trak
trak copied to clipboard
MadryLab
A fast, effective data attribution method for neural networks in PyTorch
If you try to install `fast-jl` or `traker[fast]` with `pip==23.3.2` or `pip==24.0`, you get the following error: ``` Collecting fast-jl==0.1.2 Using cached fast_jl-0.1.2.tar.gz (5.2 kB) Installing build dependencies ... done...
Hello, I have been trying to use TRAK for CelebA dataset for an age classification task using ResNet-50. However, I am unable to use the fast version as I get...
Hi, I have a classification model trained on the shape batch_size x N (=46) x channel x height x width. How can I adapt TRAK to use for that ?...
Installig `fast-jl` with Python 3.11 and `setuptools==65.5.0` gives: ``` > pip install fast-jl ... DEPRECATION: fast-jl is being installed using the legacy 'setup.py install' method, because it does not have...
I would like to express my admiration for your outstanding work. The methods and examples you provided are highly intuitive and leave a lasting impression. The effectiveness of your approach...
As the title mentioned, TRAK is experiencing numerical issues on the cuda version 11.8. - Minimum reproducible code ```python import torch from trak.projectors import ProjectionType, AbstractProjector, CudaProjector grad_dim = int(1e6)...
- minimal code for reproduce the error: ```python import torch import trak from trak.projectors import ProjectionType, AbstractProjector, CudaProjector print("trak.test_install:", trak.test_install(use_fast_jl=True)) grad_dim = int(1e6) projector = CudaProjector( grad_dim=grad_dim, proj_dim=32768, seed=42, proj_type=ProjectionType.normal,...
README.md checkpoint-1500 checkpoint-3000 checkpoint-4500 checkpoint-5500 checkpoint-7000 checkpoint-8500 checkpoint-9819 model.safetensors tokenizer_config.json training_args.bin all_results.json checkpoint-2000 checkpoint-3500 checkpoint-500 checkpoint-6000 checkpoint-7500 checkpoint-9000 config.json special_tokens_map.json train_results.json vocab.txt checkpoint-1000 checkpoint-2500 checkpoint-4000 checkpoint-5000 checkpoint-6500 checkpoint-8000 checkpoint-9500 eval_results.json...
