ddro-nq-tu / generation_config.json
kiyam's picture
Final DDRO checkpoint: untied lm_head, tie_word_embeddings=false, vocab_size=38272. Add T5ForPretrainDPO for reproducibility.
9e3a454
raw
history blame contribute delete
142 Bytes
{
"_from_model_config": true,
"decoder_start_token_id": 0,
"eos_token_id": 1,
"pad_token_id": 0,
"transformers_version": "4.53.2"
}