Skip to content

ROCm error / Kernels? #148

@Bazwyn

Description

@Bazwyn

Not sure if it's just a me problem, but after the backport issue resolution, I've received the following error when attempting to run the new build:

Starting model warm up, please wait a moment...
ROCm error: no kernel image is available for execution on the device
current device: 0, in function ggml_cuda_mul_mat_q at D:/a/koboldcpp-rocm/koboldcpp-rocm/ggml/src/ggml-cuda/mmq.cu:128
hipGetLastError()
D:/a/koboldcpp-rocm/koboldcpp-rocm/ggml/src/ggml-cuda/ggml-cuda.cu:96: ROCm error

The issue persists with similar errorcodes when I disable MMQ, FA, etc, and my best guess is that the applicable kernels were omit from the build? Request advice or clarification. The build does properly identify my 6900xt and specify gfx1030.

Additional Information:
Win 10
R7 3700x
RX 6900XT 16G
32 GB DDR4
Adrenaline 25.8.1 driver on first test > 25.12.1 on latest tests, no difference.

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