Segmentation Fault on exit with Vulkan

[TheBill2001]

Hi! It has been awhile since I last touch KoboldCpp so I haven't been testing out much. Recently, I have started using Vulkan because it is faster on my machine, and notice that #588 happens again. This time is with Vulkan only, doesn't happen on CUDA.

Here is what recorded by systemd-coredump when running a benchmark. The stack trace is abysmal this time, though.

           PID: 302990 (python)
           UID: 1000 (tuantran1632001)
           GID: 1000 (tuantran1632001)
        Signal: 11 (SEGV)
     Timestamp: Wed 2024-06-05 00:22:21 +07 (2min 32s ago)
  Command Line: python /usr/share/koboldcpp/koboldcpp.py --model ./models/NeuralDaredevil-8B-abliterated.Q6_K.gguf --contextsize 8192 --skiplauncher --usevulkan --gpulayers 22 --threads 12 --benchmark vulkan-22.csv
    Executable: /usr/bin/python3.12
 Control Group: /user.slice/user-1000.slice/session-69.scope
          Unit: session-69.scope
         Slice: user-1000.slice
       Session: 69
     Owner UID: 1000 (tuantran1632001)
       Boot ID: 0dd9e64477a34b1e8d9ed184afd1e2f1
    Machine ID: 6794e91fd4674aee9506a7cb32a7740e
      Hostname: billarch-pc
       Storage: /var/lib/systemd/coredump/core.python.1000.0dd9e64477a34b1e8d9ed184afd1e2f1.302990.1717521741000000.zst (present)
  Size on Disk: 2.3G
       Message: Process 302990 (python) of user 1000 dumped core.

                Stack trace of thread 303010:
                #0  0x000072f740ff1b80 n/a (n/a + 0x0)
                ELF object binary architecture: AMD x86-64

System:

• OS: Arch Linux
• CPU: AMD R5 3600X
• GPU: GTX 1660 6GB
• Python 3.12.3
• CUDA: 12.5.0