Timezone: »

 
Poster
Stable and low-precision training for large-scale vision-language models
Mitchell Wortsman · Tim Dettmers · Luke Zettlemoyer · Ari Morcos · Ali Farhadi · Ludwig Schmidt

Thu Dec 14 03:00 PM -- 05:00 PM (PST) @ Great Hall & Hall B1+B2 #709
in Poster Session 6 »

We introduce new methods for 1) accelerating and 2) stabilizing training for large language-vision models. 1) For acceleration, we introduce SwitchBack, a linear layer for int8 quantized training which provides a speed-up of 13-25% while matching the performance of bfloat16 training within 0.1 percentage points for the 1B parameter CLIP ViT-Huge---the largest int8 training to date. Our main focus is int8 as GPU support for float8 is rare, though we also analyze float8 training through simulation. While SwitchBack proves effective for float8, we show that standard techniques are also successful if the network is trained and initialized so that large feature magnitudes are discouraged, which we accomplish via layer-scale initialized with zeros. 2) For stability, we analyze loss spikes and find they consistently occur 1-8 iterations after the squared gradients become under-estimated by their AdamW second moment estimator. As a result, we recommend an AdamW-Adafactor hybrid which avoids loss spikes when training a CLIP ViT-Huge model and outperforms gradient clipping at the scales we test.

Author Information

Mitchell Wortsman (University of Washington)
Tim Dettmers (University of Washington)
Luke Zettlemoyer (University of Washington; Meta)
Ari Morcos (DatologyAI)
Ali Farhadi (University of Washington, Allen Institute for Artificial Intelligence)
Ludwig Schmidt (University of Washington)

More from the Same Authors