Sageattention on 2080ti
#1
by oregaoriga - opened
I get ImportError: cannot import name 'sageattn_qk_int8_pv_fp16_triton' from 'sageattention' on KJ's patch node. Should I leave the node on auto or do i need the wheel for sageattention. because I can't find a precompiled wheel anywhere.
For 2080ti I use the cuda sage attention patched for turing. I haven't checked trition in a while. https://github.com/Ph0rk0z/SageAttention2
That or sage windows, it should have all patches too.