Half-precision inference of AI weather prediction models: negligible loss of skill at a fraction of the cost

Ranked by how close their title and abstract are to this paper’s. Select some to read them together.

Closest papers

1. Close Broader context Not relevant

   FP8-LM: Training FP8 Large Language Models

   Houwen Peng, Kan Wu, Yixuan Wei et al. · 2023 · Machine Learning · preprint, not peer reviewed · cited 105 times · AI summary available

   FP8-LM develops mixed-precision training for large language models, whereas this paper tests half-precision inference in AI weather models.

   Abstract

   In this paper, we explore FP8 low-bit data formats for efficient training of large language models (LLMs). Our key insight is that most variables, such as gradients and optimizer states, in LLM training can employ low-precision data formats without compromising model accuracy and requiring no changes to hyper-parameters. Specifically, we propose a new FP8 automatic mixed-precision framework for training LLMs. This framework offers three levels of FP8 utilization to streamline mixed-precision and distributed parallel training for LLMs. It gradually incorporates 8-bit gradients, optimizer states, and distributed learning in an incremental manner. Experiment results show that, during the traini…

2. Close Same question Not relevant

   Pangu-Weather: A 3D High-Resolution Model for Fast and Accurate Global Weather Forecast

   Kaifeng Bi, Lingxi Xie, Hengheng Zhang et al. · 2022 · Atmospheric and Oceanic Physics · preprint, not peer reviewed · cited 309 times · AI summary available

   Pangu-Weather is one of the AI weather forecasting models whose fp32 and fp16 inference costs and forecast skill this paper compares.

   Abstract

   In this paper, we present Pangu-Weather, a deep learning based system for fast and accurate global weather forecast. For this purpose, we establish a data-driven environment by downloading $43$ years of hourly global weather data from the 5th generation of ECMWF reanalysis (ERA5) data and train a few deep neural networks with about $256$ million parameters in total. The spatial resolution of forecast is $0.25^\circ\times0.25^\circ$, comparable to the ECMWF Integrated Forecast Systems (IFS). More importantly, for the first time, an AI-based method outperforms state-of-the-art numerical weather prediction (NWP) methods in terms of accuracy (latitude-weighted RMSE and ACC) of all factors (e.g.,…

3. Close Broader context Not relevant

   FP4 All the Way: Fully Quantized Training of LLMs

   Brian Chmiel, Maxim Fishman, Ron Banner et al. · 2025 · Machine Learning · preprint, not peer reviewed · cited 48 times · AI summary available

   FP4 All the Way studies low-precision formats for training language models, while this paper measures reduced-precision inference for weather forecasting.

   Abstract

   We demonstrate, for the first time, fully quantized training (FQT) of large language models (LLMs) using predominantly 4-bit floating-point (FP4) precision for weights, activations, and gradients on datasets up to 200 billion tokens. We extensively investigate key design choices for FP4, including block sizes, scaling formats, and rounding methods. Our analysis shows that the NVFP4 format, where each block of 16 FP4 values (E2M1) shares a scale represented in E4M3, provides optimal results. We use stochastic rounding for backward and update passes and round-to-nearest for the forward pass to enhance stability. Additionally, we identify a theoretical and empirical threshold for effective quan…

4. Close Broader context Not relevant

   Adding NVMe SSDs to Enable and Accelerate 100B Model Fine-tuning on a Single GPU

   Changyue Liao, Mo Sun, Zihan Yang et al. · 2024 · Distributed, Parallel, and Cluster Computing · cited 15 times · AI summary available

   Adding NVMe SSDs enables large-model fine-tuning on one GPU, while this paper reduces the cost of weather-model inference on one GPU.

   Abstract

   Recent advances in large language models have brought immense value to the world, with their superior capabilities stemming from the massive number of parameters they utilize. However, even the GPUs with the highest memory capacities, currently peaking at 80GB, are far from sufficient to accommodate these vast parameters and their associated optimizer states when conducting stochastic gradient descent-based optimization. One approach to hosting such huge models is to aggregate device memory from many GPUs. However, this approach introduces prohibitive costs for most academic researchers, who always have a limited budget for many high-end GPU servers. In this paper, we focus on huge model fin…

5. Close Broader context Not relevant

   WRF-JAX v1.1.28: a differentiable regional weather model with fully coupled physics, numerically consistent with WRF and matching 128-process CPU performance on a single GPU

   Shijin Yuan, Tianyu Shen, Bin Mu et al. · 2026 · Atmospheric and Oceanic Physics · AI summary available

   WRF-JAX reimplements a regional numerical weather model on a GPU; this paper evaluates precision and cost in AI weather forecasts.

   In short

   WRF-JAX reproduces WRF’s coupled regional simulations near its internal numerical-difference baseline, matches measured 128-process CPU throughput on one GPU, and provides approximate gradients through fully coupled physics.

6. Close Broader context Not relevant

   Towards High Performance, Portability, and Productivity: Lightweight Augmented Neural Networks for Performance Prediction

   Ajitesh Srivastava, Naifeng Zhang, Rajgopal Kannan et al. · 2020 · Performance · cited 2 times · Not summarized yet

   Lightweight augmented neural networks predict hardware performance, whereas this paper measures inference speed and energy under reduced precision.

   Abstract

   Writing high-performance code requires significant expertise in the programming language, compiler optimizations, and hardware knowledge. This often leads to poor productivity and portability and is inconvenient for a non-programmer domain-specialist such as a Physicist. More desirable is a high-level language where the domain-specialist simply specifies the workload in terms of high-level operations (e.g., matrix-multiply(A, B)), and the compiler identifies the best implementation fully utilizing the heterogeneous platform. For creating a compiler that supports productivity, portability, and performance simultaneously, it is crucial to predict the performance of various available implementa…

7. Related Broader context Not relevant

   Shedding the Bits: Pushing the Boundaries of Quantization with Minifloats on FPGAs

   Shivam Aggarwal, Hans Jakob Damsgaard, Alessandro Pappalardo et al. · 2023 · Computer Vision and Pattern Recognition · cited 10 times · Not summarized yet

   Shedding the Bits evaluates reduced-precision formats for FPGA model inference, while this paper tests fp16 and other formats in AI weather models.

   Abstract

   Post-training quantization (PTQ) is a powerful technique for model compression, reducing the numerical precision in neural networks without additional training overhead. Recent works have investigated adopting 8-bit floating-point formats(FP8) in the context of PTQ for model inference. However, floating-point formats smaller than 8 bits and their relative comparison in terms of accuracy-hardware cost with integers remains unexplored on FPGAs. In this work, we present minifloats, which are reduced-precision floating-point formats capable of further reducing the memory footprint, latency, and energy cost of a model while approaching full-precision model accuracy. We implement a custom FPGA-bas…

8. Related Broader context Not relevant

   Accuracy is not compatibility: two Evo 2 checkpoints hinge on one FP8 matrix multiplication

   Xun Ding, Ziyuan Qin · 2026 · Machine Learning · AI summary available

   Accuracy is not compatibility examines how FP8 matrix multiplication affects DNA-model checkpoints, while this paper studies precision effects on weather forecasts.

   In short

   Two Evo 2 checkpoints depend on reference FP8 accumulation rather than numerical accuracy; emulating their first projection largely restores model behaviour and BRCA1 variant scoring.