Name and Version

ggml_cuda_init: found 2 CUDA devices (Total VRAM: 32621 MiB):
Device 0: NVIDIA GeForce RTX 5060 Ti, compute capability 12.0, VMM: yes
Device 1: NVIDIA GeForce RTX 5060 Ti, compute capability 12.0, VMM: yes
version: 8611
precompiled from llama-b8611-bin-win-cuda-13.1-x64

Operating systems

Other? (Please let us know in description)
Windows 10 + iOS 26.4

Which llama.cpp modules do you know to be affected?

Other (Please specify in the next section)
WebUI

Command line

llama-server -c 131072 -np 1 --host 0.0.0.0 --models-dir C:\model

Problem description & steps to reproduce

1. launch llama-server
2. visit the webui using safari on an iPhone Pro Max running iOS 26.4
3. start chatting for 3 rounds
4. the chat box becomes invisible starting from the 4th round
5. scrolling the window makes it reappear briefly

First Bad Commit

version: 8557 is affected.
version: 8555 is unaffected.
I suspect this might be related: e6f6770
#20999

Relevant log output

No response