Ananth Subramaniam
Ananth Subramaniam
An implementation of the ideal flop counter is here: https://github.com/pytorch-labs/torcheval/blob/main/torcheval/tools/module_summary.py https://github.com/pytorch-labs/torcheval/blob/main/torcheval/tools/flops.py
Hi @yiminglin-ai, could you describe your use case in more detail? For context, that field was added to support gradient accumulation in the `AutoUnit` extension. However, subsequent PRs made it...
/ok to test 1a375b19dadd218ad7e154d80d2d89d12e8b12e3
/ok to test a6e6d0e7bbb9fab917eb23881734134e91cbdda6
/ok to test 7c9d9bdbbf3eeb8f573dd213e3454c9ce8a1e8ca