matthieufp / vgamt Goto Github PK

Hi Matthieu,
Thank you for your great work. I was reading through the downloading and pre-processing of Text-only data https://github.com/MatthieuFP/VGAMT/blob/main/data/README.md#text-only-data. However, I am not sure what exactly to download from the https://opus.nlpl.eu/ website. It would be very helpful if you could provide more information on this as there are a lot of different corpora available.
Also, in the script that you have provided to pre-process the text-only data, the scripts/text_only_preprocessing.sh is missing.
Thanks.

Hi Matthieu,
Thank you for providing the code. I attempted to utilize mdetr_features_extraction.py to extract MDETR features.

The MDETR model output includes the following keys:
dict_keys(['pred_logits,' 'pred_boxes,' 'proj_queries,' 'proj_tokens,' 'tokenized']), but it does not include 'last_hidden_states.'

Error (at line 128):
last_hidden_states = outputs["last_hidden_states"].cpu()[0, keep].numpy

I would greatly appreciate it if you could guide me on obtaining the last hidden states.
Thanks