Closed smiles724 closed 1 year ago
Hi,
Thanks for the question!
I think modality='e2e' is used for conditional generation task of table-to-text, and this is some additional experiments we include in the repo, but not in the Diffusion-LM paper. So if you are trying to replicate experiments in the paper, you probably dont need to run this line. Try modality='e2e-tgt' to replicate the language modeling training of Diffusioin-LM on e2e dataset.
To answer this question, this script is MBR decoding for conditional table-text generation. For a code of very similar purpose, please refer to https://github.com/XiangLi1999/Diffusion-LM/blob/main/improved-diffusion/anlg_infill/mbr_eval.py and https://github.com/XiangLi1999/Diffusion-LM/blob/759889d58ef38e2eed41a8c34db8032e072826f4/improved-diffusion/anlg_infill/mbr_eval.py#L189.
Hi, I notice that when
modality == 'e2e
, thebatch_decode.py
calls thediffusion_lm/e2e_data/mbr.py
file. However, I failed to find this script. Is there an error or typo?