4 ms·

I was running llama on a w7900 a month ago, with 48gb of VRAM and excellent performance. ROCm support got a lot better really recently.

• Zetobal 3y ago
  Inference ≠ Training