Closed mtsokol closed 4 months ago
Hi @willow-ahrens,
I think another detail, connected to https://github.com/willow-ahrens/Finch.jl/issues/576, is that we would like to have similar input and output dtypes (rather a detail, so only if it's easy):
julia> using Finch julia> a = Tensor(ones(UInt8, 1)) 1-Tensor └─ Dense [1:1] └─ [1]: 1 julia> b = Tensor(ones(UInt8, 1)) 1-Tensor └─ Dense [1:1] └─ [1]: 1 julia> eltype(a) UInt8 julia> eltype(b) UInt8 julia> eltype(Finch.tensordot(a, b, 0)) Float64 # would expect UInt8
Hi @willow-ahrens,
I think another detail, connected to https://github.com/willow-ahrens/Finch.jl/issues/576, is that we would like to have similar input and output dtypes (rather a detail, so only if it's easy):