I was running FlopCountAnalysis on a continuous-time RNN model I created. For this and almost all custom modules I make, I get some Unsupported operators that seems like basic operators (like add and mul) that SHOULD be counted for a correct FLOP count analysis.
[2023-12-18 14:17:25,808][models._main][INFO] - Initialized a new model: NetworkCTRNN.
[2023-12-18 14:17:25,808][models._main][INFO] - Hidden size: 54
[2023-12-18 14:17:27,927][train._main][INFO] - Starting training loop...
[2023-12-18 14:17:29,696][fvcore.nn.jit_analysis][WARNING] - Unsupported operator aten::expand_as encountered 1 time(s)
[2023-12-18 14:17:29,700][fvcore.nn.jit_analysis][WARNING] - Unsupported operator aten::mul encountered 201 time(s)
[2023-12-18 14:17:29,700][fvcore.nn.jit_analysis][WARNING] - Unsupported operator aten::add encountered 200 time(s)
[2023-12-18 14:17:29,700][fvcore.nn.jit_analysis][WARNING] - Unsupported operator aten::sigmoid encountered 200 time(s)
[2023-12-18 14:17:29,700][fvcore.nn.jit_analysis][WARNING] - Unsupported operator aten::rsub encountered 100 time(s)
I was running FlopCountAnalysis on a continuous-time RNN model I created. For this and almost all custom modules I make, I get some Unsupported operators that seems like basic operators (like add and mul) that SHOULD be counted for a correct FLOP count analysis.