sgwm's picture
ViTPose: write bf16 rounding with Veltkamp splitting (fp32 only) so DirectML runs it on the GPU; weights unchanged (maiiro #970)
df127f5 verified