I was having trouble using ollama with the 5700 XT (gfx1010).
I was running with HSA_OVERRIDE_GFX_VERSION="10.1.0" ollama serve
and after trying to run a model, I was getting this error: Error: llama runner process has terminated: signal: aborted (core dumped) error:Cannot read /opt/rocm/lib/rocblas/library/TensileLibrary.dat: Illegal seek for GPU arch : gfx1010
I tried using HSA_OVERRIDE_GFX_VERSION="10.3.0"
but that caused my GPU to crash.
I noticed that the /opt/rocm/lib/rocblas/library/TensileLibrary.dat
file did not actually exist, but that other files like /opt/rocm/lib/rocblas/library/TensileLibrary_lazy_gfx1030.dat
do exist. However, there was no file corresponding to gfx1030
.
So, as a shot in the dark, I tried symlinking TensileLibrary_lazy_gfx1010.dat
to TensileLibrary_lazy_gfx1030.dat
: sudo ln -s /opt/rocm/lib/rocblas/library/TensileLibrary_lazy_gfx{1030,1010}.dat
.
To my surprise, this actually worked! Running HSA_OVERRIDE_GFX_VERSION="10.1.0" ollama serve
and then running a model like ollama run mistral
successfully loads and runs the model on my GPU.
I am wondering if the reason the TensileLibrary_lazy_gfx1010.dat
file is missing is a packaging issue, or whether it should be reported upstream?
Pinned Comments
luciddream commented on 2022-01-12 16:47 (UTC) (edited on 2024-11-07 20:44 (UTC) by luciddream)
Latest release: 6.2.4. It uses 25.93GB of disk.