Why Low-Precision Transformer Training Fails: An Analysis on Flash Attention
This addresses a persistent failure case in efficient deep learning for researchers and practitioners, offering a practical solution to enable stable low-precision training.
The paper tackles the problem of training instability in low-precision transformer training with flash attention, identifying it as caused by low-rank representations and biased rounding errors, and demonstrates that a minimal modification stabilizes training, achieving successful training in settings where it previously failed catastrophically.
The pursuit of computational efficiency has driven the adoption of low-precision formats for training transformer models. However, this progress is often hindered by notorious training instabilities. This paper provides the first mechanistic explanation for a long-standing and unresolved failure case where training with flash attention in low-precision settings leads to catastrophic loss explosion. Our in-depth analysis reveals that the failure is not a random artifact but caused by two intertwined phenomena: the emergence of similar low-rank representations within the attention mechanism and the compounding effect of biased rounding errors inherent in low-precision arithmetic. We demonstrate how these factors create a vicious cycle of error accumulation that corrupts weight updates, ultimately derailing the training dynamics. To validate our findings, we introduce a minimal modification to the flash attention that mitigates the bias in rounding errors. This simple change stabilizes the training process, confirming our analysis and offering a practical solution to this persistent problem. Code is available at https://github.com/ucker/why-low-precision-training-fails.