search

Yuanhy1997 / SeqDiffuSeq

Text Diffusion Model with Encoder-Decoder Transformers for Sequence-to-Sequence Generation [NAACL 2024]
https://arxiv.org/abs/2212.10325
88 stars 14 forks source link

problems met during inference #7

Closed BaohaoLiao closed 1 year ago

BaohaoLiao commented 1 year ago

Hi,

I could successfully train a model on IWSLT De-En. Now I have some questions about inference:

  1. decoder_attention_mask is not in the args. Should I just comment it out? https://github.com/Yuanhy1997/SeqDiffuSeq/blob/ed56ca480002b89a26ebd16b0257e1aa880f80f1/inference_main.py#L124
  2. How could I get the BLEU score after I run bash ./inference_scrpts/iwslt_inf.sh path-to-ckpts/ema_0.9999_280000.pt path-to-save-results path-to-ckpts/alpha_cumprod_step_260000.npy
  3. It seems you remove the code for DDIM. Do I have to keep the number of timesteps for inference the same as the one for training?
Yuanhy1997 commented 1 year ago
  1. We do not need the decoder attention mask arg. I should've delete that part of codes.
  2. use the bleu_eval.py script for bleu.
  3. Our code is not a method based on DDIM. You can explore to sample with DDIM.
BaohaoLiao commented 1 year ago

I can get the BLEU score now. Thank you very much.