High-resolution modeling is essential for capturing extreme weather events, particularly in coastal and mountainous regions. We introduce StormCast-WestUS, a version of the machine learning-based generative diffusion model called StormCast, trained independently on High-Resolution Rapid Refresh (HRRR) and European Centre for Medium-Range Weather Forecasts (ERA5) datasets over the western U.S. from 2015 to 2024 at 3 km spatial resolution. The model predicts 115 atmospheric state variables at 1-hour intervals, with vertical resolutions of 125 m in the boundary layer and 500 m in the free troposphere. These variables include wind fields, water vapor, and precipitation – key components for realistically representing atmospheric river structure and associated impacts. StormCast-WestUS is evaluated on five atmospheric river events that caused heavy precipitation in the Puget Sound basin of the Pacific Northwest between 2014 and 2020. Its performance is compared to simulations produced by the global Simple Cloud-Resolving E3SM Atmospheric Model (SCREAM), configured with regional refinement at 3 km resolution in the Pacific Northwest and initialized with ERA5 reanalysis. Both models capture the fine-scale atmospheric river structures and associated heavy precipitation, underscoring the importance of km-scale resolution. More detailed analysis and comparison of StormCast-WestUS and SCREAM with observations will be performed and presented.