Translation to German doesn't work in 3B model

#8
by sszymczyk - opened

While working on a T5 support in llama.cpp I noticed a strange thing - translation to German language doesn't work in T5-3B model. The model card says that it supports German language, config.json also contains "translate English to German: " prefix in "translation_en_to_de" task, but running the model with "translate English to German: The house is wonderful." prompt results in the following output: "The house is beautiful and the staff is very friendly and helpful." Why is that? I checked that it's not a problem with my implementation as the output is the same when using transformers library. Also translation to French and Romanian works without problems.

Sign up or log in to comment