Run Inference Without Docker?

[venkatesh-narayan]

Hi!

So I've already used the docker command (under Usage 1) to train and evaluate my model with REINA, and I have all the outputs from that. For further use, is it possible to do

export HF_DATASETS_CACHE=/home/reina/data; export TRANSFORMERS_CACHE=/home/reina/cache
python run_summarization.py ...

on my local run_summarization.py (ie, without using the docker command)?

Also, is Lucene needed for inference? My guess is that it isn't, because it seems like when you have the REINA dataset in cache, all reina() does is load that dataset, and then HuggingFace's Trainer takes care of the rest, but want to confirm.

[shuohangwang]

Thank you for your interest in this work!

Yeah, if you have the data on disk, no need Lucene anymore: ) BTW, you could also try step1 of usage2 to get the data first. Then you could run your own Seq2Seq model based on step1 output : )

Thanks,
Shuohang