# Online Pseudo-average Shifting Attention(PASA) for Robust Low-precision LLM Inference: Algorithms and Numerical Analysis

arXiv: https://arxiv.org/abs/2503.01873 · HTML: https://mlsanity.com/paper/2503.01873/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 17 of its 31 references that are in the map.
- First posted on arXiv 26 Feb 2025.

## Builds on (most cited first, 17 of 17)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Mamba: Linear-Time Sequence Modeling with Selective State Spaces](https://mlsanity.com/paper/2312.00752/) (2023)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [Qwen Technical Report](https://mlsanity.com/paper/2309.16609/) (2023)
- [FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning](https://mlsanity.com/paper/2307.08691/) (2023)
- [Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets](https://mlsanity.com/paper/2311.15127/) (2023)
- [LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding](https://mlsanity.com/paper/2308.14508/) (2023)
- [FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision](https://mlsanity.com/paper/2407.08608/) (2024)
- [Sora: A Review on Background, Technology, Limitations, and Opportunities of Large Vision Models](https://mlsanity.com/paper/2402.17177/) (2024)
- [Ring Attention with Blockwise Transformers for Near-Infinite Context](https://mlsanity.com/paper/2310.01889/) (2023)
- [Open-Sora Plan: Open-Source Large Video Generation Model](https://mlsanity.com/paper/2412.00131/) (2024)
- [DeepSpeed Ulysses: System Optimizations for Enabling Training of Extreme Long Sequence Transformer Models](https://mlsanity.com/paper/2309.14509/) (2023)
- [SageAttention: Accurate 8-Bit Attention for Plug-and-play Inference Acceleration](https://mlsanity.com/paper/2410.02367/) (2024)
- [Monarch: Expressive Structured Matrices for Efficient and Accurate Training](https://mlsanity.com/paper/2204.00595/) (2022)
- [Bridging the Divide: Reconsidering Softmax and Linear Attention](https://mlsanity.com/paper/2412.06590/) (2024)

## Built on by (most cited first, 1 of 1)

- [Broken Symmetry in BF16 Attention: Why FlashAttention Gradients Blow Up Late in Training](https://mlsanity.com/paper/2609.34272/) (2026)

## Abstract (arXiv)

Attention calculation is extremely time-consuming for long-sequence inference tasks, such as text or image/video generation, in large models. To accelerate this process, we developed a low-precision, mathematically-equivalent algorithm called PASA, based on Flash Attention. PASA introduces two novel techniques: online pseudo-average shifting and global recovering. These techniques enable the use of half-precision computation throughout the Flash Attention process without incurring overflow instability or unacceptable numerical accuracy loss. This algorithm enhances performance on memory-restricted AI hardware architectures, such as the Ascend Neural-network Processing Unit(NPU), by reducing data movement and increasing computational FLOPs. The algorithm is validated using both designed random benchmarks and real large models. We find that the large bias and amplitude of attention input data are critical factors contributing to numerical overflow (>65504 for half precision) in two different categories of large models (Qwen2-7B language models and Stable-Video-Diffusion multi-modal models). Specifically, overflow arises due to the large bias in the sequence dimension and the resonance mechanism between the query and key in the head dimension of the Stable-Video-Diffusion models. The resonance mechanism is defined as phase coincidence or 180-degree phase shift between query and key matrices. It will remarkably amplify the element values of attention score matrix. This issue also applies to the Qwen models. Additionally, numerical accuracy is assessed through root mean square error (RMSE) and by comparing the final generated texts and videos to those produced using high-precision attention.