search

frgfm / torch-scan

Seamless analysis of your PyTorch models (RAM usage, FLOPs, MACs, receptive field, etc.)
https://frgfm.github.io/torch-scan/latest
Apache License 2.0
208 stars 22 forks source link

Negative values for memory overhead when testing on smaller networks #37

Closed EliasVansteenkiste closed 2 years ago

EliasVansteenkiste commented 3 years ago

When I test torch-scan on small networks, I get negative values for Framework & CUDA overhead and Total RAM usage.

Any idea how to fix it?

Thanks in advance

 Trainable params: 47,073
Non-trainable params: 0
Total params: 47,073
---------------------------------------------------------------------------------------------
Model size (params + buffers): 0.18 Mb
Framework & CUDA overhead: -24.64 Mb
Total RAM usage: -24.46 Mb
---------------------------------------------------------------------------------------------
Floating Point Operations on forward: 67.61 MFLOPs
Multiply-Accumulations on forward: 34.70 MMACs
Direct memory accesses on forward: 34.57 MDMAs
frgfm commented 3 years ago

Hi @EliasVansteenkiste !

Thanks for reporting this! This is odd, could you specify a minimal code snippet to reproduce this please? (architecture included)

I suspect the RAM overhead computation failed because of an issue with nvidia-smi. But I'd need to be able to reproduce the error to investigate :pray:

frgfm commented 3 years ago

@EliasVansteenkiste any update?

frgfm commented 3 years ago

ping @EliasVansteenkiste :pray:

frgfm commented 3 years ago

Hello @EliasVansteenkiste :wave: It's been quite a while, if I can't reproduce the error, I cannot do much. Would you mind sharing how to reproduce it?

frgfm commented 2 years ago

I'm closing this issue since I don't have any way of reproducing this unfortunately :/ @EliasVansteenkiste if you have time at some point, please post more details 🙏