captum icon indicating copy to clipboard operation
captum copied to clipboard

Model interpretability and understanding for PyTorch

Results 290 captum issues
Sort by recently updated
recently updated
newest added

## 🐛 Bug I'm experiencing RAM leaks when calculating word attributions through https://github.com/cdpierse/transformers-interpret library which delegates the most of the heavy lifting to Captum. So I assume this issue is...

I'm trying to use Captum on an LSTM model. I used similiar Captum code to interpret a CNN and a BERT model with lig, but I get the above error...

Hello, I noticed that captum currently does not support the ability to directly extract feature maps, nor does it support attention-based vision models, such as Vit, MVit, etc. Do you...

Adds the the following relevance decomposition rules for LRP: alpha beta rule, z bound rule, w^2 rule and flat rule (cf. https://link.springer.com/chapter/10.1007/978-3-030-28954-6_10) as addressed in https://github.com/pytorch/captum/issues/485 for linear layers. This...

Dear developers, I encountered an error in LRP (and LayerLRP) that is caused by using the same layer (in this case a pooling layer) twice in the model. The error...

I was trying to adapt the BERT tutorial to work with BART. However, when I run the attribution, the forward function eventually breaks. After stepping through the code, I see...

I am trying to get the saliency visualization for my model which predicts 68 gray matter volumes. I want to get the saliency visualization for each of the 68 regions...

I am facing an issue with the Shap-based explainers for the XLnet model for the IMDB dataset. I am using a batch size of 1 for getting the attribution from...

Summary: - adds progress bar for computation done in `TracInCPFastRandProj.__init__`, as well as a `show_progress` argument - actual computation is done in `TracInCPFastRandProj._get_intermediate_quantities_tracincp_fast_rand_proj`. This method is changed so that the...

fb-exported
cla signed

Hi, is there a tutorial on Detectron2 mask-rcnn explainablity (specifically for *instance segmentation*)? I read another issue mentioning that it it is difficult. I would like to write a tutorial...