Open-weight & owned stack · Quantization efficiency
Developer Trains Neural Network On Unsupported AMD MI50 Hardware
A developer reported training across two AMD MI50 GPUs after patching ROCm and using Vulkan when official gfx906 support was dropped.
Read the original at reddit.comOpens the publisher's site in a new tab