Closed babangain closed 10 months ago
hyper-parameters are these ones and regarding the data we used all DA data here including the MLQE-PE data.
For making the model output a score between 0 and 1 you just have to rescale the data after concatenating everything. The rescaling is the following:
1) Find a reasonable "min value". This is done by finding all annotations with more than 1 annotator where all annotators agreed that the score was 0. Then your "min_value" is the average z-score for those segments. 2) Find a reasonable "max value". This is done by finding all annotations with more than 1 annotator where all annotators agreed that the translation is perfect (100 score). Then your "max_value" is the average z-score for those segments. 3) Apply a min_max_scaler to your data and truncate every score above 1 and bellow 0.
This rescaling won't impact your model correlations and your model will output scores in that range. This is the same rescaling used in BLEURT also.
Btw the seed used is 91 if I am not mistaken... comet-train --cfg YOUR_CONFIGS.yaml --seed_everything 91
I want to train the model Unbabel/wmt22-comet-da model from scratch. It will be very helpful for me if someone provides the list of datasets, pre-processing steps, and hyperparameters.