ag1988 / injecting_numeracy Goto Github PK

The accompanying code for "Injecting Numerical Reasoning Skills into Language Models" (Mor Geva*, Ankit Gupta* and Jonathan Berant, ACL 2020).

Python 99.68% Shell 0.32%

injecting_numeracy's Issues

pretrained model checkpoint

Hi,

I want to initialize my model with GENBERT weights and run few experiments. If you allow the pretrained model to be downloaded, where can I download it from?

Thank you,

Looking forward to the pre-trained model

Very interesting work! I'm wondering when the pre-trained model would be kindly released.

Decoding strategy

The customized class BertTransformer only supports greedy decoding?

Arguments for Textual Data generation

What are the arguments to textual_data_generation/generate_examples.py for re-creating the textual data used in the paper?

License for the code

Hi,

I want to use the code for research purposes and knowing the license under which code is released would be great. Can you please add a license file or mention the license which applies to this code?

Hi Ankit, in the paper "digit tokenization" was implemented but I cannot follow how this is done in the script. I had a quick search in finetune_on_drop.py, the argumentt indiv_digits is commented out. Could you help point this part out in the code? Thanks a lot!

Recommend Projects