-
Just wanted to let you know that you are incorrectly calculating the BLEU scores. You should split() the predicted captions as well as the reference captions. See [this tutorial](https://www.journalde…
-
What is the difference between your package's bleu implementation and sacrebleu implementation? I calculated the result differently in the two ways, Chinese expected, passed sacrebleu's zh tokenizer
-
Hi, I am a little bit confused why should we set `REFERENCE_TEST_TRANSLATE_DIR=t2t_local_exp_runs_dir_master/t2t_datagen/dev/newstest2014-deen-ref.en.sgm` . because in my mind, the reference should be…
-
we need model evaluation method.
-
Dear author,
I used the code you provided and ran this bash script on the IU Xray dataset for 70 epochs, but the performance metrics are far from those reported in the paper.
Here is my sh file,…
-
-
It seems that when the tested entity's status contains a space character, the card that should be displayed isn't.
That happens for instance with the following
type: custom:state-switch
entit…
-
Hi Bsantraigi
You have provided great tutorials. I saw your transliteration.ipynb.
In this, in Input cell "In[37]", you have put the line 'pbar.set_description(f"Epoch: {_e} >> Loss: {batch_loss/(bx…
-
Hi,
I've just found that the calculation of BLEU score during the training is incorrect.
Current implementation of evaluateBLEU function in train.cc,
just compare hypothesis and target word IDs,…
-
metric blue-4 doesn't increase more than 0.147 with coco data and 0.099 with flickr8k data, is there any way to improve the result. thank