Fix: tambah decoder_start_token_id + clean generation_config

Upload tokenizer untuk model QG t5_simplified

Retrain: fix input format + clean generation_config

initial commit