paper-with-me

홈 › Papers

Revisiting BFfloat16 Training

2021-01-01 · Pedram Zamirai, Jian Zhang, Christopher R Aberger, Christopher De Sa

State-of-the-art generic low-precision training algorithms use a mix of 16-bit and 32-bit precision, creating the folklore that 16-bit precision alone is not enough to maximize model accuracy. As a result, deep learning accelerators are forced to support both 16-bit and 32-bit compute units which is more costly than only using 16-bit units for hardware design. We ask can we do pure 16-bit training which requires only 16-bit compute units, while still matching the model accuracy attained by 32-bit training. Towards this end, we study pure 16-bit training algorithms on the widely adopted BFloat16 compute unit. While these units conventionally use nearest rounding to cast output to 16-bit precision, we show that nearest rounding for model weight updates can often cancel small updates, which degrades the convergence and model accuracy. Motivated by this, we identify two simple existing techniques, stochastic rounding and the Kahan accumulation, to remedy the model accuracy degradation in pure 16-bit training. We empirically show that these two techniques can enable up to 7% absolute validation accuracy gain in pure 16-bit training. This leads to 0.1% lower to 0.2% higher matching validation accuracy compared to 32-bit precision training across seven deep learning applications.

📄 PDF Abstract BibTeX

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Revisiting Batch Norm Initialization

2021-10-26 · Jim Davis, Logan Frank

Batch normalization (BN) is comprised of a normalization component followed by an affine transformation and has become essential for training deep neural networks. Standard initialization of each BN in a network sets the…

Revisiting Recurrent Networks for Paraphrastic Sentence Embeddings

2017-04-30 · ACL 2017 7 · John Wieting, Kevin Gimpel

We consider the problem of learning general-purpose, paraphrastic sentence embeddings, revisiting the setting of Wieting et al. (2016b). While they found LSTM recurrent networks to underperform word averaging, we present…

SentenceSentence EmbeddingsTransfer Learning

Revisiting Training Strategies and Generalization Performance in Deep Metric Learning

2020-02-19 · ICML 2020 1 · Karsten Roth, Timo Milbich, Samarth Sinha, Prateek Gupta 외

Deep Metric Learning (DML) is arguably one of the most influential lines of research for learning visual similarities with many proposed approaches every year. Although the field benefits from the rapid progress, the div…

Metric Learning

Learning detectors quickly using structured covariance matrices

2014-03-28 · Jack Valmadre, Sridha Sridharan, Simon Lucey

Computer vision is increasingly becoming interested in the rapid estimation of object detectors. Canonical hard negative mining strategies are slow as they require multiple passes of the large negative training set. Rece…

Revisiting Tones in Twic East Dinka

2017-11-01 · PACLIC 2017 11 · Yu-Leng Lin