There is a discrepancy in the number of triggers

[ysf123456]

Taking sst2 as an example, according to the word_countall.json you provided, with selected_trigger_num set to 20, the number of triggers counted is:
wandb: trigger_num_0 65708
wandb: trigger_num_1 2429
wandb: trigger_num_2 78
wandb: trigger_num_3 6
This is somewhat different from what was given in the paper. Can you help me clarify it?

[yjw1029]

When we reproduced the results using the Github code in June 2023, we were still able to replicate the results from the paper. The records in Wandb are as follows:

The command recorded in Wandb was:

Since the trend of the data you provided is not significantly different from the actual data, it is possible that some random factors, such as modifications to the random number generator, platform differences, or other random elements, led to a different trigger selection than the one we used for the reported experiment. You may try the following:

1. Check whether the running command matches the command mentioned above.
2. Verify whether the running environment is consistent with the environment we provided.