Issues
- 1
Grad Sample Module: Use full backward hook to save activations and backprop values.
#652 opened by ParthS007 - 4
ValueError: Per sample gradient is not initialized. Not updated in backward pass?
#645 opened by Lhsakfheudsj35407 - 2
Spectral normalization in Opacus
#646 opened by 100sunnygirl - 2
Error in DPOptimizer: Inconsistency between batch_first argument of PrivacyEngine and DPMultiheadAttention
#650 opened by tklausen - 1
IndexError: pop from empty list
#624 opened by shanjin2014 - 1
- 1
Backward pass with Opacus and ViT model. Batching rule not implemented for aten::_chunk_grad_outputs_efficient_attention.
#595 opened by sebasrb09 - 1
Wrapper references can be easily replaced, consider using properties instead
#649 opened by andreatgretel - 0
BatchMemoryManager for training private GANs
#647 opened by Caarinaaa - 15
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat2 in method wrapper_CUDA_mm)
#612 opened by FryLcm - 5
`BatchSplittingSampler` return wrong length
#640 opened by lsc64 - 13
- 0
- 2
OverflowError: cannot convert float infinity to integer
#637 opened by Tnia1 - 5
Opacus with torch_geometric.nn and GCN's
#588 opened by SagerKudrick - 3
- 5
- 2
Negative Epsilons at high deltas
#585 opened by tudorcebere - 2
- 1
- 2
Integrating Opacus for a custom pytorch model raises errors but works fine on its own
#639 opened by nhianK - 4
- 1
- 0
Error occurred when executing GroundingDinoSAMSegment (segment anything): Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat2 in method wrapper_CUDA_mm)
#642 opened by Ymshkz - 2
LLM finetuning with Opacus
#632 opened by KKNakkav2 - 1
Implementing augmentation multiplicity using Functorch [Error: AttributeError: 'Tensor' object has no attribute '_forward_counter]
#626 opened by mshubhankar - 1
UnsupportedModuleError: [IllegalModuleConfigurationError('Model needs to be in training mode')]
#629 opened by jS5t3r - 4
- 3
- 7
The privacy engine make_private_with_epsilon function leads to memory error even for a very small network.
#602 opened by telegraphroad - 2
- 2
Can opacus only inject differential noise in only one layer of parameters or some target layers?
#615 opened by hacker-jerry - 3
DPMultiheadedAttention is not a full drop in replacement for nn.MultiheadAttention
#596 opened by jfb54 - 8
- 2
- 3
- 2
- 2
Cannot compute epsilon
#591 opened by 01manoj20 - 0
- 6
ImportError: cannot import name 'DifferentiallyPrivateDistributedDataParallel' from 'opacus.layers'
#597 opened by Ethan-Chen-plus - 1
Real-time change of clipping thresholds
#594 opened by Jenson66 - 1
- 1
How to integrate Opacus with Nvidia NeMo
#593 opened by yangjundy - 1
WeightedRandomSampler is invalid
#600 opened by LanpingTech - 0
More less the same value over different EPS values
#587 opened by zedoul - 1
- 3
- 1
How to fetch the original model again?
#582 opened by TsingZ0 - 4
- 2
TypeError: 'type' object is not subscriptable
#577 opened by GuYueJiahui