Will you add metadata to the dataset.
#3
by
kyuuki
- opened
Hi like to train a model with your dataset but the metadata are missing.
Will you add them soon or how do you use this dataset.
Many Thanks.
Probably I should, but there's no secret to it - it's just a text completion dataset for translation pre-chunked into 1024 tokens using the LLaMA3 tokenizer. You can ignore the ignore_loss column because it's always empty.
One thing I should mention is that the translation pairs were taken from the ParallelFiction dataset, focusing on the dialogue parts. I didn't do any additional quality filtering besides trying to ensure the English lines align with the Japanese lines.