negar-foroutan / multilms-lang-neutral-subnets

[EMNLP 2022] Discovering Language-neutral Sub-networks in Multilingual Language Models.

negar-foroutan

multilms-lang-neutral-subnets's Issues

Variable definition problem

Code error message:
These two variables are defined after the reference.
"multiLMs-lang-neutral-subnets\helpers\data_processor.py", line 47, in init
num_examples = int(len(tokenized_text)/ block_size)
UnboundLocalError: local variable 'block_size' referenced before assignment

"multiLMs-lang-neutral-subnets\helpers\data_processor.py", line 67, in init
logger.info("Saving features into cached file %s", cached_features_file)
UnboundLocalError: local variable 'cached_features_file' referenced before assignment

Missing parameter

File "LT_ner.py", line 415, in train
if args.pruning_method == "component_wise":
AttributeError: 'LTHTrainingArguments' object has no attribute 'pruning_method'

Can not find this parameter definition in the code, how to solve it？

Inquiry about Function Dependency in analysis_tools.py

Could you please clarify whether the get_mbert_mask_ones_overlap(mask_list) function is capable of directly computing the similarity between two subnetworks without relying on other functions? (Because I found other functions in this file that I didn't use, I was a little confused)
If it does depend on other functions, could you kindly provide some insight into which functions are necessary for its operation?