Flavio Bizzarri

@brainrepo Could you now try the Wuerstchen model with metal by setting 1024x1024 as values? These are the results using mkl: 128x128: 9 minutes and image does not match the...

Try using these proguard-rules: ``` -keep class org.apache.** { *; } -keep class ch.qos.** { *; } -keep class io.netty.** -keep public class org.slf4j.** { * } -keepclassmembers class org.apache.http.**...

@nikola-jokic maybe [tokio::select](https://docs.rs/tokio/latest/tokio/macro.select.html#) + [tokio_utils::CancellationToken](https://docs.rs/tokio-util/latest/tokio_util/sync/struct.CancellationToken.html#) should do the trick [Example](https://github.com/newfla/crab-kafka/blob/main/src/statistics/mod.rs#L237)

@mrkapqa so I should try using these options `sudo amdgpu-install -y --usecase=rocm,dkms,hip --opencl=legacy` ?

`sudo amdgpu-install -y --usecase=rocm,dkms,hip --opencl=legacy` doesn't seem to work for me on the latest 5.4.1 E: Unable to find packet clinfo-amdgpu-pro E: Unable to find packetopencl-legacy-amdgpu-pro-icd I'll try the command...

> @newfla Hi, is this resolved on the latest ROCm? If so can we close this ticket? Hi, unfortunately I no longer have access to that PC and can't provide...

Hi, I can confirm that the same issue occurs compiling candle. See https://github.com/huggingface/candle/issues/1882

I encountered the same problem. [Pipeline log](https://github.com/newfla/fourviere-podcast/actions/runs/8408620117/job/23024994424)

I can confirm the issue on cuda 12.4 with both cudnn 9.1 and 8.9.7 It would be nice if some of the developers could look into this issue

> We've just the upgrade for the latest version of the cuda bindings (cudarc 0.11) in #2165 It might be good to try again though it's worth pointing out that...