MiniMax M2.1 Working?

[d-shehu]

Is anyone able to get this model to run?

Mode: MiniMax-M2.1-Q4_K_M
Source: https://huggingface.co/unsloth/MiniMax-M2.1-GGUF?show_file_info=Q4_K_M%2FMiniMax-M2.1-Q4_K_M-00001-of-00003.gguf

It works mostly okay with release b7501 and 64K of context. Occasionally it gets stuck infinite loops, repeating itself or a piece of code with longer session but one shot worked fine.

I made a comment in this thread which seem to be the only MiniMax M2.1 issue I could find.

#18268

I tried a bit of binary search with releases 7501 -> 7772. However, every in-between version I tried either wouldn't load the model and seg-faulted or resulted in garbled output.

All my other models work fine with 7772. Latest drivers for Cuda & Vulkan, using RPC, etc.