transformers

f007617e - Fix flex_attention block mask creation when `get_seq_length` returns a tensor (#46802)