Saved in:
Bibliographic Details
Main Authors: Abdin, Marah, Aneja, Jyoti, Behl, Harkirat, Bubeck, Sébastien, Eldan, Ronen, Gunasekar, Suriya, Harrison, Michael, Hewett, Russell J., Javaheripi, Mojan, Kauffmann, Piero, Lee, James R., Lee, Yin Tat, Li, Yuanzhi, Liu, Weishung, Mendes, Caio C. T., Nguyen, Anh, Price, Eric, de Rosa, Gustavo, Saarikivi, Olli, Salim, Adil, Shah, Shital, Wang, Xin, Ward, Rachel, Wu, Yue, Yu, Dingli, Zhang, Cyril, Zhang, Yi
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2412.08905
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915060960460800
author Abdin, Marah
Aneja, Jyoti
Behl, Harkirat
Bubeck, Sébastien
Eldan, Ronen
Gunasekar, Suriya
Harrison, Michael
Hewett, Russell J.
Javaheripi, Mojan
Kauffmann, Piero
Lee, James R.
Lee, Yin Tat
Li, Yuanzhi
Liu, Weishung
Mendes, Caio C. T.
Nguyen, Anh
Price, Eric
de Rosa, Gustavo
Saarikivi, Olli
Salim, Adil
Shah, Shital
Wang, Xin
Ward, Rachel
Wu, Yue
Yu, Dingli
Zhang, Cyril
Zhang, Yi
author_facet Abdin, Marah
Aneja, Jyoti
Behl, Harkirat
Bubeck, Sébastien
Eldan, Ronen
Gunasekar, Suriya
Harrison, Michael
Hewett, Russell J.
Javaheripi, Mojan
Kauffmann, Piero
Lee, James R.
Lee, Yin Tat
Li, Yuanzhi
Liu, Weishung
Mendes, Caio C. T.
Nguyen, Anh
Price, Eric
de Rosa, Gustavo
Saarikivi, Olli
Salim, Adil
Shah, Shital
Wang, Xin
Ward, Rachel
Wu, Yue
Yu, Dingli
Zhang, Cyril
Zhang, Yi
contents 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 is based primarily on organic data sources such as web content or code, phi-4 strategically incorporates synthetic data throughout the training process. While previous models in the Phi family largely distill the capabilities of a teacher model (specifically GPT-4), phi-4 substantially surpasses its teacher model on STEM-focused QA capabilities, giving evidence that our data-generation and post-training techniques go beyond distillation. Despite minimal changes to the phi-3 architecture, phi-4 achieves strong performance relative to its size -- especially on reasoning-focused benchmarks -- due to improved data, training curriculum, and innovations in the post-training scheme.
format Preprint
id arxiv_https___arxiv_org_abs_2412_08905
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Phi-4 Technical Report
Abdin, Marah
Aneja, Jyoti
Behl, Harkirat
Bubeck, Sébastien
Eldan, Ronen
Gunasekar, Suriya
Harrison, Michael
Hewett, Russell J.
Javaheripi, Mojan
Kauffmann, Piero
Lee, James R.
Lee, Yin Tat
Li, Yuanzhi
Liu, Weishung
Mendes, Caio C. T.
Nguyen, Anh
Price, Eric
de Rosa, Gustavo
Saarikivi, Olli
Salim, Adil
Shah, Shital
Wang, Xin
Ward, Rachel
Wu, Yue
Yu, Dingli
Zhang, Cyril
Zhang, Yi
Computation and Language
Artificial Intelligence
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 is based primarily on organic data sources such as web content or code, phi-4 strategically incorporates synthetic data throughout the training process. While previous models in the Phi family largely distill the capabilities of a teacher model (specifically GPT-4), phi-4 substantially surpasses its teacher model on STEM-focused QA capabilities, giving evidence that our data-generation and post-training techniques go beyond distillation. Despite minimal changes to the phi-3 architecture, phi-4 achieves strong performance relative to its size -- especially on reasoning-focused benchmarks -- due to improved data, training curriculum, and innovations in the post-training scheme.
title Phi-4 Technical Report
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2412.08905