captum

captum copied to clipboard

## ❓ Questions and Help import torch import torch.nn as nn import pandas as pd from sklearn.preprocessing import LabelEncoder, StandardScaler from sklearn.model_selection import train_test_split class TitanicDNNModel(nn.Module): def __init__(self, num_sex, num_embarked,...

quanxing

Hi everyone, I am trying to use captum's TCAV for a pre-trained TinyLlama model (text generation model). I have followed the tutorial (https://captum.ai/tutorials/TCAV_NLP) but am facing the following error related...

khushaalkurswani

j-t-1

j-t-1

Initialize parameterized test attributes

1

Summary: Make Pyre happy. The actual values are overridden by parameterized_class args Differential Revision: D73884605

sarahtranfb

Futures support with cross-tensor attribution 2/n

2

Summary: Copypasta, refactor in next diff Differential Revision: D73466780

sarahtranfb

Add enable_cross_tensor_attribution flag to attribute_future

3

Summary: reserved Differential Revision: D73464680

sarahtranfb

Failing tests on `aarch64-darwin`

1

## 🐛 Bug Some tests are failing on `aarch64-darwin`, e.g.: ``` > self._compute_cavs_interpret( [["striped", "random"], ["ceo", "random"]], True, -1.0, # acc is not defined, this field will not be asserted...

drupol

Fix total_forwards calculation in ablation/permutation for cross-tensor attribution

1

Summary: Previously, we could iterate over the feature masks and get the feature count from the ID range in the mask. Now mask ID/indices are global Differential Revision: D72480911

sarahtranfb

Making LLMAttribute work with BertForMultipleChoice models

3

## 🚀 Feature Allow LLMAttribution goodness to be applied to BERT models for multiple choice tasks ## Motivation following up on [suggestions from aobo-y](https://github.com/pytorch/captum/issues/1523#issuecomment-2705610883) ## Pitch Integrated gradient attribution techniques...

rbelew