[Question] How to keep the consistency of ActivationStore

[luckyeric320]

Hi, I am trying to use ActivationsStore method to generate batched tokens but I fail to maintain the consistence of these tokens for each run in Jupyter Notebook.

Here is the code:

from sae_lens import ActivationsStore activation_store = ActivationsStore.from_sae( model=model, sae=sae, streaming=True, store_batch_size_prompts=8, train_batch_size_tokens=4096, n_batches_in_buffer=32, device=device, )

I set the random seed for random, numpy and pytorch but the results keep changing each time I run the cell. Is there a way to keep the consistencey of batched tokens getted from ActivationsStore?

Also, I am wondering whether it is possible to change the dataset for ActivationsStore so that I can control the content of the dataset, probably making it related to certain concept or theme.

I am new to this library and would very appreciate any help or advice.