks source link

Make linter output cleanup #41

Closed markcoletti closed 6 months ago

markcoletti commented 6 months ago

Currently, the linter output looks like this:

Linting file: pytorch_conditional.py
Found non-deterministic function Conv1d at line 12, column 9
Found non-deterministic function Conv2d at line 13, column 9
Found non-deterministic function Conv3d at line 14, column 9
Found non-deterministic function ConvTranspose1d at line 16, column 9
Found non-deterministic function ConvTranspose2d at line 17, column 9
Found non-deterministic function ConvTranspose3d at line 18, column 9
Found non-deterministic function ReplicationPad2d at line 20, column 9
Found non-deterministic function bmm at line 22, column 9
Found non-deterministic function index_put at line 24, column 9
Found non-deterministic function put_ at line 26, column 9 with accumulate=True that will be non-deterministic if used with a CUDA tensor
Found non-deterministic function scatter_add_ at line 28, column 9
Found non-deterministic function gather at line 30, column 9
Found non-deterministic function index_add at line 32, column 9
Found non-deterministic function index_select at line 34, column 9
Found non-deterministic function repeat_interleave at line 36, column 9
Found non-deterministic function index_copy at line 38, column 9
Found non-deterministic function scatter at line 40, column 9

This may be a little hard to read, so something that's a little more tabular might work better.