Misc. bug: Gibbersish output on AMD Ryzen 9 8945HS w/ Radeon 780M Graphics since commit: 3d82dbcbce2c

[marijnfs]

Name and Version

Commit: 3d82dbc

ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = AMD Radeon Graphics (RADV PHOENIX) (radv) | uma: 1 | fp16: 1 | warp size: 64 | shared memory: 65536 | matrix cores: KHR_coopmat
version: 4928 (732b5fb)
built with cc (GCC) 14.2.1 20250207 for x86_64-pc-linux-gnu

Operating systems

Linux (EndeavourOS / Arch)

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

AVX2 on AMD for sure

Command line

Problem description & steps to reproduce

I'm getting complete gibberish output on my AMD Ryzen 9 8945HS, AVX is turned on.
example gibberish:
3
ot
ot
othere are not-
ot
otp
otroverflow. 3

After bisecting, I determined this started at commit 3d82dbc
Which adds Q4_K quantization for x86 AVX2.
I indeed use a Q4 model with avx2 on, so there seems to be something buggy in there!

Can someone confirm this?
The model I used is: bartowski_DeepSeek-Coder-V2-Lite-Instruct-GGUF_DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf

First Bad Commit

Commit: 3d82dbc

Relevant log output

[github-actions]

This issue was closed because it has been inactive for 14 days since being marked as stale.