Models take a long time to be loaded and get stuck on any input

### What is the issue? I've setup ollama to use my 6750XT on nobara/fedora system using fedora ROCm packages (6.3.1) and I can see two issues happening: 1 - It is using just a small portion of my vram (3.8 from 12) 2 - Whenever I input any command/token to the llm GPU goes to 100%, but it get stuck and never "answer" (it just keep showing that "loading" animation) on the terminal Ollama logs with AMD_DEBUG=3 and OLLAMA_DEBUG=1 [ollama_log.txt](https://github.com/user-attachments/files/20526824/ollama_log.txt) Rocminfo [rocminfo.txt](https://github.com/user-attachments/files/20526825/rocminfo.txt) ### Relevant log output Stuck after this ```shell :3:rocvirtual.hpp :67 : 1131975724d us: Host active wait for Signal = (0x7c927a1ff600) for 10000 ns :3:rocvirtual.cpp :484 : 1131975741d us: Set Handler: handle(0x7c927a1ff580), timestamp(0x7c90d45fe6f0) :3:rocvirtual.hpp :67 : 1131975743d us: Host active wait for Signal = (0x7c927a1ff580) for -1 ns ``` ### OS Linux ### GPU AMD ### CPU AMD ### Ollama version 0.9.0