1 paper
Marah Abdin, Jyoti Aneja, Harkirat Behl +24
We present phi-4, a 14-billion parameter language model developed with a training recipe that is centrally focused on data quality. Unlike most language models, where pre-training…