KnightZhang issues

Results 5 issues of


                                            KnightZhang

Input_mask and number_indices doesn't match, because of [cls] at the beginning

``` def convert_single_mathqa_example(example, is_training, tokenizer, max_seq_length, max_program_length, op_list, op_list_size, const_list, const_list_size, cls_token, sep_token): """Converts a single MathQAExample into an InputFeature.""" features = [] question_tokens = example.question_tokens if len(question_tokens) > max_seq_length...

Question about why "number_indices" plus 1

Hello, In "dropreader.py". Why you plus 1 for the indices of each number in the passage? And why append 100 and -1 in the following code? I am confused because...

Just want to Thanks for this!

Thanks for your excellent work!

Question about the layer representation from the first token of the X

Hello, Thanks for the easy-read code. I have a question about the code, could you please help me? In your paper, you said "we use the hidden state of the...

[BUG] Inconsistence implementation between the MoE in this repo and MoE in transformer version (Huggingface)

Looks like the implementation of MoE in transformer version (Huggingface) is wrong. The intended flow should be: - For each token, get outputs from topk experts - Weight these outputs...