Will GGML support ROCm platform?

[Promesis]

I have 3 AMD Radeon s:

• RX 7900XTX
• RX 7900XT
• RX 6950XT

Although I use my RTX A6000 and Tesla A16 to train and inference, but I want to use my AMD's computing power.

[Green-Sky]

there is some afford being made here ggerganov/llama.cpp#1087

[SlyEcho]

I have to test again, but RX 6950 XT should work (for ggerganov/llama.cpp#1087)

[Promesis]

@Green-Sky thank you.

btw it takes me 6 hours to see all the comments 😄

[Promesis]

But leave it as a feature request. Maybe using HIPIFY the platform migration will be easier. But still waiting for ggml to support ROCm originally.