.NET MAUI (Windows) inference 9× slower than WinForms using LLamaSharp

[mfsigauravk]

When running the same LLamaSharp inference code on Windows:

WinForms app: Response time ≈ 10 seconds

.NET MAUI (Windows): Response time ≈ 90+ seconds

The model, parameters, and hardware are the same. Only the UI framework differs.

Steps to reproduce

Create a WinForms app targeting .NET 8.0 or .NET 9.0 on Windows.

Add LLamaSharp and LLamaSharp.Backend.Cpu packages.

Load model + run:

await foreach (var text in executor.InferAsync(prompt, inferenceParams))
{
// consume output
}

Repeat exactly the same code in a .NET MAUI (Windows) project.

Expected behavior

Similar response time (≈10 seconds) on Windows for both WinForms and .NET MAUI apps.

Actual behavior

.NET MAUI (Windows) takes ≈90 seconds or more to produce the same response.

Environment

.NET 9.0

LLamaSharp version: (insert your version)

Windows 10/11 x64

CPU: (insert CPU)

Build configuration: (Debug/Release)

Additional context

Model is loaded once and reused between calls.

Question

Is there a known performance difference between LLamaSharp running in a .NET MAUI (Windows) environment vs a WinForms app?
Any tips for achieving WinForms-level inference speed in MAUI?