65a

Results 41 comments of 65a

There's a reference diffusers config in a PR as well: https://huggingface.co/black-forest-labs/FLUX.1-dev/discussions/3/files

I might also be encountering this. I have several nodes which are not primarily storage oriented, but run ZFS filesystems. Starting somewhat recently, all of them have the ARC consuming...

Wasn't sure how to add to the API changelog... #15108 is a forward compatible API change to support multimodal in the embeddings and completion endpoints. It also adds the capability...

The patch below solves the problem above, but I'm not sure how generalizable it is. Perhaps a start to further discussion. ``` diff --git a/language/go/fileinfo.go b/language/go/fileinfo.go index 38a9566..69ae727 100644 ---...

`GPU_MAX_HW_QUEUES=1` in the environment seems to be a workaround. This should receive an upstream fix, but hopefully that helps ROCm developers narrow the problem.

Sort of similar to [#2844](https://github.com/ROCm/ROCm/issues/2844) except in this case, lower mclk states are listed, but neither used nor reached even when ROCm SMI is used to configure the lowest possible...

I believe I am using the latest VBIOS: ``` [user@machine ~]$ rocm-smi --showhw ===================================== ROCm System Management Interface ===================================== ========================================== Concise Hardware Info =========================================== GPU NODE DID GUID GFX VER...

@IMbackK this is for the Mi210, not the Mi100. Hoping the story is different here, being a newer card (Aldebaran vs Arcturus).

@IMbackK by `runtime-pm` do you mean `amdgpu.runpm=N` or something else? I noticed that some of the kernel runpm code will default to r/o for sysfs entries unless runpm is either...

This seems relevant: https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.h https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c runpm=0 doesn't help, I'll look at pptable stuff next, unless AMD has some better guidance.