issues
search
MadryLab
/
trak
A fast, effective data attribution method for neural networks in PyTorch
https://trak.csail.mit.edu/
MIT License
180
stars
26
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Questions about LDS Calculation and Examples in CIFAR and QNLI
#79
hhf-hd
opened
2 days ago
0
Fix: Change BERT attention implementation to `eager` in the QNLI example
#78
Yupei-Du
closed
1 week ago
1
Errors when running QNLI demo
#77
aski02
opened
2 weeks ago
2
text-classification issue
#76
HARISHSENTHIL
opened
1 month ago
0
Request for Complete Training Code for QNLI Task
#75
hhf-hd
closed
1 month ago
0
TRAK on 5D shapes
#74
Bas-2k
opened
2 months ago
0
`CLIPModelOutput` gradient computation wrong dimensionality
#73
dwahdany
closed
2 months ago
1
fix: CLIP out to loss grad dim
#72
dwahdany
closed
2 months ago
2
Add NaN checks for inverse computation. PyTorch 2.0 - 2.4 deadlocks on matrix inversion
#71
BrunoKM
closed
2 months ago
1
TRAK and fast-jl use deprecated `setup.py install` method
#70
BrunoKM
opened
3 months ago
1
`fast-jl` doesn't work with `pip` version of 23 and above
#69
BrunoKM
opened
6 months ago
2
RuntimeError: CUDA error: the provided PTX was compiled with an unsupported toolchain.
#68
simplelifetime
closed
6 months ago
3
Fast_jl undefined symbol
#67
Bas-2k
closed
5 months ago
4
Request for a ChatBot Example
#66
Taha-Bahadori
closed
7 months ago
1
Trouble Matching Margin Values in cifar_quickstart.ipynb
#65
amberyzheng
closed
7 months ago
2
0.3.2
#64
kristian-georgiev
closed
10 months ago
1
[added feature] regularization term for inv(xtx) calculation
#63
heale04
closed
10 months ago
1
Remove dependency of proj_matrix during scoring
#61
JD-ETH
closed
10 months ago
0
fix finalize features floods CPU memory: remove log and load mapped gradients to device
#59
JD-ETH
closed
10 months ago
1
[Bug Fix] fix `is_grads_dict` for `IterativeGradientComputer`
#58
TheaperDeng
closed
10 months ago
3
Support the LDS metric
#55
Jiaxin-Wen
closed
1 year ago
1
0.3.1
#54
kristian-georgiev
closed
1 year ago
0
Error when execute `finalize_scores`
#53
Jiaxin-Wen
closed
1 year ago
2
Regarding the model input during `traker.featurize`
#52
Jiaxin-Wen
closed
1 year ago
5
Regarding the implementation of `TextClassificationModelOutput`
#51
Jiaxin-Wen
closed
1 year ago
4
0.3.0
#50
kristian-georgiev
closed
1 year ago
0
0.2.2
#49
kristian-georgiev
closed
1 year ago
1
RuntimeError: input.dtype() == torch::kFloat16 || input.dtype() == torch::kFloat32 error on QNLI
#48
rishabluthra
closed
1 year ago
1
How to implement get_output() function for segmentation task?
#47
yangchaohua
closed
1 year ago
1
Order of Arguments in LLM can mismatch
#46
awe2
closed
1 year ago
2
Parallelized score finalization?
#45
nickk124
closed
1 year ago
4
Gradient_Computers grad.dtype shouldn't inherit from batch[0].dtype?
#44
awe2
closed
1 year ago
2
Blockwise matmul for scoring
#43
kristian-georgiev
closed
1 year ago
5
CLIP Fixes updates
#42
andrewilyas
closed
1 year ago
0
all NaN or all zero scores?
#41
nickk124
closed
1 year ago
6
0.2.1
#40
kristian-georgiev
closed
1 year ago
0
0.2.0
#38
kristian-georgiev
closed
1 year ago
0
Can trak be applied to diffusion models?
#37
xszheng2020
closed
1 year ago
0
fixed model output function when computing gradients in float16
#36
AlaaKhaddaj
closed
1 year ago
1
Clean up docs & tutorials to match >=0.2.0 methods
#35
kristian-georgiev
closed
1 year ago
1
Migrate from `functorch` to `torch.func`
#34
kristian-georgiev
closed
1 year ago
2
Save scores in memory-mapped files and allow loading existing scores from the `TRAKer` class
#33
kristian-georgiev
closed
1 year ago
1
0.1.3
#32
kristian-georgiev
closed
1 year ago
0
Add a colab where one can download pre-computed TRAK scores and visualize them.
#31
kristian-georgiev
closed
1 year ago
1
Save scores in half-precision format by default
#30
kristian-georgiev
closed
1 year ago
3
Support dataset sharding
#29
scoutsaachi
closed
1 year ago
5
0.1.2
#28
kristian-georgiev
closed
1 year ago
0
Allow changing saver and score computer directly from `TRAKer`
#27
kristian-georgiev
closed
1 year ago
1
CudaProjector gives `CUDA error: too many resources requested for launch` on V100
#26
kristian-georgiev
closed
1 year ago
1
Fix bug with custom model output and add test
#25
jvendrow
closed
1 year ago
1
Next