Skip to content

coredump when trying to run qwen3-tts on the rocm version #2355

Description

@khirasier

if i try to run the qwen3-tts model through kobold i get this message once it tries to generate.

rocblaslt error: Could not load "/tmp/_MEIBKN92O/library/TensileLibrary_lazy_gfx1201.dat"
Segmentation fault (core dumped) ./koboldcpp-linux-x64-rocm

i also tried to just run qwen3-tts from cpu instead and it just gives me a coredump with no error log of any kind. trying to run on vulkan instead causes the tts to run off of cpu again even with use gpu checked.

kobold version 1.117.1

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions